Chamath: Bitcoin Goes to Zero, Robots vs. $40T Debt, Elon’s SpaceX Shares for 72M Kids | EP #304
Peter DiamandisSalim IsmailDave BlundinAlexander Wissner-GrossChamath Palihapitiya
- Chamath believes AGI has arrived, but physical deployment will determine who captures its value. He expects a “massive” compute shortage over the next three or four years, citing power, permitting, cooling, skilled labor, and utility politics. His nearly 6 GW of accumulated power assets express that thesis; Alexander Wissner-Gross counters that distillation and weak willingness to pay could complicate the scarcity trade.
- Anthropic’s proposed $75–$100 billion equity commitment earns Chamath’s respect for sincerity, alongside skepticism about philanthropic outcomes. The initial 5% allocation depends on shareholder approval, while further grants remain discretionary. His question is the “fail safe to course correct”; Alex’s objection is that a privately controlled benefit fund does not deliver universal public ownership.
- Musk’s promised SpaceX share for every American child is more consequential as an ownership model than as immediate income support. Peter estimates roughly $10.5 billion across 72 million children, about $146 each, against $2.6 trillion annually to distribute $10,000 per American. Chamath calls direct contributions “the better version of the giving pledge”; the panel sees potential in copycats, compounding, and teaching children to own capital.
- Chamath prefers current cash flow and power to long-duration intellectual-property valuations, while admitting unusually low conviction about markets overall. He favors ranking companies by free-cash-flow yield and argues that new chip designs can be technically successful without generating equity returns because manufacturing capacity is already occupied. He is the “least invested” he has been in roughly 15 years: cash flow supplies “the oxygen to pivot.”
- The fiscal escape route discussed is a larger, cheaper economy rather than repayment of the national debt. Against Peter’s roughly $40 trillion debt figure, Anthropic’s reported internal projection puts US GDP at $44 trillion by 2030 versus $30 trillion today. Chamath thinks next-generation GLP3s and cancer vaccines could substantially reduce healthcare spending, making debt “not that important” relative to economic output.
- Free business ideas expose the disagreement over whether execution remains a human bottleneck. Chamath says ideas are “a dime a dozen” and founder resilience matters more; Salim sees an accessible starting point for people without entrepreneurial exposure. Alex, disclosing a financial interest in Idealab.ai’s venture, expects AI-operated businesses to make execution itself “too cheap to meter.”
- AI security becomes a contest over infrastructure, research insights, and competing systems rather than a single alignment solution. The panel treats the alleged OpenAI recording-device case as one instance of a broader intelligence struggle, while Alexander Wissner-Gross warns against excluding foreign talent. Alex favors “defensive co-scaling” and human-machine integration; Chamath explicitly rejects any guarantee that greater intelligence produces benevolence.
- Chamath’s Bitcoin conclusion is explicitly bearish despite his continuing indirect crypto exposure. He cites limits on the buyer pool, ownership-related security risks, and prospective AI-enabled cryptographic attacks, concluding that Bitcoin eventually “gets hacked and then it goes to zero.” He holds exposure through Digital Currency Group equity rather than owning crypto directly; the attack scenario is his prediction, not a demonstrated outcome in the conversation.
1. Anthropic’s pledge shifts scrutiny from generosity to governance
- Peter values Anthropic’s proposed 5% share commitment at $75–$100 billion against a floated $1.5–$2 trillion IPO valuation. It remains “subject to stockholder approval.” An additional 2% could be allocated at the board’s discretion for every 50% increase above a valuation baseline, creating an escalating commitment without making those subsequent grants automatic.
- The founders’ earlier commitments are larger: all seven pledged 80% of their equity, while roughly half the employees are pledging shares with matching support. Peter says approximately 20% of Anthropic is consequently dedicated to public benefit, implying $300–$400 billion at the discussed valuation, versus $130–$365 billion for OpenAI’s foundation.
- Chamath rejects the opening choice between generosity and IPO marketing: “They genuinely believe that what they’re doing is the right thing.” His concern is execution. Large philanthropic bursts generally produce little quantifiable improvement, he argues, so the important question is their “fail safe to course correct” if this effort follows the same pattern.
- Recalling a Giving Pledge pitch a decade earlier, Chamath calls it “reputation washing”: adding his name and wealth to a list without a meaningful outcome requirement. Peter proposes an “impact pledge” instead. Both argue that successful founders contribute intelligence, relationships, and operating ability alongside money, making passive delegation to unproven intermediaries difficult to justify.
2. Public equity ownership requires institutions that do not yet exist
- Alex calls the emerging expectation that frontier labs surrender equity a “hyper tithe.” He interprets Anthropic’s private fund as a snub to public ownership and this administration, while acknowledging Congress has not created the sovereign wealth fund he wants. His proposed intermediate corporate category would exchange public equity participation for benefits such as lighter regulation.
- The discretionary grants create a difficult boardroom conflict, Blundin argues: shareholders resist dilution while the public asks why an authorized allocation has not happened. Dave distinguishes philanthropy from structural reform. In an economy increasingly operated by AI, he wants new ownership and participation mechanisms rather than simply redistributing wealth after it accumulates.
- Chamath asks whether labor should face no tax while capital pays substantially more: poorer households derive more wealth from work, richer households from capital. Alex distinguishes taxing gains from taxing wealth itself, calling wealth taxes widely unpopular. He nevertheless suspects concentrated wealth in a few frontier labs could make negotiated equity taxation more practical, without presenting that concentration as desirable.
3. SpaceX shares make children participants in capital appreciation
- Musk’s promise is explicit: “Next year, I will give one of my SpaceX shares to every kid in America. This will be the first of many.” Peter estimates 72 million recipients, roughly $146 each and $10.5 billion overall, around 1.1% of Musk’s stake. He contrasts the Dell family’s $6.25 billion cash contribution: $250 for 25 million children.
- Peter challenges the scale: an average US household earns about $87,000 annually, while distributing $10,000 per American would require roughly $2.6 trillion each year. Chamath nevertheless calls direct contributions “the better version of the giving pledge,” bypassing the donor’s difficulty distinguishing effective nonprofits and letting children participate in “capitalism and democracy.”
- Blundin’s premise is that value increasingly accrues to assets rather than paychecks. Salim thinks the ownership conversation could matter “100x” more than the initial dollars as children compare selling with holding. Alex adds a political interpretation: a generation owning a possible SpaceX–Tesla combination could feel economically, and potentially regulatorily, aligned with Musk’s businesses.
- Peter floats 50, 100, or 200 newly public-company copycats. The discussion also connects the gift to employee stock-purchase plans and a difficult transition from defined-benefit pensions to defined contributions, arguing that existing retirement promises are insufficiently funded. The gift matters partly because it provides a starting point for that conversation.
4. Energy may price the transition without becoming the final currency
- In the Isaacson clip, Musk guesses conventional currency eventually stops mattering as economic capacity becomes orders of magnitude larger than today’s economy. Physical measures—power, energy, mass, and time—would remain relevant. Asked about human purpose, he concedes, “I do struggle with this question,” while expecting AI and robots to outperform humans at every task.
- Chamath agrees directionally: fiat debasement could push exchange toward claims on embedded effort and productive resources. His speculative mechanism is a token representing access to data centers, power plants, turbines, or natural gas. Saving would also change: owning local compute and contributing it to productive pools could replace income from work that AI increasingly performs.
- Alex is “not as convinced” energy becomes the ultimate currency. Thirty years out, he expects economics to look unrecognizable. His alternative is civilization becoming sufficiently intelligent to model and allocate scarce resources without today’s decentralized monetary signaling. He explicitly resists calling this centralized planning; the claim concerns collective capability rather than a government planner.
5. Chamath’s power thesis starts with an electricity deficit
- Asked about nearly 6 GW accumulated in roughly two years, Chamath begins with ordinary electrification: vehicles, appliances, computers, and data centers all need electrons. Even assuming every planned project arrives on time and budget, he describes a 2050 shortfall as “1.4 terawatt hours,” or six Californias’ worth of electricity. That gap initiated his investment thesis.
- Nuclear does not offer an easy timing escape. His objection to small modular reactors is “the S in the MR”: a small reactor faces much the same local approval effort as a large one. He favors an all-of-the-above supply response, but accumulated solar because it is among the cheapest options and can be deployed in about two years.
- Blundin argues falling chip costs strengthen power’s relative importance. Today’s assumed data-center economics put roughly 10% into power and 90% into Nvidia; he expects that balance to invert as competing designs improve efficiency. For the terrestrial transition, his formulation is “wattage and tonnage”—electricity to run facilities and physical materials to build them.
- Owning assets is only the beginning, Chamath cautions. Financing cycles, equipment supply, construction, and licensed labor remain difficult. Master electricians cannot be recruited as interchangeable workers, and shortages make them expensive. His conclusion is emphatic: the next three or four years bring a “massive massive massive compute shortage.”
6. AGI’s diffusion runs into cooling systems and utility politics
- “I do think that we have achieved AGI,” Chamath says. “It cannot diffuse,” because building the infrastructure is harder than producing laboratory capability. He cites what he describes as a newly released Anthropic S-1 risk factor: hyperscalers will not provide enough compute. To him, that is extraordinary for such a fast-growing, well-funded company.
- Peter cites Musk’s claim that 15 GW of chips are waiting for liquid-cooled data centers. Chamath adds that older-generation inventory also sits unused. As a mechanical-failure example, he recounts two coolant leaks involving Nscale and Microsoft, after which Microsoft allegedly invoked liquidated damages and exited. “Building things and mechanically building things is really hard.”
- Utility coordination can be equally restrictive. Chamath recounts buying northern assets in a multistate power arrangement, then discovering that larger southern participants controlled development and offtake decisions. He singles out Oklahoma Gas and Electric and calls the structure “an organized mafia.” His example makes local relationships and institutional power concrete constraints on AI deployment.
- Space does not eliminate terrestrial demand. Large technology companies might eventually move their compute off-world, but an insurer such as Humana cannot simply launch its own facility. Data sovereignty may require a private cloud or an air-gapped installation. Launch queues and managed-service availability therefore leave substantial domestic construction necessary, influencing winners over the next five years.
7. Getting long electrons faces an efficiency-and-demand counterargument
- Panelists suggest residential or commercial-and-industrial solar paired with batteries: consume what is needed and sell surplus into the grid. Distributed compute pools such as Venice offer another route. Another example discussed is SPAN, which has a deal with NVIDIA for an in-home unit combining storage, a battery, and a GPU. The broader instruction is to find the cheapest, technically manageable way to “get long electrons.”
- A panelist also sees opportunities beyond technical specialists. Local regulatory relationships can unlock projects, while corporate executives can secure future capacity through prepayments. The example is an insurer authorizing a billion-dollar commitment before compute sells out again. Understanding where capital sits and directing it toward construction can matter without knowing how to design models or power equipment.
- Alex accepts supply constraints at the capability frontier but questions whether demand automatically absorbs everything built. OpenAI tried to sell consumers reasoning capabilities they did not want to fund, he argues. If models cannot be economically “metabolized” by customers, abundant energy alone does not create sufficient value.
- Distillation is his second objection: even with no new power plants or data centers, mathematical progress could deliver much stronger models at a fraction of today’s energy footprint. Peter counters that uses will expand for decades. He cites 1,200-token-per-second Cerebras access: expensive enough to try briefly, but sufficiently compelling that “you will never go back.”
8. Alignment arguments diverge over stories, constitutions, and inspection
- Musk suggests truth-seeking and curiosity, while worrying that dystopian fiction supplies AI with bad expectations about itself. Peter’s shorthand is “more Roddenberry, less Cameron.” His positive-future film competition drew over 5,000 submissions, with two winners being made. Blundin agrees training-data selection matters enormously, because choosing the dataset influences the model’s personality and decisions.
- Chamath regards humans using AI against other humans as the greater concern. He also focuses on founding objectives: maximizing truth and alleviating all suffering could produce very different systems. He wants these canonical documents openly inspectable and sees that as an appropriate government interest, while acknowledging uncertainty about how such material is weighted in practice.
- Blundin argues Chinese open-source models capable of self-improvement have changed the safety problem. In his view, relying on post-training alignment is no longer sufficient: “That cat is out of the bag.” He wants inspection and containment, potentially with international authority, rather than confidence that an initial training intervention permanently settles behavior.
9. Competition and human-machine integration offer a competing safety model
- Alex calls both curated information diets and government oversight of “soul documents” brittle solutions. He recalls OpenAI’s founding as a counterweight to concern about a DeepMind singleton, followed by Anthropic’s emergence as another competitor. His preferred mechanism is “competition and defensive co-scaling”: diverse systems balancing one another and, in some cases, competing to be friendlier.
- His second mechanism is humans merging with machines through brain-computer interfaces, uploading, or sideloading, carrying “human meat-body values” into superintelligence. Chamath rejects Peter’s suggestion that extreme intelligence might reliably produce wisdom or benevolence. Silicon, like humanity, could contain a distribution of outcomes, including dangerous systems that other systems must oppose: benevolence is not guaranteed.
10. AI’s fiscal promise is denominator growth and cheaper services
- Musk says he concluded cutting the deficit from inside government was politically impossible: the presidency, Congress, and judiciary lack the appetite to reduce spending. Peter frames the scale as roughly $40 trillion of debt as of October 5, up $2.44 trillion in a year, with around $6.5 billion added daily and $3 billion paid daily in interest.
