The U.S. Votes on Data Centers, Big Tech Impact, Google’s AI Response | Sharp Tech with Ben Thompson
- The data center backlash has reached the industry’s presumed safety valve, Texas. Sharp reports that 43 of 44 local measures opposing construction failed, while recalls warned officials that approving projects could cost them their jobs. Texas had been expected to house 25% of US capacity by 2030—the answer to opposition elsewhere was “we’ll just build it in Texas”—but its incoming governor campaigned on a moratorium. Existing construction cushions 2027–2028; the greater concern is 2029 onward.
- Tech’s political problem is distrust of the future it is selling, which fact checking alone cannot resolve. Thompson argues that “what you’re actually dealing with is sentiment,” reinforced by messaging that AI will replace workers and concentrate wealth. Separately, Sharp urges companies to compensate communities—“If you want to get a data center built, write checks. Put money in people’s pockets”—but asks whether that can overcome philosophical opposition.
- Restrictions could entrench the hyperscalers that opponents want to punish. Existing sites, power connections, cash generation, cheaper credit, and stronger equity currencies give incumbents advantages over frontier labs seeking new infrastructure. Thompson’s illustration: $200 billion of Amazon capex against $180 billion of cash generation still leaves an enormous funding advantage over loss-making labs. OpenAI and Anthropic might endure while ownership changes: “The question is who are the equity holders?”
- An externally imposed capex slowdown could give revenue time to catch up with investment. Thompson’s conditional upside is that easing capital pressures could let “everyone” win. Nvidia could sell fewer chips in absolute terms while strengthening its relative position: constrained power makes “the most tokens per watt” worth a premium. Existing neocloud capacity could also appreciate even as expansion becomes harder.
- Google’s search disappointment could represent a monetization transition rather than a terminal business failure. Alphabet’s reported decline of more than 13%, erasing over $500 billion, followed disappointing search and advertising results. Thompson sees a possible replay of Meta’s Stories and Reels transitions: attention moves before advertising tools and monetization catch up. Sharp’s objection is substantial—Google must replace “the greatest business that has ever existed”—and Thompson acknowledges his poor forecasting record on Google.
- Google may be using search to finance an infrastructure transformation. Sharp compares search’s potential role with See’s Candies funding Berkshire Hathaway’s railroads: lower margins can accompany greater absolute profits. Thompson’s “Google Capital Company” idea is that Google could become an AI infrastructure provider while transforming search into an interface and eventual agent. Whether that ultimately makes financial sense remains uncertain.
- Reddit’s roughly 30% selloff illustrates the cost of losing Google’s distribution bargain. Sharp regards management’s direct-destination argument as optimistic spin; Thompson calls it a “valiant attempt to state reality.” The response is higher revenue per user through logged-in experiences and better advertising, alongside licensing continually refreshed content. The familiar 1% posting, 9% responding, 90% lurking split matters: contributors already log in.
- AI’s more credible entrepreneurial payoff may be many $100,000 businesses rather than a few billion-dollar solo companies. A school-lunch app illustrates software tailored to a small community rather than a mass market. Thompson advises using AI as “a machine gun you can train on your weaknesses,” freeing time for strengths. Payments, authentication, hosting, and security remain unresolved opportunities around this “group chat” era of software.
1. Texas’s role as the automatic escape route is now uncertain
Sharp reports that 43 of 44 local measures opposing data center construction failed; North Dakota was the exception. Thompson considers recalls of mayors and commissioners particularly consequential: officials now know approvals can cost them their jobs, including decisions that never reach a public ballot. In the Madison market, candidates from both parties attacked each other over alleged data center support.
Texas is the larger shock. Gina Hinojosa was elected governor after promising a moratorium on new data center construction until the legislature could act; Thompson notes that the legislature meets only once every two years. The episode cites a projection that Texas would hold 25% of US data center capacity by 2030, with expectations of a still larger share over time.