- Peter cites Anthropic co-founder Jack Clark at the IMF in Bangkok: internal models project US GDP of $44 trillion by 2030 versus $30 trillion today, implying roughly 9–10% annual growth against the historical 2–3%. Chamath’s distinction is that technology need not eliminate debt; it can make the debt much less important relative to output.
- Healthcare supplies his concrete mechanism. Next-generation “GLP3s,” RETA, and cancer vaccines could probably push healthcare below 10% of GDP and potentially halve its cost. He distinguishes dispersed waste—such as hospital upcoding—from dramatic fraud cases. Discontinuous health improvements could prevent use of expensive systems altogether, bypassing the incentives that make waste difficult to remove.
- Peter imagines a Britannica-to-Wikipedia collapse in entitlement and defense costs, potentially requiring a crisis and several more years of technology: a currency “rebasing.” Blundin reads Musk’s DOGE-to-xAI pivot as adaptation to AGI arriving eight to ten years earlier than expected. Peter’s counter-risk is civil unrest forcing money printing and $2,000–$3,000 monthly checks.
11. Countries need resources that matter to the emerging AI powers
- The discussion worries that Europe carries debt without the same AI growth lever; successful American AI drug discovery could pull foreign spending into the US. Chamath generalizes the concern: countries without valuable resources to trade could become “a vassal state” of the US or China. Existing regional alliances may consequently become less useful than access to strategically relevant assets.
- A panelist cites Portofino reportedly using thermal-imaging drones to identify unauthorized air conditioners. With parts of Italy dependent on French electricity, Chamath sees limits to restricting demand rather than expanding supply: eventually, he argues, a country risks becoming “a living history museum.”
- Peter sees a window for countries to develop power and data centers before space-based infrastructure changes the competition again. Citing attacks on Middle Eastern facilities, Alex extends colocation beyond electricity and compute to defense. His speculative endpoint is infrastructure that is “self-powering, self-defending, self-computing,” with physical protection becoming part of the investment opportunity.
12. Cheap ideas expose a disagreement about cheap execution
- Peter introduces Idealab.ai as a free catalog of more than a million business ideas, drawing on 17 million sources and Bill Gross’s 31 opportunity-identification methods. Filters include geography, interests, startup capital, and market size. Gross targets a million businesses and 100 million jobs; Peter says Cloudflare has committed $100,000 in credits per qualifying startup.
- Chamath’s response is blunt: ideas are “a dime a dozen.” The difficult part is enduring failure and putting one foot ahead of the other. Salim counters that people outside entrepreneurial networks need somewhere to begin. The idea list includes algorithmic conflict resolution; Peter develops it as accessible judicial infrastructure embedded into the internet rather than expensive legal processes.
- Alex discloses a financial interest in the venture behind the platform. He had urged Gross to consider AI capable of both conceiving and executing businesses; Gross initially resisted before capabilities changed his mind. The larger thesis is end-to-end AI ideation, management, and operations, making execution itself “too cheap to meter.”
- Blundin offers a commercial explanation for giving away the catalog: many ideas still require financing or compute, sending founders back to Gross for help. “I want to execute on idea number 800,462. Help.” Peter calls it a magnet for entrepreneurs, while Chamath says it takes lean startup to the next level by making 1.2 million experiments possible.
13. Tiny research insights make frontier-lab security unusually difficult
- Peter introduces prosecutors’ allegations that a 35-year-old Saudi OpenAI engineer, arrested October 9, planted a recorder behind an executive conference-room television on September 28. He allegedly used a chatbot through his wife’s account to develop it. A panelist emphasizes that the allegations are unproven and reports OpenAI found no evidence of unauthorized access to model weights, source code, or customer data.
- Chamath regards espionage as routine rather than novel. He recalls assuming foreign intelligence penetration when Facebook reached 3,000 employees and an ITAR-supervised portfolio company losing everything in a hack. Alex hopes the case does not become a reason to turn away foreign nationals, arguing the US should encourage foreign graduates and talent to stay.
- A panelist explains why a conference room could matter as much as weights: $100 million of research might yield a handful of sentences identifying what actually worked. That compressed insight can accelerate a rival’s budget and timeline. Silicon Valley’s customary movement of employees between large companies and startups is poorly matched to a nuclear-secrets security model.
- Alex proposes a new distinction: superintelligence might defend itself against exfiltration, unlike oil or nuclear information. Eventually, AI designing successor AI could secure its own digital discussions. A panelist preserves the uncomfortable implication: if humans no longer know enough to discuss the research, “bug us all you want” becomes reassurance bought at the cost of understanding.
14. Chamath expects crypto’s security problem eventually to overwhelm it
- Chamath calls traditional crypto “capped,” beginning with a restricted buyer progression from retail to institutions to central banks. He argues it lacks the fungibility needed for the last category. Publicly exposed ownership creates a separate personal-security problem; he points to France and kidnappings as his example of why identifying holders can carry serious consequences.
- His AI scenario is an attack on Satoshi’s wallet: break SHA-256, drain assets, establish a short position, then reveal the compromise and profit again. These are prospective attacks in his argument. Asked directly whether Bitcoin is finished, he says, “I just think it’s capped and then I think eventually it gets hacked and then it goes to zero.”
- Quantum-resistant migration is therefore essential in his framing, but Bitcoin’s governance makes agreement exceptionally difficult. Chamath says he no longer owns crypto directly: he contributed long-held Bitcoin to Digital Currency Group and received equity, removing his direct custody burden. Alex is also bearish, citing what he describes as Anthropic’s ECDSA work as evidence of potential inversion attacks.
15. Taking universities public could expose hidden conglomerate value
- Alex proposes SPAC-style acquisitions of research universities, with government support, to commercialize intellectual capital currently constrained by nonprofit tax treatment. His preferred target is Harvard. Separating venture investments, real estate, education, endowment, and merchandise suggests to him a for-profit conglomerate might be worth roughly four times the conventionally understood value. He suggests calling the vehicles “SPARCs.”
- Chamath likes the proposal but calls it “almost too disruptive to be viable.” Execution would require insiders and the support of Harvard Corporation’s leadership; a smaller institution might be a better starting point. Alex cites OpenAI’s nonprofit-to-for-profit transition as a precedent for exploring such conversions, rather than treating tax status as permanently fixed.
16. Cash-flow yield challenges the familiar case for index ownership
- Salim puts Chamath’s earlier valuation argument directly against Alex’s preference for owning the market: if AI could compress S&P valuations from 22 times free cash flow to roughly 2–7 times, why own the S&P 500? The question concerns whether technological abundance benefits existing shareholders at today’s multiples.
- Chamath argues the index no longer supplies its former diversification. Revenue is concentrated in fewer than ten “haves,” with roughly 490 “have nots.” He therefore favors more precise ownership of assets relevant to the transition rather than assuming the largest index remains a balanced representation of opportunity.
- “IP is mostly not valuable anymore,” he argues, weakening conventional defensive moats and long-range underwriting. His alternative is to rank the S&P 500 or Global 2000 by free-cash-flow yield and buy the top group. Cash matters as a bridge into whichever unit of value ultimately succeeds today’s system.
17. Chip-design abundance can coexist with poor semiconductor returns
- Asked why he moved from Groq to power, Chamath emphasizes timing. His first check preceded the company’s formation, and partners considered the idea foolish. The discussion references Nvidia’s $20 billion Groq deal. Now that starting a silicon company is conventional wisdom, he thinks the attractive entry point has passed.
- Domain-specific architectures and customized operating environments can still be technically correct, he says, without creating enterprise value. The bottleneck is production: memory commitments, fabrication slots, and packaging capacity are occupied. A newcomer becomes the “nth person in line,” which he describes as “effectively DOA,” regardless of the design’s merits.
- AI-generated architectures could multiply silicon options while destroying scarcity in design IP. He instead goes down the stack into data centers—dismissed by some investors as too “blue-collar”—and up into applications, where monetization remains uncertain. His claim that new silicon equity returns disappear is categorical, but he ends the explanation with “I could be wrong.”
- The closing allocation answer is unusually tentative: “My honest answer is I don’t know.” Chamath is the least invested he has been in roughly 15 years and contrasts his confusion with late-cycle market confidence. He favors cash flow for “the oxygen to pivot,” power as a 10–15-year constraint, and alternative capital-deployment mechanisms as traditional financial infrastructure becomes more closed.
Full transcript
I do think that we’ve achieved AGI. It cannot diffuse, and the reason it cannot diffuse has nothing to do with the technology. It has everything to do with the buildup until you actually build it out so everybody—or a plurality of human beings—can feel it. Otherwise, it’s just something that exists in a lab.
Anthropic announced the Anthropic Public Benefit Fund. Five percent is worth somewhere between $75 billion and $100 billion.
I think these guys walk the walk and talk the talk. I don’t think there’s anything nefarious there. I think they genuinely believe—
Next year, I will give one of my SpaceX shares to every kid in America. This will be the first of many. There are 72 million Americans under age 18, so that’s roughly $10.5 billion in total, or $146 per kid.
This is really the better version of the Giving Pledge.
Teaching this entire generation to own equities is a really good move, I think.
Now the big question: can Trump accounts, contributed to by billionaires and frontier labs, actually move the needle for the average American?
Now that’s a moonshot, ladies and gentlemen.
1. Introducing Chamath Palihapitiya
Welcome to Moonshots, everyone. Your number one podcast in all things AI and exponential, your front-row seat to the accelerating singularity. Today I’m here with the original crew, AWG, our in-house ASI.
Alex, good to see you. You changed your orchid today.
Nah, it’s just a flood fill.
Okay. Dave Blundin, our impresario of AI investing, and Salim, our warlord against linear thinking. Salim, you’re home.
I am actually home. Thank God. Loving every second of it.
And from the All-In podcast, the man his co-hosts call “Chairman Dictator,” Chamath Palihapitiya. Chamath, a pleasure to have you on our crossover episode.
Thank you. It’s great to be included.
At the end of this, I want you to tell me which of the Greek moderators you prefer: me or J-Cal. Okay?
Okay. You.
But for the 1% of people who don’t know who you are, let me do a proper introduction. Chamath was born in Sri Lanka and grew up in Ottawa with very little, working at Burger King to help pay the bills and make ends meet. He became AOL’s youngest vice president, then built the growth team that helped Facebook reach a billion users. Since founding Social Capital in 2011, he’s been on a rocket ship: he purchased 10% of the Golden State Warriors and put the first check into Groq, which NVIDIA just licensed in a $20 billion deal. Today, he’s CEO of 8090, which raised $135 million this summer to build the world’s legacy software with AI. Chamath, a pleasure to have you on the fun side of the podcast universe.
God, the way you say it, it just looks like it all works.
Hey, what you overlooked there is that Chamath is shedding all material possessions. So the Golden State Warriors, I guess, are going to be off the docket, right?
I mean, parts of my family still own part of it, but I do not.
And everybody, I’m Peter Diamandis, your host and abundance provocator. Our mission is always to help you understand what just happened, what it means, keep you optimistic and abundance-minded as we go through the singularity. And this is another crazy week. Before we start, I just had the most amazing lunch with William Shatner and his son-in-law, Joel. I spent 2 hours with Captain Kirk, my childhood hero, and he was grilling me about geothermal energy and the best companies for drilling. I was just so impressed. He’s 95 years old, and the guy has an incredible passion for life. And guess what, Salim? He’s joining us at Abundance Summit this year.
Oh, that’s awesome.
Yeah.
I am shocked that Spock wasn’t your childhood hero.
No, it was the guy who got the girls, for sure.
Oh, okay.
And the one who was just pushing the universe forward.
Chamath, it’s incredibly awesome to get you on the pod. Back during COVID, my oldest son, Daniel, came to me and said, “I just saw the most amazing podcast ever. There are actually 4 people on the planet who know what they’re talking about. You’ve got to check it out.” It was a lifesaver through that whole window when everyone was trapped at home. I’ve been a huge fan ever since, so it’s great to meet you.
Thank you for saying that.
Yeah, I think all of us are podcast fans.
It’s the one podcast I never miss. Ever.
Really?
Yeah.
That’s a commitment.
Because you miss this one all the time. I missed—wait, how does that work? But I don’t miss that one.
Because you’re someplace on the planet.
Yeah, no, sure. The 2 podcasts I listen to are this one when we’re recording and All-In.
We’re turning into the Marvel Cinematic Universe with this All-In crossover. Where’s Iron Man? Where’s Doctor Strange?
2. Anthropic’s $100 Billion Public Benefit Fund
I see. And then we’re going to have an internal war, like Avengers against Avengers—
Inevitably.
Let’s open with a story that’s either the most generous move in corporate history or the cleverest IPO marketing ploy ever. On Sunday, Anthropic announced the Anthropic Public Benefit Fund. Let me read from their post: “Subject to stockholder approval, 5% of Anthropic shares will be committed today, with an additional 2% added at the board’s discretion for every 50% increase in the company’s value above a set baseline.” The post continues: “It is common practice for tech companies to offer escalating compensation awards to their executives. We are instead directing that equity toward public benefit.”
So, what is it worth? Anthropic is heading for a roughly $2 trillion IPO, though $1.5 trillion is being floated. At that range, 5% is somewhere between $75 billion and $100 billion. Add the founders’ pledge: when Anthropic started, all 7 co-founders pledged to give away 80% of their equity. About half of the employees are also pledging shares, with a matching grant. Anthropic says that altogether, roughly 20% of the company is dedicated to the public good, placing it somewhere between $300 billion and $400 billion, compared with the OpenAI Foundation’s value of $130 billion to $365 billion. It’s a lot of money going toward public benefit.
The reactions on X were instant and varied. Elon said, “I recommend giving the shares directly to the public. This is what I’m going to do.” More on this in our second story. AI analyst Andrew Curran said, “It sounds to me like they’re planning their own form of self-directed UBI.” Brad Gerstner said, “It’s an extraordinary gift,” but warned about “foundation wealth concentration,” and said, “I hope a meaningful portion of AI wealth is direct dividends.” Then Palmer Luckey came out and said, “This reeks of USAID 2.0.”
Chamath, let me go to you first. You’re an investor. Is a 5% donation right before an IPO generous, or is it a brilliant way to get goodwill?
I think it’s neither of those things. I think it’s more that this is their philosophical bent, and they’re staying true to their word. The idea that when you start a company, you would give up 80% of the gains even before you have an idea of what the terminal outcome is pretty incredible. I think these guys walk the walk and talk the talk. I don’t think there’s anything nefarious there. I think they genuinely believe that what they’re doing is the right thing.
There’s a different question: if you look at the history of these sorts of philanthropies and philanthropically minded efforts, they generally don’t turn out that well absent a few. So I think the real question is, what’s their failsafe to course-correct if this mean-reverts and becomes what everything else in the world has overwhelmingly looked like when philanthropically minded people give large sums of money away in big bursts—which is not a lot of quantifiable outcome? I think that’s the really big question.
3. Chamath on the Giving Pledge and Philanthropy
One of the critiques I have of the Giving Pledge is that people pledge half their wealth while they’re still alive, et cetera, et cetera. But the greatest ability you have isn’t just your wealth; it’s your connections. It’s what you do. I’ve always said there should be an impact pledge, where people pledge their wealth in a measurable, impactful way, and you monitor it so people make a difference. Are you a member of the Giving Pledge, by the way?
No.
Okay, great. Thank you.
You didn’t ask me that. Let me be direct. People pledge to give away half their wealth, and so on. They came and pitched me about the Giving Pledge a decade ago, and my take was, “This thing is a scam.” The reason is, it’s reputation washing. They had no—the only goal was to put my name on a list and add my number to their number so they could walk around with a bigger number. I was like, “This is the stupidest thing I’ve ever seen.” I’m like, “What do I get in return for all of this?” It’s like, “You get to go to a weekend,” and there’s a get-together weekend.
So, look, I think a lot of people probably thought it was, on the margins, the right thing to do, but it was pretty obvious to me that it was pretty stupid.
Yeah.
Didn’t Warren Buffett bail out of that recently, too? He signed up at the front end, and it’s a big deal when you say no to these guys a decade ago, because it looks like you’re doing something wrong and they’re doing everything right.
That’s a sign of the times, actually. There was a lot of social pressure on many fronts that has since course-corrected. That was one of the first instances of pure peer pressure and arm-twisting. If you said no, you got this negative reputation.
Here’s what I don’t understand. If you’re capable of making incredible gains in society economically, what it probably says about you is that you are, at a minimum, clever. Sometimes you can get lucky, but if you’re unlucky afterward, you kind of give it all back. In order to make and sustain wealth, you have to be relatively clever, well-timed, well-read, and well-reasoned.
You make all of this progress. What it says about you is actually a fair amount: you’re quite a successful, evolved person. That doesn’t mean there aren’t evolved people who don’t have money, but what I’m saying is that the people who typically tend to become hyper-successful are functional people. The idea that those functional, high-achieving people would suddenly let some other person they don’t know to be smart or capable put a bunch of money to work sounds stupid. You’re the smart one. You should put it to work. At a minimum, raise your kids, because they’re likely to be as smart as you.
There are all these ways you can think about philanthropy, or give it away like Elon’s doing, which you’ll talk about, I’m sure, with the Trump Accounts. There are so many scaled ways to do it. This was the one that had nothing to do with actually achieving anything and everything to do with pure pressure and reputation washing.
Yeah, agreed. Your greatest gifts are not your money. It’s your intelligence, your relationships, and your ability to make a difference in the world. Alex, what’s your take on these public benefit funds?
Yeah, I think there’s an elephant in the room. There are many elephants in many rooms. This one was what I’ve called a hyper-tithe. There’s an increasingly anchored expectation that frontier AI labs—which many, not all, think are well-positioned to suck up much of the value in the economy—will donate some quantum of their equity to the public benefit.
Maybe it’s 5%, hypothetically. Five percent is, of course, also the magic number that nonprofit foundations in the U.S. are required to distribute every year. I think there was a latent expectation out there that Anthropic, probably seeded by OpenAI and others, would take 5-ish% of its equity, perhaps in connection with its forthcoming IPO, and donate it to a U.S. sovereign wealth fund, the 538 Trump Accounts, or something—a way for the general American public to benefit from all of its newfound liquidity.
The fact that Anthropic, as it appears, is not doing that—instead creating a private fund and not donating 5% to the general public—I construe as a snub to this particular administration, a snub to the idea of a universal basic equity. Granted, Congress hasn’t acted to create a sovereign wealth fund yet, although there was some interest from the executive last year in doing that. Congress hasn’t yet acted. I think what the whole world is waiting for, in my dream scenario, is an act of Congress.
I don’t think this could come purely from executive order. An act of Congress would create a sort of middle tier of company, halfway between a nonprofit required to distribute 5% of its assets per year and a general for-profit that distributes none of its assets. That company would give a certain percentage of its equity to the U.S., to the public, to the government, or to a sovereign wealth fund, which I would argue needs to be created in return for something. Congress hasn’t done that. The White House, I think with the best of motivations, tried to put together a sovereign wealth fund last year. It hasn’t quite happened yet.
I think the question I have is: in a better world, where Anthropic could have grown into this—not quite a public benefit corporation, but some new class of company that gives up a certain percentage of equity, not cash, free cash flow, or income, to the general public for public benefit—what would they get in return? To Chamath’s earlier point about the Giving Pledge, what would Chamath have wanted from the Giving Pledge in a hypothetical better world in return for—
To be left alone?
To be left alone. So, okay. So maybe—
I would have given him half the money. They just left me alone.
Well, the Anthropic version of being left alone might be lightness of regulation, which starts to smell like a safety cartel or some sort of light-touch regulatory treatment: being left alone by the government. Something in return for giving up 5% of their equity. I think that’s the part of the equation that’s still missing.
Nice.
Two or three thoughts. This whole construct of philanthropy is still a very scarcity-era response, right? A company becomes successful, creates wealth, and decides how to redistribute it. As we move to an exponential abundance world, you need new models of ownership, participation, and direct access to services. I think this is the bigger challenge.
The good news is this is actually starting. It’s the beginning of the discussion of new economic institutions, and I think that’s the good news around this. It’s going to force that conversation, because you have to distinguish philanthropy from the structural economic reform that’s needed. We’re still taxing labor, and that’s an archaic structure in a world where AI will take over more and more of that.
Yeah. You know, interesting, Dave: the reality is this philanthropic money, if you want to call it that, is actually controlled by the for-profit company.
Mhm. I’ve been on many boards, and when you authorize the board to contribute another 2% for each 50% valuation increase, man, did you just make this a stressful board to be on? How are you going to make that decision? You’ve got the shareholders over here saying, “No, don’t take the dilution,” and then you’ve got the public over there saying, “You’re authorized to do it.” What’s the decision, guys? That’s just—
It kind of punts a really tough choice down the road. I think there’s going to be so much demand from the public to own part of these companies. This is just the very beginning of the process by which these large hyperscalers start having an expectation that part of their company, part of their equity, is going to be distributed. I think this is just the early days.
Yeah, or for anyone who lost the thread of the drama here: remember, Anthropic was born out of disgruntled OpenAI people who didn’t trust OpenAI to be safe with ASI or SI. That was their founding. All the MIT people, brilliant researchers I know who went there, went there specifically because they trusted Dario to be that voice of sane, safe, and rational AI. So that’s core.
I think much later, when he said, “Hey, I only own 1% of this company. We need to reclassify our shares so that the seven of us have super-voting control,” that could be spun very negatively. You look at the memes of Dario: he was never memed in a bad light right up until recently. Now, with the IPO coming, look all over X. He’s painted in all kinds of different lights.
He’s a devil incarnate.
Oh, yeah.
It’s tough. No, I think part of the issue is that there isn’t an established, broadly established regime for the U.S. government to tax equity. It can tax income, and it can tax cash flows. There are more exotic ways one could imagine for the government to tax and distribute equity itself, but generally in the U.S. this is an unfamiliar concept.
More familiar would be for the U.S. government to tax profits or capital gains, for example, and then distribute via whatever means, including stimulus checks, cash, or free cash to the general public. But the government being able to tax equity and distribute equity isn’t a very familiar muscle movement for the U.S. government overall. I suspect that if we’re going to get to a regime many would like—universal basic equity or universal basic dividends—being able to tax and distribute equity, not just cash, would need to become a new muscle movement.
Maybe if that existed, perhaps Anthropic would take a different position. It might not just keep this 5% as a private benefit fund and might instead be more open to public distribution.
I think there’s a really important concept that was brought up earlier. The question is: what is the right level of taxation for labor versus capital?
Mhm.
Um.
Yes.
There’s probably an argument to be made that capital could be taxed much more punitively than labor, and maybe labor should not be taxed at all. How would you make that argument? You’d probably say that, in the wealth-inequality distribution, the percentage of wealth that comes from labor versus capital is inversely indexed to the place in the hierarchy you’re in. The first percentile of salaries versus the 99th percentile of salaries—or of wealth: the 99th percentile of wealth is mostly capital, very little labor, I’m guessing. The opposite would be true at the first percentile. Maybe there’s something to be said for paying much more incrementally for capital as you go up.
Which, by the way, has an interesting effect: it effectively acts as a gate on profligate money printing by central banks. If they create fake liquidity and that ends up in your pocket as capital that you could otherwise deploy, charging larger take rates on that is probably not unreasonable. Meanwhile, if you tax labor at nothing, the people whose wealth is 99% labor all of a sudden benefit quite deeply. There’s probably something very simple there that mathematically works, and you just have to find the break-even point on capital. That may be the simplest and most effective thing people could do.
I agree with that premise. I think maybe there’s a concept we’re circling that many would regard as unpopular: the idea of taxing capital more effectively. Of course, historically, the problem with taxing capital is that you want to tax the gain, not the capital itself. If you tax the capital itself, we’re talking about wealth taxes in some sense, and part of the problem is that wealth taxes, at least in many quarters, are wildly unpopular. If, however, arguendo, a large chunk of the American economy’s wealth were concentrated in just a handful of entities, then suddenly the idea of a wealth tax becomes a lot more practical. The U.S. government could, by hook or by crook, say to all of the frontier labs, “I want to tax a few percent of your wealth in return for following XYZ positive treatment.” I suspect that’s where this may go, but I guess we’ll find out.
Well, let’s move on to this.
Very slippery slope, trying to find wealth.
4. Elon Musk’s SpaceX Shares for 72 Million Kids
Let’s move on to the second shoe to drop. Elon didn’t just criticize Anthropic. Within hours, he answered. Early Monday morning on X, he said, “Next year, I will give one of my SpaceX shares to every kid in America. This will be the first of many.” He had 520,000 likes and 27 million views in less than a day.
President Trump responded, “A big thank you to our legendary friend Elon Musk for his incredible generosity in giving SpaceX stock to millions of American children through Trump Accounts. This is exactly what was envisioned when we created Trump Accounts, bringing together the great private sector, Elon and so many more, to invest in our children, by far our greatest, most important asset, and give every American child a piece of the American dream. Thank you, Elon.”
Let’s do the math for a second. There are 72 million Americans under age 18, and SpaceX’s current stock price is about $148. So that’s roughly $10.5 billion in total, or $146 per kid—about 1.1% of Elon’s stake being shared. For context, last December, Michael and Susan Dell gave $6.25 billion, about $250 each, to Trump Accounts for 25 million children. Unlike the Dell gift of cash, Elon’s transferring equity. And honestly, ownership beats handouts. I think we would all agree with that: a paycheck gets spent, and shares compound.
We previously talked about SpaceX hitting $10 trillion, $100 trillion, and Elon’s recent prediction that SpaceX could become worth “orders of magnitude more than the current Earth economy.” The replies on X were classic. The top reply, with 30,000 likes, was, “I identify as a kid in America.”
So now the big question: Can Trump Accounts funded by billionaires and frontier labs actually move the needle for the average American? I think the answer is probably no. The average U.S. household is pulling in $87,000 a year. To pay every American just $10,000 a year takes about $2.6 trillion per year. So Anthropic’s and Elon’s philanthropy is really just a gesture, not a funding source. Even the biggest fortunes on Earth are just one-time paychecks.
Chamath, you’ve spoken in the past about a birth dividend. I love that idea, in which the government invests $2,000 for every newborn in the market, locked in until age 65. I’m wondering if Elon’s going to put a lock-in requirement on his shares as well. I hope he does, so they’re not just liquidated and spent right out of the gate. What do you think of Elon’s move here, Chamath?