Thompson recalls a senior industry participant dismissing a canceled Wisconsin project because everything could move to Texas. Abundant natural gas, including gas otherwise flared, had already supported Bitcoin mining; Texas also operates its own grid alongside the eastern and western grids. Land, energy, and a presumed friendly government made “we’ll just build it in Texas” the backstop. That political assumption is now unsettled.
2. The backlash targets AI’s promised future
Thompson’s framing: misinformation often supplies ammunition for an existing preference rather than creating it. “You’re trying to refute facts when what you’re actually dealing with is sentiment.” Debunk one water or electricity claim and another objection can replace it, because the underlying force is distrust of technology and discomfort with AI. Thompson sees people voting against “this version of the future.”
Sharp offers the slogan, “If you love advertising, you’re optimistic about AI.” Thompson explains the connection: good advertising identifies a problem people have learned to tolerate and shows a solution. Consumer AI companies have failed to make that connection convincingly. Sharp’s pushback is that they are building exceptionally expensive technology while their clearest public application remains replacing large portions of the workforce.
The deeper mistake, Thompson says, is treating people as interchangeable inputs ranked on an IQ scale. Humans are “need generation, desire generation machines”: they invent wants, businesses arise to satisfy them, and demand changes. Job-replacement messaging implicitly tells the public, “All of you are worthless.” Sharp finds the backlash understandable given that message and the concentration of wealth among big tech companies.
Sharp leaves durability open: movements with a memetic character can fade over two or three years, although those years matter enormously for construction. Thompson thinks broader anti-tech sentiment will persist and data centers will remain on ballots. The globalization comparison sharpens his point: people could not vote when factories closed; here, they have a veto over a future they dislike.
3. Communities need benefits before the virtuous cycle arrives
Thompson calls corporate America’s underlying error “the reduction of morality to legality”: exploiting every permissible loophole invites regulations that constrain innovation alongside abuse. Amazon’s headquarters auction exemplified the posture—one of the world’s most powerful companies soliciting subsidies from competing cities. Data center companies should have anticipated public unease and demonstrated community benefits before demanding inducements as the price of locating there.
Loudoun County, Virginia, supplies the affirmative example. Sharp identifies it as the country’s richest county; Thompson says its property taxes have gone down. Sharp adds, with an explicit hedge, “I believe power prices have gone down.” Sharp traces the county’s growth to historic connectivity, proximity to Washington and the populous East Coast, access toward Europe, and available power. Thompson adds that the resulting industry developed a specialized construction ecosystem.
Thompson emphasizes that Loudoun represents a roughly 30-year virtuous cycle. Sharp cautions, “You can’t sell the long run” while voters bear immediate costs; Thompson adds that communities should be paid for the privilege of running up their electricity bills. Sharp distinguishes election incentives from governing incentives: officials also risk watching valuable industrial projects go elsewhere.
Sharp predicts communities will reverse the negotiation: “You want a transactional model, you’re going to get a transactional model good and hard.” His practical prescription is direct compensation—“If you want to get a data center built, write checks. Put money in people’s pockets.” He agrees companies can accept that baseline but doubts it guarantees approval where residents oppose AI philosophically or resent tech’s wider effects on society.
4. Existing projects buy time, while new investment may move abroad
The hosts distinguish new permitting from projects already underway, citing roughly 130 Texas projects under construction. Thompson expects much of that capacity still to arrive in 2027 and 2028; the concern shifts toward 2029 onward. That limits the immediate construction impact without removing the possibility that today’s political decisions create a later bottleneck.
Going abroad is “probably likely to happen,” Thompson says; the hosts cite a $15 billion Google project in Malaysia. Canada and Mexico enter the discussion, but permission, power, financing, and security still matter. Thompson had expected substantial Middle Eastern construction. He recalls an AWS region in Bahrain, he thinks, whose customers were told their data was gone forever—an example raised amid concerns about security.