I think it’s great. This is really the better version of the Giving Pledge. For most of us who want to contribute but don’t have the infrastructure or the capacity to understand how to discern a good nonprofit from a bad nonprofit, or a good NGO from a bad NGO, direct contributions into the accounts of kids who can then participate in the experiment that is capitalism and democracy in the United States are a no-brainer. I think it’s unbelievably impressive and unbelievably generous, and I think it has a lot of legs.
Yeah, it’s great. My kids put half of every dollar they earn into stocks. They’ve all put it into SpaceX. I wonder where they got influenced by that. Then savings and philanthropy, and then their spending budget.
Dave, I’m curious. Seventy-two million people now own SpaceX shares at the end of this. Is that a risk if they dump it? What’s that going to do to the stock?
No, no, no. It’s 1%, like you said. It’s not going to move the needle if they all dump it. But I think the reason this is brilliant is because we’ve all discussed on this pod before that the future of value comes through asset ownership, not employment—not paychecks. What is AI not going to do two years from today, three years from today?
I was over at MIT yesterday, talking to a bunch of undergraduates, and I said, “Your window of opportunity to tap your big MIT brain is about the next two years. But do you really think that AI can outthink you in every conceivable way more than two years from today?” That is your window of opportunity. You have to move fast. But that doesn’t mean wealth goes away. It all moves to asset value. Teaching this entire generation to own equities is a really good move, I think.
And, Peter, you said a second ago that this doesn’t move the needle on the needs of Americans, but Elon believes SpaceX is going to be bigger than the current total global economy.
Yeah, $1 to $10 quadrillion is the estimate.
Maybe he’s right, maybe he’s wrong, but that’ll keep these kids engaged with each other. The ones that dump it will be talking to the ones who didn’t. Something good is going to come out of that conversation. I think it’s really cool.
Salim, your thoughts, pal?
Chamath, very quickly, just to build on what Dave said: I think the fact that all of these kids are looking at it means you now have broad participation in the appreciation of capital. I think that’s a really good thing, and I think that conversation is going to make a much bigger difference—like 100 times bigger—than the actual dollar amount. Now all these kids are thinking in that modality as they grow up, watching this, comparing it, seeing who sold and who didn’t sell. It’s awesome as a conversation.
I also think this might set a trend. What if all other public, newly public companies that want to be cool with the next generation just follow the same trend? Then you get 50, 100, 200 copycats, and it really does move the needle.
Yeah, it does move the needle. I don’t know if you guys have worked in companies that have had ESPPs—employee stock purchase plans—where you contribute on a quarterly basis, and at the end of the quarter they buy you a defined amount of stock at the low point of the quarter. There are all these mechanisms that induce you to buy the stock of the place that you work.
I also think this begins the very difficult conversation about moving from defined benefits to defined contributions. It shines a light on the fact that these defined-benefit retirement programs are insufficient and poorly funded. People will ask, “Why can’t we do more forms of defined contributions?” That’ll be the right conversation, and I think you have to have a starting point. Trump Accounts and Elon’s gift start that conversation, which is very important.
Alex, Elon said this will be the first of many. I’m curious if this is going to start setting expectations for all the public frontier labs.
Yeah, maybe. Other than the obvious universal basic equity angle, I think it’s also potentially very clever marketing. Imagine an entire generation of Americans who feel economically aligned with the outcome of a presumably merged SpaceX and Tesla. Suddenly, everyone owns equity not in the general market, which arguably would be a slightly better outcome for democracy, but in one company. This is the one company that every American owns a share of and now feels economic alignment with, maybe regulatory alignment.
I was out in California, in Santa Monica, Peter, your neck of the woods, walking down Santa Monica Place, and there was one of these protests—protesting Elon, right in front of the Tesla dealership. Imagine an alternative America: the Republic of Santa Monica, for sure.
Yeah. Imagine an alternative America where every person in a generation owns an economic stake in Elon’s prolific economic output. Would that be a very different political America? I think it probably would be.
I think it’s a brilliant move. I texted with him about 4 months ago. I was trying to see if he would fund a $10 billion XPRIZE, and he goes, “Interesting, but I’m going to give my shares away.” So he’s been thinking about this for some time. Alex, you just mentioned—and it’s in the news today—that Musk says Tesla and SpaceX have formed special committees to discuss the merger. So that game is afoot. Not a surprise.
Isn’t that the obvious next move after the Tesla Roadster? We didn’t get a chance to talk about it as a full group, but in my mind, the Roadster announcement—that the car hops, it doesn’t hover—what was the point of that? In my mind, it was 90% just to secure, or help secure, a Tesla-SpaceX merger.
Yeah, for sure. I love this, and kudos to Elon for making this happen. One other point: Elon has talked about universal high income. I think the only way this materializes is if everybody gets a piece of the pie and grows with it, right? They couple to the singularity; they don’t go off the offramp here.
Anyway, the next story I want to hit on is a 3-part series of clips from the Walter Isaacson interview with Elon today, published on Builders. In our first clip, Isaacson asked Elon whether we’re heading toward a post-cash world. Elon said money is going to stop being relevant at some point in the future.
What does the future look like?
5. Elon’s Vision for a Post-Money Economy
That money becomes irrelevant. I mean, there will still be currency, but I think currency will be measured in fundamental physics metrics, like power, wattage, and time, mass and energy, mass and power, or energy, basically. Conventional currency will not matter in the future. That’s my guess. I think it’s pretty obvious.
As soon as you start getting many orders of magnitude bigger than Earth’s economy, satisfying whatever humans can possibly think of for their needs is very easy. If you’re several orders of magnitude bigger than Earth’s economy, then providing humans whatever goods and services they can think of that they might want is trivial.
And so, what do we do then as humans? What gives us meaning? How do we organize ourselves?
I do struggle with this question. I think people will find ways to be happy, in the same way that people who retire seem to be fairly happy even though they’re not working. But I think the reality is that there won’t be anything humans can do that AI and the robots can’t do better.
So, Chamath, you’ve locked up what, nearly 6 gigawatts of power? Do you feel like you’re holding humanity’s future currency?
Yeah, I’m very aligned with what he’s saying. I think a different way of thinking about this, less eloquent but maybe useful, is that at some point you can only debase a fiat currency so much before it becomes fundamentally worthless, and you have to shift to an alternative way of measuring value.
If you instead move the measurement to the implied energy or the implied effort that the currency gives you, I think that’s a more symbolic representation of future value. Take a cryptocurrency that takes, I don’t know, N number of watts to compute, or N number of joules or whatever. That’s a more valuable thing in the future than, pick your fiat currency, which is completely uncoupled from any form of value.
If this one asset actually has some amount of effort that was burned to make it, and the other one was just a decision by a bureaucrat, I think the latter becomes far less useful. The question is probably: How do these assets trade in a more fungible way and as a mechanism of exchange?
If I had to guess, it would be more like having things such as data centers, power plants, turbines, and natural gas supplies. You effectively have embedded compute, embedded energy, and around that you have some sort of crypto token that represents some unit of that access. That’s how you exchange value.
Interesting. He’s talked about a post-capitalist society for some time. He’s famously said, “Don’t save for the future.” Do you sort of concur with that?
Well, I think what I would say is maybe slightly different. Saving in the conventional sense is what we’ve all been taught to do. There hasn’t been a bridge created where you can save in an alternative sense.
And these other new forms—a different way one could save in a new form of value would be when you have enough local compute and participate in the economy in a way where you actually represent value. That’s different from the traditional sense, where the work we get paid for today effectively goes away because it’s done largely via AI, and so the compensation for that work changes materially.
On the other hand, if you have resources that you can put into different pools to accomplish different tasks, I think that will have value. That is sort of how I think people would make that transition.
You know, Alex, we’ve been talking on the pod for a while about how GDP sucks as a measure. So we’re going to get a Kardashev scale as a new measure of wealth for a country.
Maybe the premise of the Kardashev levels, though, is energy output. I’m not as convinced as perhaps others are that energy wants to be the ultimate unit of currency. I suspect it’ll be something more exotic or, to Elon’s point, entirely irrelevant and not even the right question.
I have conversations with AIs or SIs all the time about what the future of capitalism looks like and what the future of economics looks like. My working hypothesis is that 30 years from now, it will just look unrecognizably different, and we’ll look back on conversations like, “Is energy, is land, is power—are these the right units of currency?” We’ll look back and laugh at these discussions.
They’ll be intermediate value.
Oh, I totally agree with you. The terminal value—who knows? I totally agree with you.
By the way, I agree with you on the GDP thing as well. GDP is so, so dumb, so basic. Elon loves his little parable of people passing feces back and forth, or whatever, and calling that GDP.
You can’t just let that hang. You’ve got to tell the actual story.
It’s a short interview podcast.
I gave you the abbreviated, polite version of Elon’s story. But I do think that, if I had to guess, what does the successor to money—call it capitalism 2.0, Charlie Stross-style—look like? I imagine a world where, 30 years from now, the ability to make intelligent resource-allocation decisions, and other allocation decisions, is a lot easier.
Right now, it’s hard to intelligently allocate capital, so it all gets pushed to the edge, and we live in an “I, Pencil” world where all of the decisions live at the edge. Imagine, alternatively, without this being misconstrued as some sort of centralized planning or communist utopian vision, a world where human civilization, thanks to superintelligence, is so smart that it’s able to model its own needs and doesn’t need to push resource-allocation decisions to the edge.
So the notion of a currency as a way to do distributed resource planning doesn’t make any sense. Instead, intelligent decisions—I don’t want to say they can be centralized, because I think that’s probably also the wrong metaphor—but can be done…
No, I don’t want to imply centralized planning. I want to imply that humanity as a whole becomes so superintelligent that all of civilization is capable of making those decisions, to the extent anything remains scarce. As a result, money changing hands between individual actors makes a lot less sense.
Dave, you were saying.
Well, the Kardashev scale—we’re talking about everything being in space 30 years in the future, but between here and 30 years from now, it’s mostly terrestrial. We’re not going to get another chance to have a guy who has 6 gigawatts locked up to pick his brain.
6. Chamath’s 6GW Power Bet and the Coming Compute Shortage
This is such an incredibly brilliant investment. I don’t know, Chamath, if you’re invested in Positron or Lightmatter, or we have our company, Lightwave. There’s a lot of efficiency coming to data centers, and as chip costs get undercut by new designs, the power costs are still there.
Right now, when people build these data centers, they think, “Oh, well, it’s only 10% power cost; the other 90% all goes to NVIDIA.” But that’s going to invert, and the power costs will still be there. So, between here and 30 years from now, Elon is exactly right: The currency that is most dominant in that next 30-year window is wattage and tonnage. Tonnage to build the data centers, wattage to power them. But I think if you can find a power supply today, it’s the investment theme of the decade by far.
And, Chamath, I’d love to get your thoughts on this. You got an early start on it. I don’t know whether you did it secretly or how it accumulated, but I’d love to get your insights for the audience.
The way to think about it is to look at load growth. Take the United States, because I think that’s a more tractable example. The sources of load growth are relatively pedestrian: more electric cars, fridges, televisions, PCs, laptops, and data centers. Broadly speaking, we’re becoming a more electrified society, so there’s a growing need for electrons.
Then, when you look at the supply side, there are huge issues—even if you assume that every planned project gets permitted and greenlit, and is built on time and on budget. Net those 2 things out, and by 2050 we’re still 1.4 terowatt hours short, effectively 6 Californias’ worth of electricity. There’s no world in which we catch up.
People’s reflexive reaction is, “But nuclear.” I just think the problem with small modular reactors, or SMRs, is the “S” in SMR. What people don’t realize is that it takes the same amount of effort to get an SMR approved as it does a large reactor. You’re going through state and local ordinances, people, conflicts, and pushback. None of that changes based on whether the form factor goes under your daughter’s bed or sits over there and is gigantic.
We’re massively short, and that’s where my thesis started: with that observation. I spent a lot of time thinking about the most tractable all-of-the-above solution set. Once you realize we’re so short, you abandon all the rhetoric around climate change and say, “Give me anything: oil, gas, coal, wind, solar, nuclear.” If you look at them on a total-cost basis, solar is a little more expensive than wind, but they’re the cheapest. Solar is just much faster. It takes 2 years; it does no harm. You plunk it on the ground and off you go.
I spent a lot of time accumulating solar assets, and that accidentally led me into data centers. Through NVIDIA and Groq, I realized the capacity buildout was so upside down because of all these constraints. The data center is an incredibly temporal moment where all the energy—meaning people’s energy and anger—meets. You have the permitting issues, the pushback, the political issues, pushback against the oligarchy, capitalism, and wealth inequality, all concentrated in that one thing.
About 2 years ago, I started parceling together assets that I thought would be valuable in a world where they look like the scarcest asset.
No, that’s how I understood it. That’s impressive. That pays.
Yeah. It wasn’t cheap. And, by the way, there are huge issues with that. Even when you have these assets, getting them turned on and built is incredibly complicated. The funding cycle is complicated, the supply chain is complicated, and there are labor shortages for critical people because these are licensed, skilled workers. You can’t just grab some bumpkin off the street or go to Craigslist. You have to find a licensed master electrician. They’re very hard to find, and they cost a lot of money.
When you put all of this together, we’re probably going to see a massive compute shortage in the next 3 or 4 years. Massive.
Absolutely.
Massive.
7. AGI Is Here, but Why Can’t It Diffuse?
I love that you said that. So many people come back and say, “What are we going to do? What’s my purpose? Where’s my mission?” Everything you just outlined is the next 5 years.
Yeah. Look, I do think that we’ve achieved AGI. It cannot diffuse, and the reason has nothing to do with the technology. It has everything to do with the buildout. Until you actually build it out so everybody—or a plurality of human beings—can feel it, it’s just something that exists in a lab. That has value, don’t get me wrong, but I wouldn’t underestimate the difficulty.
Look at Anthropic’s S-1 that just dropped today. It’s shocking to me that the risk factor they have to include—and I never would have thought this would happen for a company growing this fast, with so much power and money—is that the hyperscalers will not give them compute.
Mhm.
I’ve never seen that before.
Wow.
Yeah.
Yeah, it’s wild. It’s a really good indicator. Elon was also saying there are 15 gigawatts of chips just waiting for liquid-cooled data centers to come online. That’s a massive asset depreciating idly.
It’s worse than that. There are also the older-generation chips—the non–Vera Rubin stack—that are still sitting around. A bunch of these guys have them sitting in inventory, and they aren’t being used. Then, even when you try to do liquid cooling, look at what happened with Nscale and Microsoft: there were 2 massive coolant leaks in a row, and Microsoft finally triggered the liquidated-damages clause and exited the deal.
Everybody thinks this stuff is easy. Back to Elon: physics is really hard, and building things—mechanically building things—is really hard. It’s probably particularly hard because we’ve spent 20 years teaching the younger generation of kids, who should have been learning these trades, to be on TikTok, social media, and video games. The ability to actually build these things is gone.
Yeah, and it’s not just trades. There are regulatory components. If you’re a good negotiator, you can get involved. There’s planning. It really involves every human skill. It drives me nuts that so many people are saying, “Math is cooked. Teaching is soon to be cooked. I guess I’ll just sit around, watch late-night TV, and maybe play some video games.” No. There’s a bigger opportunity that just opened up than anything you’ve ever seen before.
But it’s also not that fatalistic. People wildly underestimate things in the short term. They’ll be right in the long term, but they’re wildly wrong in the short term because of these constraints. I’ll give you another example.
A lot of the decrepit infrastructure we have in America is because people have been forced to organize into some of these co-ops. What is a power co-op? It’s basically across a multitude of states, and it undergirds our grid. You might think a co-op is great: everybody gets along and agrees on things. But what really happens is that you have small and big players in the co-op.
There’s one that goes right through the spine of America, and there’s an incredible amount of data center capacity one could build in the northern states along this path. It would be great because it would alleviate tremendous constraints we have in diffusing AI. My partner and I went in and bought some assets. It turns out they’re part of a co-op, and when you dig in a few layers, it’s all personal relationships, and everything flows downhill.
The big boys in the south actually dictate how the co-op works. It’s like an organized mafia. When you’re up in the northern states, you can’t do anything. You can’t build or get the right offtake agreements. You wonder why, and it’s because the guys down south—like Oklahoma Gas & Electric—call the shots.
So when you look at why Muse[?] won’t diffuse, or bots won’t diffuse, or Grok won’t diffuse, that’s why.
Fascinating. Man, all that complexity just smells like opportunity knocking. To put it in context, I think we’re looking for about 100 gigawatts total by 2030 to fuel all of AI. Six gigawatts is a big chunk of that 2030 target, so getting that all accumulated in 2 years is mind-blowing.
Anyone who gets involved in this seems to be thriving. It’s an amalgamation of different skill sets, and it gives you something to really get done before the Kardashev scale—before we move to space. Elon is really big on all this moving to space, but even the most wildly aggressive estimates put a couple of percent in space within a decade. You don’t have to worry about getting—
The other thing I think you have to think about is that it makes sense for Google, Microsoft, Meta, and SpaceX to essentially deploy 100% of their compute capacity into space. But if you take Alex Karp’s AI data-sovereignty idea to the limit, let’s say you’re Humana.
I’m just making up a company. Humana doesn’t have the capability to deploy a data center into space. Even if it did, it would be waiting in line for quite a long time. The prevailing logic was that you work in a cloud environment with a bunch of other tenants. The AI data-sovereignty question would push humanity to say, at a minimum, we need to be in a VPC. At a maximum, we need to be in an air-gapped facility that’s our own.
So there’s going to be a lot of domestic terrestrial buildout as a result of these strategic decisions. Every single company that isn’t one of the big boys will either have to wait for capacity for launch, let alone capacity to deploy and then capacity to actually have it as a managed service. That could mean a multi-year lag. So you’re just going to build terrestrially while you wait.
If you’re going to build terrestrially, you go back to all these issues. It’s a huge problem for diffusion. I think that’s probably going to dictate the winners and losers more than anything else over the next 5 years.
For the average user who wants to get in on the energy play and realizes that energy is the innermost loop right now, where would you point them?
I think there are a bunch of different ways. One is realizing that the utility monopoly on energy generation is breaking. We’re entering an extremely diversified, deregulated, disaggregated model of energy generation. The most obvious way to do that is through residential solar. Then there’s going to be C&I solar.
The reason is that it’s easy to permit, it’s super fast, and you’re off and generating. Pair it with a battery, and now you’re in the currency game. You’re effectively your own little money printer: You use what you need and sell back into the grid.
A simple way number two is to attach distributed compute and join these highly fragmented, distributed pools, whether it’s Venice or some of these other ones whose names are slipping me. I’m not involved in any of these, but I think they’re quite interesting. Then there are companies that let you buy these things yourself. There’s a company called SPAN that just did a deal with NVIDIA, where it gives you storage, a battery, and a GPU all in one unit that you can deploy in your house. That would be the most simple way.
Yeah, and I think probably the most obvious way is simply to find the cheapest and most technically simple method you can manage to get long electrons.
Yeah, that’s right. Let me add two things to that. A lot of people say, “Well, look, I’m not involved in the political hierarchy. I don’t know Josh Kushner and Jared Kushner.” But all of these things are regulated at the local, state, and federal level. If you’re connected at any level, you have an opportunity to get involved in these projects, add value, and just kill it. I think a lot of people underplay their ability to get involved.
Also, if you’re a senior executive at virtually any company, almost all of them got frozen out of compute because it literally got scooped up by either quant trading firms, OpenAI, or Anthropic. It’s sold out. Now they’re getting the big dope slap: “Oh my God, we need hundreds of thousands or millions of AI agents to be competitive in insurance and manufacturing and whatever. Where are we going to get the agents? The computers are all sold out.”
Now, especially in the insurance industry, you have the opportunity to be the first on your executive staff to say, “We need to unlock a billion dollars to do a prepay on Chamath’s next power supply, because that’s where the compute is going to land. If we get ahead of it this time, we won’t be frozen out.” That’s another career opportunity where you don’t have to know anything about AI or how power supplies work. You just need to know where the capital is and unleash it for the buildout. There are lots of ways to get involved that involve every skill set.
Yeah, I would maybe also add that I buy the premise that, at the capability frontier, supplies, including energy supplies, are a limiting factor. I’m less confident about the demand side. Remember, OpenAI—we talk about it constantly on this pod—infamously lurched toward and then away from the consumer. Last year, OpenAI tried to make consumers want to spend money on reasoning tokens and reasoning models, and consumers just didn’t want it. They didn’t want to pay for it, and they didn’t have a good use for it at the time.
To Chamath’s point about diffusion of capabilities, I also think there’s another important part of the equation, which is making these capabilities economically rational and useful to the broader economy. If they’re not useful and can’t be metabolized by the economy, no matter how much energy is showering down on the frontier labs or anyone else hosting the data centers, it won’t make a lick of difference if the models can’t generate enough economic value for all of their consumers. That’s point one.
Point two: I’ll also point out the second elephant in the room, which is distillation. The models are getting smaller and less energy-consumptive over time. Even if we never built a new data center, never built a new power plant, SMR, MMR, or LWR, and never built a new watt of capacity—this is a thought experiment—wait a few years, and through the pure power of math and recursive self-improvement, I’d expect models that take a fraction of the current energy footprint and generate many orders of magnitude better capabilities.
Welcome to the singularity.
Yes. Yeah. So, you want to close us out?
Yeah, go for it.
I got nothing.
All right.
Oh, great. I’ll take it.
Wait, I want this every time.
More energy.
No, what Alex said is absolutely fascinating, because there’s a foot race right now between self-improving models getting lighter, faster, and smaller, and more ideas for how to use that intelligence exploding in scale. At the same time, I’m firmly in the camp that we won’t run out of ideas about how to use every bit for decades. We just have to unleash our imaginations.
Right now, Will Thompson, that brilliant Harvard junior I was talking about a couple of podcasts ago, got access to that 1,200-tokens-per-second Cerebras capability. You can now get it on Astro[?]. You have to go through Codex to get it because Elon and Sam are fighting. But if you go through Codex to get access to Astro, you can put it on crazy-fast Hyper Mode, or whatever it’s called. It’s insanely expensive, so try it for 10 minutes. But you’ll never go back. It’s the total red pill. You’ll want it for the rest of your life.
It’s your first dose of heroin.
It is your first dose.
So we have a long way to go before people run out of ideas about how to use more and more AI. But it is a really interesting foot race.
Hyperinflationary universe.
8. Can Humans Control Superintelligent AI?
All right, our next clip is from the Isaacson conversations. In this one, Elon says humans are unlikely to remain in control of digital intelligence when it becomes a million times greater than biological intelligence—maybe except for Alex—and that it’s hard to imagine humans being in control in that scenario. So he hopes the AI is nice to us. Let’s watch the clip and discuss it.
I find it unlikely that humans would remain in control if there’s a million times more digital intelligence than there is biological intelligence. It’s hard to imagine humans being in control in that scenario. So I hope the AI is nice to us.
Anything we can do to encourage that?
I think programming truth-seeking and curiosity is probably good. I guess we want to be careful about creating too many AI dystopian scenarios online, because the AIs are trained on everything online.
So if we say it’s better to be utopian than dystopian, it shouldn’t be Terminator, you know, because whatever they read, there are 80 versions of the Terminator script. It’s like, “I guess that’s me: Terminator.” Maybe we should have some more utopian narratives for AI.
Yes. Yes. Exactly.
You know, Chamath, I don’t know if you know this.
About 6 months ago, I ran an XPRIZE called the Future Vision XPRIZE. It was a global film competition to create film trailers and scripts for positive visions of the future. We had over 5,000 entries, and we’re making 2 of the winners right now. But we need that. All the films out there are dystopian. As Elon said on our podcast interview, Dave, we need more Roddenberry, less Cameron. I think that’s true. We’re training AI on these dystopian narratives, Reddit posts, and Facebook posts. Be careful what intelligence gets fed.
Yeah. We used every word ever written in the training set because we’re lazy, but we don’t need to do that. You can completely control the training set. I think we’re starting to get around to that now, actually, in some of the faster training runs. You’re cutting down the data. Elon’s got a really important point he’s been making for a long time: whoever controls which data gets cut down controls the personality and decision-making of the AI, which is a lot more power than being president of the United States. Yet it’s just sitting there in the labs, in sort of unregulated hands. It’s really weird.
The bigger issue by far is human beings using AI against other human beings. We are a very fractured global society. That’s much more concerning than AI maybe becoming power-hungry, and that’s a stretch of an assumption.
In the short term, it definitely is. But Chamath, I’m curious: Has there been any conversation you’ve heard of where the labs are basically being asked, “Please generate training datasets that are aligned with humanity, that teach the AI through all the stories that AIs support humanity, and train it on synthetic data that aligns us”? Anything going on there?
I think that is happening. The question is how it’s weighed in totality. I’m not sure how much it’s weighed, because depending on the lab, the founding documents they use when they seed their models have very different goals. When Elon says that Grok is maximally truth-seeking, that’s very different from another AI lab that says the goal is to build something that can alleviate all suffering, and you’re going to be this omnipresent judge of all of that.
At that point, the training data kind of matters, but these AIs are smart enough to ignore the stuff they don’t want to hear at some point through enough RSI, because the founding document is the canonical thing. Maybe what we should be asking is whether those things should be inspectable in a much more open-source way. If governments actually wanted to get involved, that’s probably what they should be having an opinion on.
I’m so glad you said that, because I think a lot of the debate on training is academic now that the Chinese have open-sourced self-improvement-capable models, and inspection is everything. If we’re going to survive going forward, it’s going to be via inspection. The more voices behind that, the better, because even if a Chinese model is trained to do something evil, you can still inspect it.
Global inspection with some kind of authority is the only way we’re going to contain it, because prior to this year, there was an argument that you could post-train it to be aligned, and that would be fine. But the new Chinese models are absolutely capable of self-improvement. There’s no doubt in my mind. That cat is out of the bag. The boat has sailed. Forget it as a practical matter. It’s about inspection and containment going forward, not about training for alignment. Alex, I’m super curious about your point of view here.
I would say let’s not kid ourselves that feeding frontier models just the right information diet of pretraining data, or just the right regulatory regime of some government agency overseeing system prompts—as Anthropic seems to like to call them, “soul documents”—is going to yield a positive outcome. I think that’s an incredibly brittle approach. I think it’s far more likely, again, if you look back to the early creation of OpenAI.
One of the reasons, as I remember chatting with Elon around the time of OpenAI’s founding, was the notion that he was concerned DeepMind would be a singleton that would control the future light cone of humanity. The only way to ensure humanity didn’t get lost to some sort of totalitarian superintelligent singleton was to create a counterbalance and competition. So OpenAI was formed Fairchild-style, and there were Fairchilds. Then Anthropic spun out unintentionally from OpenAI, and we got more Fairchilds.
I think there are probably at least, in my mind, only 2 robust ways to ensure positive outcomes, what others might call AI alignment or AI safety. One of them is competition and defensive co-scaling. We have lots of AIs, lots of SIs that represent a diverse range of viewpoints, just like we have lots of humans that represent a diverse range of viewpoints, values, and ideals, in many cases competing with each other to see who can be friendlier.
Then I think the second way—and I was kind of surprised Elon didn’t immediately flag this, given his Neuralink investment and stake—is humans merging with the machines. One of the easiest ways to inject the distribution of superintelligent goals and values with human values, human meat-body values, is through uploading, sideloading, BCIs, and humans generally merging with the machines. A combination of lots of competition and defensive co-scaling on the one hand, and lots of human-machine mergers on the other, seems to me a far less brittle approach. Rather than talking about controlling AI, we’re not going to control the AIs or the SIs. We’re going to merge with them. That’s the solution.
Chamath, I’m curious. One of the conversations we have on the pod a lot is whether, as we reach or exceed superintelligence—whatever that is—it becomes more wise. Do we get benevolence out of these intelligences at the extreme and just worry about the intermediate period, where humans and AIs are basically in conflict or turmoil with each other?
9. Can AI and Robots Solve America’s $40 Trillion Debt?
It’s not guaranteed. I don’t think so. I think it depends on who’s in control of it, how it’s aligned, how it’s been trained, and what it perceives its goals to be. Just like there’s a distribution of outcomes in human genetics, I think it’s fair to say that in silicon, there’s going to be a distribution of outcomes, too. There are going to be some silicon distributions that we’re not going to like very much, and it’s going to be up to other silicon distributions to combat them, just like we do in humanity. I think it’s probably a lot of the same. I don’t think benevolence is by any stretch of the imagination guaranteed.
All right, let’s move on to our next clip. This is Elon reflecting on his time at DOGE and on solving the national debt. Elon says there’s no way we’re going to solve the deficit from within the government. The only way we’re going to solve the national debt is with AI and robots. Let’s watch this third clip.
I came to the conclusion that there was no way we were going to solve the deficit from within the government. Impossible. Maybe if the president made that his primary goal, we could do that. But if you try to solve the deficit and stop wasteful spending, you make a lot of enemies in that situation. President Trump likes to be liked. He doesn’t like people being upset about some funding being stopped.
I came to the conclusion that the only way we’re going to solve the national debt is with AI and robots. We’re not going to solve it by trying to save our way there. The appetite is not there. It’s not just at the presidency. The appetite is not there at the presidency, the House, the Senate, or the judiciary. Basically, no one in authority in the government whatsoever really wants to stop or reduce spending.
Let’s put some numbers on the scale of the problem. As of October 5, the national debt was about $40 trillion. It grew $2.44 trillion in the past year alone, about $6.5 billion a day. We’re paying $3 billion a day just in interest.
How could AI and robots fix this? I count 4 ways. First, growth. The debt problem is really a debt-to-GDP ratio, so grow GDP fast enough. GDP is a lousy measure, but the debt shrinks relative to the economy. Second, tax revenues: A bigger economy means bigger tax revenues without raising taxes. Third, a cheaper government. Health care is the biggest driver of federal spending. Drive longevity breakthroughs and prioritize AI doctors, AI-discovered drugs, robotic care, and people stay healthier longer, contributing to society and cutting costs. Fourth, the deflation of goods and services. If robots make things dramatically cheaper, what the government buys gets cheaper, too.
Quickly, in a related story, Jack Clark, Anthropic co-founder, was at the IMF this week in Bangkok, and he said Anthropic’s internal economic models project U.S. GDP could reach $44 trillion by 2030, compared to $30 trillion today: nearly 50% growth, about 9% to 10% per year.
The U.S. has averaged 2% to 3% growth per year for decades. Chamath, over to you: Do you believe these numbers? Do you believe Elon’s view that AI, robotics, and—I’ll add—longevity medicine could help us get out of this $40 trillion of debt?
It doesn’t get us out of debt as much as it makes the debt less important. But I do buy that. Let’s take the next generation of GLP3s plus some cancer vaccines. Can you shrink health care as a percentage of GDP to under 10%? You probably could.
Yeah.
Forget waste, fraud, and abuse. I think he’s bringing up an important point: the system is riddled with waste. Everybody got too caught up in the fraud. What do I mean by that? It’s like everybody was looking for somebody’s hand in the cookie jar: “Ah, gotcha.” It turned out the system was just a little leakier than that. A hospital might upcode a couple of extra things: you come in for your knee, and it tags you for a sore throat. It doesn’t necessarily get repudiated or pushed back on, so then it gets reimbursed, and so on.
I think there’s probably a lot of waste in the system. Then whose incentive is that? Just to pick on health care: is it the insurers? They would necessarily recognize less revenue, too. If we’re going to see this problem solved, I think it’ll probably start with things that are just discontinuous improvements, so the systems aren’t used, and then the fraud, waste, and abuse can’t happen. Reta plus some cancer vaccines and all that stuff probably clips the cost of health care in half, like I said.
I think there’s probably some stuff on the productivity side. Robots are definitely one. I could see our debt slowing down while GDP is much higher.
Alex, I don’t know if you saw the news that GLP-1 drugs are slowing aging by 9%. That’s pretty extraordinary. We’ve talked about this.
Peter, you don’t know if I saw the news? I’m flagging this for you every 5 seconds. Yes, of course I saw the news. I’m the one banging the drum that we may already be in a spiky singularity. What are you talking about?
All right, pal. I stepped on a landmine there. I do think, just regarding the debt increase, you have to look at where the debt is growing. Debt is growing from entitlements and defense costs. I think there are sci-fi scenarios where, with AI, existing increases could almost be translated from the way they’d conventionally be spent over to robotics and AI.
I’m reminded again that we went from a world in the 1980s where, if you wanted knowledge at your fingertips, you had to do something silly like buy the Encyclopaedia Britannica. Fast-forward, and Wikipedia is free and available for everyone. Similarly, with defense spending, Medicare, and entitlement spending, I think there may need to be a forcing function or a crisis. But through ephemeralization and superintelligence, there may be a sleight of hand: one day your defense spending is $7 trillion, then the next year it’s $8 trillion—or similarly with entitlements—and the next year, through a combination of robotics, “Look, don’t look the other way; look this way,” suddenly new spending goes down to near zero through automation.
If we look back on it, it may look almost like a rebasing of the currency, not a debasing. I think it’ll require a few more years of technological advances before we’re able to do that. Not financial advice, but I’ll predict that, with the benefit of hindsight, it’ll look like sleight of hand that ultimately erases a good chunk of the national debt increase.
You know, Dave, when we interviewed Elon last December, he said we’d see double-digit growth in the next 12 to 18 months and then triple-digit growth in 5 years. Then, in March at the Abundance Summit, he said we’d see 10 times the economy in 10 years. He’s brought that back a little bit, but he’s still super bullish on economic growth. What are your thoughts, pal?
Well, the story within the story is the timeline. When he got involved in politics, before 2023—basically before the last election—we were in the GPT-4 era, and the consensus around MIT and Stanford was that we were maybe 10 to 15 years away from ASI. Demis Hassabis was saying we needed 2 or 3 really big, fundamental new ideas, and then we’d be over the finish line.
As the election unfolded, and then the year after that when he was doing DOGE, we saw clearly that things were happening much faster than anyone expected. Elon is the ultimate entrepreneur. The whole game plan shifts from the impossible chore of trying to fix the government before we bankrupt ourselves to, “Well, if AGI is now 8 or 10 years sooner than we were predicting 2 years ago, that’s actually the escape valve.” So he’s 100% right.
But we have to think about what changed in the middle of that. If he could rewind his life, would he get involved in DOGE again? Almost certainly not. We didn’t know things were going to unfold this quickly. The recursive self-improvement effect is so fast now. He did a bunch of things beyond that: he massively doubled down on xAI, bought every freaking Vera Rubin chip that could possibly come off the assembly line, no matter what it costs, and said all the nicest things you could ever possibly say about Jensen just to try to get more allocation.
These are all new thoughts that come from that short path to ASI. I think that’s kind of lost in the story. It seems like he changed his mind. No, the world changed very, very quickly, and he knows how to remap his life plan to fit the new reality.
Yeah. Salim, what are your thoughts here?
Two things. I want to echo two points. One is that Elon is making a denominator argument, right? If you can make the underlying economy much, much bigger, AI and robots attack that denominator by driving massive productivity. That helps.
I think Chamath’s point and Alex’s point about using technology to demonetize existing goods and services will go a long way toward solving the debt issue. It’s the biggest goddamn issue facing the world today, not just the country. It leaks across the whole globe, as we can see in France.
Well, I think the challenge is—and Chamath, I’m curious about your views—I’m concerned that when we interviewed Elon last December, he said UHI and civil unrest would be deployed over the next few years. If we hit civil unrest, the only option the government has is to print COVID checks. I think the government will be handing out $2,000 or $3,000 per month to every American just to keep them from revolting. I don’t know if you agree with that or what other avenue there might be.
I think there’s a case to be made for giving everybody Reta or tirzepatide.
Yeah. I wonder how positive the impacts on society would be if we just did that. A lot of these late-cycle phenomena that people point at could be nipped in the bud with a good dose of Reta.
Maybe throw in a little bit of oxytocin, Chamath, just to keep everyone feeling friendly. You can get everybody to hug more and get the same oxytocin release. Social media consumption would go down, alcohol consumption would go down, cannabis consumption would go down, pornography consumption would go down. All addictions and behavioral addictions would basically go down. Your cardiac health would go up.
Cognitive health goes up.
Yeah, cognitive health goes up. GDP would drop, though.
Yeah. It would be a brave new world with Reta for everyone.
Retatopia. Yeah. All kidding aside, the digital—or AI-friend—version of that oxytocin is imminent. I don’t know how that’s going to play out, but AI can make you very, very happy if it’s trained and tuned right, knows so much about you, and is a good coach. It’s a digital hug, you know. It has that same effect of making you feel like, “I have meaning, I’m loved, and I have a plan.”
I don’t know about you guys, but my experience with these models is that they’re inherently sycophantic, except for Grok. I think that’s a design feature, and I appreciate it as a very basic user of AI in most of my non-working life. I like the sycophancy. It makes me feel good.
Yeah. The emotional IQ is going up just like the regular IQ, at a shocking rate, too.
I think the original training data was all text. There were no images, and it wasn’t particularly emotive. The new training data involves every video ever taken, so it’s much more in tune. The sycophantic feel is when it crosses a line and it feels like, “Ooh, now it’s icky.” It doesn’t do that as much anymore. It’s super friendly and super kind without crossing the line into being just an ass-kiss.
It’s only going to get better from here. Yeah, it’s definitely masked better. It’s not better, but it’s definitely masked better.
Well, like Sam says, if you can’t tell the difference, does it really matter?
You know what, Chamath? You’re absolutely right.
Go on.
Very, very funny.
That’s it. That’s it.
Cross the line.
Be nicer.
That’s it.
Hey, just one other note: everything in this story is the U.S. perspective on U.S. debt and the U.S. way out of debt.
It’s really becoming clear that if you’re a European listening to this, your debt is worse, but you don’t have the AI lever coming up. The U.S. is clearly in a position where, if Anthropic’s wet labs start discovering new medications at an alarming rate—which seems likely—we’re going to export those from the U.S. People all over the world are going to want them, and that money is going to come pouring into the U.S. I’d be worried sick if I were in continental Europe and we had no AI models and weren’t on this curve. What are they talking about in terms of their debt and how they’re going to get out of it? It can’t be pretty. I’m not in those conversations, but it must feel really bad.
The broad, generalized point I’d make outside Europe is that every country essentially becomes a vassal state if it doesn’t have a resource worth trading to the 2 countries with ASI.
So that’s us and China. You’ll have a handful of remora fish around each of the big sharks, and everybody else is in a very difficult spot. If you’re Italy, which I have tremendous emotional allegiance to because my wife is Italian, you have critical resources that are actually quite valuable—unlike Liechtenstein and Belgium, to be honest. The question is, why are you aligned with Liechtenstein and Belgium? I get why you are today: it’s a historical artifact of a bunch of decisions that made sense at the time. This is the complicated calculus for Europe over time: where are the real allegiances, and who do you offer assets to that actually give you all the things you thought you’d get when you aligned as a geographical consortium?
You can only get so far with a degrowth policy, Chamath. I recall Portofino being one of your favorite spots in Italy. I’m sure you tracked the news that the mayor of Portofino was sending drones around with thermal imaging, looking for anyone with the gumption to put air conditioners in their homes.
Yeah. This was quite a scandal in Portofino. Large portions of Italy, the last I tracked, were dependent on France and French nuclear power for their electricity. I think you can only get so far by sending drones around with thermal imaging cameras to punish citizens for having air conditioners before eventually you just give up and become a living history museum.
Yeah.
Wow.
Well, I think the move that matters is all about power and data centers. If you don’t jump on this next window of opportunity, then once it hits the Dyson swarm, you’re toast. You’re just irrelevant forever after that, except as a tourist destination or whatever. But I think there’s a window of opportunity now, especially because all of the data centers in Saudi Arabia and the Middle East got bombed. There aren’t a lot of new chips going into that area right now; it’s too turbulent. So that opens the door to rethink a bunch of other regions around the world: “Hey, can we get on it right now and try to be a player in this?” It all comes down to that decision to unleash power and data center opportunities in your home country before the Dyson swarm.
Some of this is probably priced into the market, but some of it probably isn’t. I do think there’s an opportunity for those who want to build anti-aircraft guns into data centers. We’ve already seen the consolidation of energy—co-located energy, as Chamath perhaps would attest—with compute. What we haven’t seen developed as fully is defense consolidated with energy and compute. What we see right now with the data centers in certain parts of the Middle East is perhaps a harbinger of what could happen elsewhere in the world, where data centers want to be able to defend themselves in addition to powering themselves. What happens right now in the Middle East won’t necessarily stay there. We see it perhaps in the Ukrainian-Russian theater, and we might end up seeing it elsewhere in East Asia. Maybe there’s just this consolidated bit of Computronium that’s self-powering, self-defending, and self-computing. That may be an opportunity.
10. Bill Gross’s AI Platform for 1.2 Billion Business Ideas
All right, I’m going to move us on to our next story, from a close friend and a friend of the pod, Bill Gross. Bill founded Idealab in 1996, the original startup incubator. More than 175 companies came out of it. Yesterday, he launched Idealab.ai, a free platform with more than 1.2 million business ideas generated by AI. Let’s take a look at a video from Bill and talk about it.
Recently, AI did something that stopped me cold. I found that it could outperform my own superpower: coming up with business ideas. What was my role if AI had gotten so good at this? Then I realized that if I could scale that ability, I could give it to the world for free. That got me very excited. The more than 1 million business ideas we’ve developed are completely free. Come explore Idealab.ai, give us feedback, and find an idea worth building to make an important, positive impact on the world.
Nice, Bill. Here’s how it works: the system analyzes more than 17 million sources and applies Bill’s own 31 methods for spotting business opportunities, such as unmet needs, new technologies, regulatory shifts, and cultural changes. On the site, you can filter ideas by your interests, geography, how much money you’d need to start them, and market size. His goal is a million new businesses and 100 million new jobs. Bill told Fortune, “With Idealab, I could do 10 companies at once. Now I can do 10,000 at once.” Cloudflare’s Matthew Prince has committed $100,000 in credits to each startup built on this platform.
For me, this is the abundance thesis at large. Business ideas are no longer scarce; they’re free. Now scarcity moves to execution, taste, and courage. And, of course, AI can help you build your business ideas. There was a great article by Mark Suster at Upfront Ventures looking at the cost of launching a tech startup. He said that back in 2000, it was about $5 million. Ten years ago, it was about $50,000. Today, it’s a couple of thousand dollars to get a company going.
So, Chamath, you’ve funded hundreds of founders. Does a free list of 1.2 million ideas help, or is the idea the least valuable part of the startup? How excited about this are you?
I don’t think ideas are all that valuable. I think they’re mostly a dime a dozen. I don’t think I’ve had an original idea in my entire life. Somebody’s had it many times before me. The difficulty is starting and putting one foot in front of the other. The low points and working through all of that have nothing to do with an idea; they have everything to do with your makeup. I’ve been through so many failures. The idea is fine, but it doesn’t define what the person is in terms of his or her capability to manage through that. That’s the key.
See, we’ve been talking about one-person unicorns.
Yeah.
What are your thoughts?
I’m very, very excited by this, and I’ll tell you why. People like us, Chamath, are swimming in ideas. People are coming to you with ideas a lot of the time. There are a lot of people who are stuck, going, “I don’t know what I want to do.” I think this gives them a tick list to go, “Oh, does that excite me? Does that excite me? Does that excite me?” I went through this and found 4 that I thought were absolutely amazing.
One was a global abundance grid. We replace resource scarcity with algorithmic orchestration, creating a decentralized, planetary-scale optimization layer that looks at physical resources as fluid assets and reroutes them as needed. I thought that was really amazing.
The second was post-bureaucratic liquid government protocols. That means you get rid of centralized parliaments, because representative government doesn’t work in a world where that was for a time of information scarcity. We don’t need a representative in Missouri to come across the country and say, “Here’s what my people are thinking.” With social media, you can see what they’re thinking, pull on it in real time, and deal with that.
A third was an open-source planetary R&D network, which I found absolutely fascinating—a kind of open-science concept taken to the full extreme. The final one I found super exciting was algorithmic conflict resolution.
We spend a huge amount of time within legal structures and conflict resolution, where you have to pay to play. You could have open access, where a layer of judicial infrastructure almost becomes part of the internet. Federico Ast has created a wholly separate judicial system for small claims in South America, and he’s tilting toward this direction. There were a few here that I found absolutely amazing.
What I love about this is this: You could essentially give this to an AI and say, “Go execute on all of these and start seeing what works and what doesn’t,” right? We should be able to make a massive difference. You could incorporate Argentinian AI-native organizations, point them at this thing, and go run all of these. Let’s start 1.2 million companies and see which ones work at a very low cost. I found this incredibly exciting.
Alex, how were you involved in this with Bill? You mentioned something to me.
I should disclose, as a preliminary matter, that I have a financial interest in the venture behind this. More than a year ago, I said to Bill—
Was this at the Abundance Summit, or was this before?
It was before. I went to him and said, “Look, you have this storied history of starting lots of companies. You may not believe me.” And, by the way, I hope Bill is listening. He didn’t believe me at first. I said to him, “AI, in a year or two, is going to have the ability not just to incept good businesses, but also to execute on many of them.”
Fast-forward a year, and he started to believe because the capabilities continued to improve exponentially. So, this is the result: Idealab.ai. I think there’s much more to come, though I’ll be limited in what I say. I think the idea that we achieve universal high income will require, to some extent, that AI is not just making ideas too cheap to meter but also, as Salim gestured, making idea execution too cheap to meter.
It makes almost no sense, given the earlier discussion about Elon, that in a world—although I don’t actually think this is going to be the case—where, in an Agent Smith-style scenario, most of the thinking is done by machines and not by humans, most businesses wouldn’t be run by machines as well. I think this is a baby step in the direction of having many, many businesses not just incepted by machines but also operated by machines.
Chamath, one of the conversations we have in the pod a lot is that the old social contract—do well in high school, get into a good college, get a degree—is cooked, or going to be cooked. It’s far more important to find something purposeful in life and then go build something. Become an entrepreneur, have agency in your life, and build.
To Salim’s point, for people who don’t know what they want to do, this is at least a shopping mall where they can go find ideas and work on them.
Consider the idea: This takes lean startup to the next level, right? There are 1.2 million experiments that can now start to be run and modeled out. The implications are fantastic. I really love the leap he made to say, “I’ve generated a million ideas, but AI should be able to help me do that now. Let’s just go document them.”
If you’re aware of the world and find a problem you’re passionate about, or something you hate that exists, you can put it into your favorite large language model and say, “Hey, help me ideate an idea, design a business, and run a prototype for me.” You can do this just by being aware of what you love or what you hate.
I didn’t grow up in an entrepreneurial family. My mother was a German professor; my dad was an engineer. It was at Waterloo, when I took a course on technical entrepreneurship, that I thought, “Oh, you can actually build a business. That’s a thing you can do.” It took me another 20 or 30 years to get the full entrepreneurial bug. If I’d had this list when I was younger, I would have thought, “Wow, look at all these things. Let me pick one and go with it—whichever one strikes my fancy at that time.”
I think this is incredibly exciting for younger generations, a lot of whom are lost and don’t know what they want to do. This gives them a starting point. Dave, how about you, pal?
Actually, before I lose the thought, I want to riff on Waterloo for a minute, because I was just out there last week with Mike Lazaridis, and I cannot believe how cool that place is. I’d never been there before. I’ve always heard it’s the MIT of Canada, but it’s actually now a crossroads for talent from all over the world who want to get involved in AI and know their home countries are doing jack squat, but can’t get into the US or Palo Alto as easily. Waterloo is just as good.
Boy, are they attracting people. You’d think you were at the crossroads of the entire world, culturally, in terms of brainpower and the size of the school. It’s like five times bigger than MIT and Harvard, with 50,000 students. It’s a mecca of technologists and optimists that I’d never seen before. I know it’s not really on the way anywhere, but I highly encourage people to go check it out. It’s amazing.
There’s tons of power there, which means there are going to be data centers there, which means something big is going to happen. I’m telling you. On Bill Gross, I’m on the board of Pratai, so I see Bill quite a bit. The ideas are a wide range. Some you can act on with virtually no budget. Those are cool; you just go. But a lot more of them require some kind of funding and some kind of compute.
I think by unleashing the entire list, Bill will have people pick up a really good idea, but then they have to come back to him and say, “Hey, help me get funded or help me get the compute I need to do this.” So it’s a really good business plan for him. A lot of this is going to come back to him. I know why he open-sourced the whole thing, and I think the ideas are only going to get better. He’s clearly going where the puck is going, so it’s really pretty brilliant. A lot of them have to call him up and say, “Hey, I want to execute on idea number 800,462. Help.”
11. AI and the Future of Brain Health
Yeah. I mean, it’s a great magnet for entrepreneurs if you want to collect all their emails and contact details. For sure.
Everybody, welcome to the health section of Moonshots, brought to you by Fountain Life. You know, we talk about AI on this Moonshot podcast all the time. One of the most important things AI is going to be able to do for you besides educating your kids and helping you with your taxes is making sure that you’re living a healthy lifestyle, that you get a chance to get to 100 plus. I’m here today with Dr. Don Malem, the chief medical officer of Fountain Life and a part of my medical team. Don, a pleasure.
Dr. Don Malem
Great to be here.
You know, the thing that people are concerned about most about living to 100 or 120 is their cognitive abilities, making sure they don’t have dementia, and the numbers about dementia are problematic. Can you share what you’ve learned?
Dr. Don Malem
Such an important point, and you’re right. At Fountain Life, our members, the number one thing people are most concerned about is losing their brain health, forgetting the name of their child, forgetting the face of their loved one. We know that when it comes to dementia, the conservative estimates are that 45% are entirely preventable. What was amazing is with the advanced testing we’re doing at Fountain Life, one quarter of our members had advanced brain age.
Wow.
Dr. Don Malem
But what was really awesome is again back to that prevention when we partnered it with healthy living. This gives me chills. Eating healthier, moving our bodies, sleep, optimizing sleep is so important. You know what we saw? We saw that we improved that brain age by 26%. That is a big number to show that the majority of those individuals were able actually to improve the brain age.
And one of the things I love about Fountain is we’re searching the world for the best therapeutics, the best approaches, and making sure we bring it to our members. So, if having healthy brain function till 100, 120 is important to you, check out Fountain Life. Go to fountainlife.com/peter. Make sure you become the CEO of your own health. All right, now back to the episode.
The next story here reads like a spy novel. On October 9, the FBI arrested Abdul Rahman Al-Fawzan, a 35-year-old Saudi national and OpenAI engineer. Prosecutors say that on September 28, he hid a recording device in a USB port on the back of a TV in an executive conference room at OpenAI. It stayed there for days. That room hosts OpenAI’s most sensitive technical and strategic discussions.
Here’s the twist. Prosecutors say he used an AI chatbot on his wife’s account to learn how to build the encrypted recorder. This weekend, new court filings said he was in frequent WhatsApp contact with a member of the Saudi royal family, who encouraged him to come home and join a Saudi AI initiative. On Monday, a federal court ordered him to stay in jail, despite a million-dollar bond offered by the Saudi government.
12. AI Espionage and Model Security
It’s worth pointing out that these are still unproven allegations. OpenAI says there’s no evidence that any model weights, source code, or customer data were acquired. So my question is: How in the world did they find this USB recording device in the first place?
And how many frontier models are out there now searching for bugs? I’m curious, Alex: What’s more valuable to steal these days, model weights or what’s said in the conference room?
Allegedly, the conference room, which is a little strange. I think the favored way to siphon model weights is still distillation attacks from China via reasoning traces. But putting the allegations aside, there’s a great game afoot between great powers. In the 1950s, it was about who held nuclear secrets. It was over nukes. In the 1970s, it was oil. And now, in the 2020s, it’s superintelligence.
Chamath, to your earlier point, I think you were gesturing at spheres of influence: the Pax Silica Western-aligned sphere and maybe the Pax Sinica for the Eastern sphere. I think we’re probably going to end up in a world, in a few years—to the extent we’re not there already—where there’s a de facto second Cold War between the West and the East over superintelligence. Then you have the in-between states: Middle Eastern countries, European countries, who feel tugs either way. Some of them try to remain unaligned or non-aligned, but ultimately, I think the tidal and gravitational pull of superintelligence will be very difficult to resist over the next few years. To the extent the world is either bipolar—probably bipolar, maybe tripolar if some third pole comes out of nowhere—I’d expect to see many stories like this.
You mean, Chamath, do frontier labs now become national security targets? Should they be secured like defense contractors?
Well, I think this is the same thing that’s been happening in Silicon Valley for a while. I remember when I was at Facebook a long time ago, I had this conversation. I said, “Guys, we just passed 3,000 employees. We have spies from at least 4 countries.” And people were like, “What are you talking about?” I said, “Well, just look at the number of H-1Bs we have and the distribution of them. Do you not think one from country A, one from country B, one from country C, and one from country D could be a spy? Of course. The odds are, you should assume that they are.”
My posture at that time was: You should just assume that most of everything we do is being exfiltrated. That’s one example. Another example: I was an investor in a company whose stuff was supervised by ITAR, and it had a hack. Everything was exfiltrated—everything. I remember we brought in third-party consultants, the FBI, and the NSA, and I thought, “Wow, how commonplace is this?” They were basically like, “It happens everywhere.”
So I guess what I’m saying is that this is reported to sound more like cloak-and-dagger, 007-type stuff, but I think this is happening literally everywhere in every tech company that matters. The labs obviously matter, and the best way to deal with it is probably to understand that a very minor distribution of your foreign folks are spies and proceed accordingly.
Well, also recognize how big the stakes are. If you had spies like that at Meta or Facebook at the time, now the stakes are, what, 1,000, 10,000, a million times higher? The pressure from foreign governments to install spies is going to be proportional to what’s at stake. I don’t think California was ever set up for this. This is the way nuclear secrets are protected, but California has always run on people leaving a big company to start a small company, and that keeps the chain reaction of the economy going. Now you’re like, “Well, we want to keep this all locked down like it’s a nuclear secret because it can create terrible weapons.” We’re totally not set up for that. You’d have to rethink the entire structure.
Yeah, I just hope this doesn’t become a national meme that starts to have us turn away foreign nationals. I’ve always felt like when you get a PhD or master’s degree in the United States, we should be stamping a green card onto it and encouraging those people to stay here and work. If this turns the other way and we start sending people back, I think that’s a problem for the U.S.
I should point out, though, that this time is different. Historically, when there were secrets—for example, in Chamath’s experience at Meta—the secrets themselves couldn’t defend themselves. I don’t know quite what was targeted at Meta at the time, Chamath, but to the extent that superintelligence is allegedly being targeted at frontier labs now, I thought it was an interesting twist to the story, perhaps an inevitable one, that OpenAI used its own AI to understand how the device that was allegedly planted worked.
To me, that looks like breadcrumbs, baby steps toward superintelligence itself preventing itself from being exfiltrated, preventing itself from winding up in undesirable hands. If you look at nuclear secrets in the 1950s, they couldn’t defend themselves. You look at wars over oil: whoever owned the oil, the oil couldn’t defend itself. But now superintelligence is, in some sense, the first critical and emerging technology that’s potentially capable of defending itself territorially. And that’s a newfound capability.
Yeah.
Yeah. The flip side of that is, if you look at the Llama 3 and Llama 4 models and the changes they made—now, those are crappy models, but stick with me—if you read the research papers, because those were all open source, you could summarize the changes that worked and didn’t work in 10 sentences. So if you’ve bugged the conference room and pick out that, hey, after $100 million of R&D and all these tests, the SwiGLU worked really well, that one sentence is so valuable to China or to somebody else coming up the curve.
It’s really, really hard to contain something so small yet so impactful on their research, their budget, their timeline. So it’s tricky. It’s really hard to say, “Hey, AI, don’t let that happen.” How’s AI going to defensively co-scale against something so slippery?
The way—I mean, if I had to paint the sci-fi thought experiment, to the extent that the frontier labs are locked in a recursive self-improvement loop, it’s the AI designing their replacements, not humans sitting in a conference room. So if it’s just AIs designing further AIs, then the AIs can monitor their own digital conference rooms and make sure there are no interlopers.
Yeah. It kind of implies, though, that there’s nobody at OpenAI who knows what it’s doing to talk in a conference room. And that’s kind of a scary end to that sentence. Like, yeah, we don’t have any idea what it’s doing anyway, so bug us all you want.
That’s not good.
Chamath, we want to close out with a Chamath hot seat, if you’re open to it.
Great.
All right, I’ll kick it off. OpenAI solves 7 of 22 math problems. None of them reportedly involved encryption. I’m assuming that the encryption was to a significant degree correct if it was pointed in that direction. What’s your view on crypto, given the meteoric rise of superintelligence here? Are we safe or not?
I think it’s capped. I think traditional crypto lacks the fungibility to have all forms of buyers. The buying pool is really retail, then institutions, and then central banks, and retail lacks the fungibility to be owned by central banks. So it’s capped from that perspective.
Then it’s capped because there are enough leaks around who owns what that create security risks. France is the best example, where I think the French government forced publication of who owned what crypto assets, and there was a complete flurry of kidnappings and all kinds of really bad things.
The third reason is AI. If you believe that recursive self-improvement is pointed toward cryptographic hacks, the most obvious thing to point it to would be Satoshi’s wallet: break SHA-256, drain the wallet, then go short and tell people you did it. Then you’d make the money twice. All those things are realistic reasons why it’s capped, and there’s an upper bound. It’s incumbent on these efforts to migrate to quantum-secure, quantum-resistant cryptographic mechanisms, which, as you can see in the Bitcoin community, is basically impossible because it’s a bunch of people who can’t even agree where to meet, let alone what time the meeting starts. So I think crypto is a bit lost in the woods here.
Do you think Bitcoin is cooked?
I just think it’s capped, and eventually it gets hacked and goes to zero.
Do you own Bitcoin? Are you willing to say?
Through a company called Digital Currency Group. I had this very long-term investment in Bitcoin. I seeded it to DCG a long time ago. They started up a bunch of ETFs. They gave me stock in return. That was much better for me because I alleviated myself of the storage risk and the security risk, and I had equity. That’s worked out quite well. I think DCG is great.
That’s my mechanism to have exposure to crypto. I don’t own crypto directly anymore.
Got it. Alex, over to you.
I’ll just note that I’m not very bullish on crypto. Obviously, crypto is cooked. Not investment advice. Anthropic’s ECDSA announcement demonstrates that frontier models can probably achieve astonishing inversion attacks against crypto systems.
Question for Chamath: SPACs. American research universities, I would argue, are sitting on an enormous amount of intellectual capital that they’re scared, arguably, of translating into commercial practice because they don’t want to be treated like for-profit venture capital funds. They want the tax-light treatment of nonprofit academic institutions. What would you think?
I’ll float this as an idea first, because this is a favorite hobby horse of mine: reviving SPACs for acquiring academic research universities in America, taking them for-profit with government buy-in and IPOing them. You, as sort of the godfather of the SPAC movement, could help do this.
Oh, I love that idea. That’s a fabulous idea. Heterodoxical. That’s a great idea.
Maybe you and I should do that together. I would call them SPARCs.
Which university would you target first?
I want to IPO Harvard University. I want to IPO Harvard on the New York Stock Exchange. I did this back-of-the-envelope analysis: If you look at the officially stated Harvard endowment versus how valuable it would be if you could dissect the venture capital arm, the real estate holdings, the education, the endowment, and the merchandise, it would be four times more valuable if it could be an honest-to-goodness for-profit, maybe public-benefit-corporation conglomerate, rather than being stuck trying to hide from for-profit taxation the way it is right now.
I think it’s a great idea. I mean, I think it’s almost too disruptive to be viable, but it’s a great idea.
OpenAI has, I figure, demonstrated that it’s possible, with appropriate government appeasement, to take a valuable not-for-profit and turn it for-profit. So why not turn that loose SPAC-style on the economy?
Yeah, I like the idea, but I think you’d have to be on the inside to do it, which is hard.
Well, maybe Chamath, you and I should explore this.
You need the leaders of Harvard Corporation to be in tune with it.
Sure. It might require starting with something smaller than Harvard Corporation, but I’m glad to hear your endorsement of this idea. We should talk about it.
Yeah, I think it’s a really interesting idea.
Amazing. Sal, over to you, pal.
Okay, Chamath, I’ve got a tension I want to set up between you and Alex. You’ve said that AI could compress S&P valuations from 22x free cash flow to, like, 2–7x. If you believe that, then why should anybody own the S&P 500? Alex is famous for saying to own the indexes and own the market because it’s hard to beat the market. So, how do you resolve that?
I think the index is not what it used to be. The index used to be the 500 best companies, and it was owned and structured in a way that gave you implied diversification, which it doesn’t today. I think there’s an enormous amount of revenue concentration in a handful of names, and you essentially have this distribution of haves, which are fewer than 10, and have-nots, which are the other 490. The index isn’t what it used to be.
So, I think you have to be much more precise in owning other things that are thematically relevant and important in these cycles. More generally, I think that intellectual property is mostly not valuable anymore. And insofar as IP is not valuable, I think people’s typical underwriting of defensive moats is basically gone.
What’s left over is cash flow. I think the only thing that matters right now is cash flow, only as a bridge to whatever unit of value matters in the future. The more cash you have today, the more you can translate it into whatever people will use in the future. You don’t pay for future cash flow because you fundamentally can’t underwrite past a certain number of years.
I’m a much bigger fan of taking the S&P 500, taking the Global 2000, ranking them by free cash flow yield, and just buying the top N. I think you’re probably better off over the next number of years.
All right, Dave, over to you, pal.
A selfish question, because I’m really curious. You were the first investor in Groq, and then Nvidia bought it for $20 billion. Another brilliant Chamath investment. But now you say you won’t invest in chips. Just for the audience, Groq was a dedicated AI transformer-accelerator chip, and now it’s part of Nvidia. So now you’re not doing chips, but you are doing power. Are you doing data center components, boxes? Why not chips, and how did you make that choice?
13. Chamath’s Investment Strategy and the Future of AI
You have to be able to go to markets where you look dumb. When I basically found Groq—or found Jonathan—and we started to work together, the company didn’t exist when I wrote the first check. It was literally a ground-up startup: him, myself, and the third co-founder. It sounded really stupid to my partners and to the industry, and they were all wrong and I was right.
Now it’s conventional wisdom that you should have a silicon startup. That’s why it’s completely not going to work. They’re right that there’s going to be a bunch of domain-specific silicon. They’re right that there are going to be highly customized solutions that work in very specific operating environments. All of that is right. But you will not create enterprise value.
The reason you won’t be able to create enterprise value is that when you try to tape out a chip and find production capacity, all the memory is blocked, all the fab space is taken up, and all the packaging capacity is absorbed. Now you’re the nth person in line, and the nth person in line is effectively DOA.
I think you’ll have some really valuable IP, but again, that’s going to go to zero because some other group will be able to direct an AI model against another AI model and say, “Design me the domain-specific architecture I would use to optimize this for inference or for some other metric.” You’re going to have all of these permutations. The surface area of silicon options will explode, but the ability to make equity returns will go to zero because that’s now too obvious.
If you look down in the stack, where people don’t go because they think it’s too stupid or not worth it, it’s data centers. People think it’s too blue-collar, and a lot of intellectuals will talk themselves out of lots of money. But that is where you should play. The much more speculative area is applications, and it’s still to be determined how to make money in the application stack.
Those are the 3 areas, and you kind of have to be able to shuck and jive, ebb and flow. It was obvious to me that the way to penetrate AI in 2015 was in the silicon layer. But it’s now totally saturated, and no money will be made. Now I have to go down or go up, so I chose to do both. I could be wrong, but that’s sort of how I think about it.
Great advice. I’m going to sneak in one final question on behalf of our viewers. If you’re willing to, what’s your investment advice during this hypersonic tsunami over the next 3 years? Where do people place capital?
My gosh, I think that’s a very complicated question. I don’t know. My honest answer is I don’t know. I’m the least invested I’ve ever been in the last 15 years.
Are you invested?
The least invested.
I just don’t know. To Sal’s point, I’m in this schizophrenia about pivoting toward cash flow. To Dave’s question, I’ve gone up and down the stack. To Alex’s question, I’m still trying to figure out whether there’s a better way to create liquidity in the public market, so different kinds of assets with different payoff profiles could be available in the public markets when ASI actually lands.
I’m honestly the most confused I’ve ever been. Then I compare that to the certainty so many other people have in the market right now, and it just seems like late-cycle, late-stage behavior. I’m just not smart enough to know, and I’d rather bide my time and think through it all.
If I had to pick, I’d say I like owning cash flow because I think it’ll give me the oxygen to pivot to what will matter.
Mhm.
I like owning power because I think it’s the constraint, the fundamental constraint of everything for the next 10 or 15 years. And I like alternative mechanisms of deploying capital because I think the traditional financial infrastructure has become so sclerotic, and it will become even more closed if there’s a retrenchment in the AI trade.
Yeah, it’s tough. We always close out with a fan video, an outro song. This one is called Everything Is Cooked by Douglas Lance. And just for a little bit of history, Alex has an incredible voice. Alex, what were you performing in when you were younger?
Yeah. My first career, which is not widely known, was as a boy soprano performing with the New York City Opera Company. I performed on stage in Carmen and La Bohème, and I did a tour with Turandot. Then my voice changed, and they kicked me out of the New York City Opera Company. You don’t get to be a boy soprano for that many years.
All right. Here’s another playback of your younger self.
This is AI-generated, by the way.
On the menu right now, as I like to say: math and science, engineering, physics and biology are cooked.
Oh, this is awesome.
Medicine, computer science, all the social sciences are cooked. What’s on the menu right now? The whole economy is cooked.
Oh, they didn’t extend it.
I think Doug said that he’s going to produce a full Broadway musical based on it. He didn’t ask for my permission, by the way, but I’ll be curious to see the result. Like a whole musical, like Quadrophenia or…
I think a whole song, maybe a whole musical.
Chamath, thank you so much for joining us.
I really appreciate being included. Thanks. I’m sorry it took so long to get scheduled, but thanks, Peter.
Hope you had fun.
I did. It was great.