The possible domestic dividend is power generation that survives an AI investment bust. Sharp says the US has been living in “power poverty” for roughly 50 years, dating back to the oil crisis, and contrasts today’s efficiency mindset with an earlier assumption of abundant energy. If economically extravagant AI spending produces enduring generation capacity, its long-run value could be enormous. That payoff remains conditional, but Sharp believes substantial power projects are already moving toward completion.
5. Space infrastructure becomes less absurd when launch capacity already exists
Thompson rejects the image of warehouses floating in orbit. His proposed unit is an individual rack: an NVL72 is approximately the size of a current Starlink satellite, he says. Cooling remains difficult, requiring massive radiators, but satellite deployment and laser communications already exist. Agent-serving inference could also tolerate more latency than a person waiting for an immediate chatbot response.
The economic starting point matters. SpaceX is already developing Starship and expanding launch capacity; Sharp argues that this creates a need for useful payloads. Sharp characterizes Starlink and SpaceX as “the ultimate hammer looking for a nail company.” He argues that launch can be treated as a marginal cost for some decisions, while explicitly acknowledging that it is not actually free. Starlink’s apparent profitability also depends partly on internal launch pricing rather than an arm’s-length premium.
Thompson’s shipping analogy separates operating economics from sunk investment: a container costing $1,000 at the margin can rationally ship for $1,100 even if depreciation makes its accounting cost $2,000. Reported losses can coexist with positive cash flow; debt service can ultimately cause bankruptcy. COVID’s $17,000 containers illustrate why commodity operators endure lean years—and why investing only after prices recover can be too late.
Sharp argues that launches build on the learning function; Thompson adds that Musk’s overarching goal is Mars. Verizon, AT&T, and T-Mobile should worry about a rival whose seemingly irrational cellular economics support a different objective, Thompson argues. Sharp frames orbital computing as a “5 to 10 to 15 year solution.” Thompson emphasizes that most data centers will still be built on Earth for the foreseeable future.
6. Hyperscalers hold the strongest cards in a constrained buildout
Frontier labs have two potential integration paths, Thompson explains: combine models with consumer-facing applications, or combine models with purpose-built chips and infrastructure. The latter requires capital and new facilities. Anthropic’s Broadcom chip agreement does not remove the need to construct places to deploy those chips. Existing hyperscalers begin with assets and relationships that the labs must assemble.
Legacy facilities are not easily converted into modern AI facilities: older designs generally use air cooling and lower density, while current GPUs require much denser, water-cooled installations. Conversion can amount to rebuilding. Nevertheless, incumbents already control land, power connections, local relationships, and sites communities accept. Sharp cites Meta discussing expansion of existing facilities as an alternative to entirely new construction.
Thompson’s financing illustration is Amazon spending roughly $200 billion on capex against $180 billion of underlying cash generation. Exceeding that cash generation does not erase the advantage of having it. Frontier labs start from negative cash flow. Successful existing businesses also support cheaper borrowing, allowing Amazon to obtain attractive Anthropic arrangements partly because it can finance infrastructure more cheaply than its customer.
Equity is the final funding source, and incumbents have an advantage there too. Buying newly issued Amazon shares purchases exposure to a successful existing business alongside speculative AI investment; buying Anthropic equity depends on Anthropic succeeding. Hyperscalers also possess greater lobbying capacity and more resources to compensate communities. If those become decisive construction inputs, opposition reinforces incumbent advantages rather than weakening them.
7. A compulsory capex pause could relieve the financing race
Nobody can easily slow down voluntarily. Amazon risks ceding capacity to Google; Microsoft’s brief restraint already produced shortages, Thompson argues. Frontier labs face the same competition in training: exceptionally cheap inference does not eliminate the need to spend heavily to remain at the frontier. Anthropic cannot independently decide that OpenAI, Google, or another competitor will stop advancing.
An external constraint could let everyone pause without surrendering competitive position. Canceled projects could mean flatter capex, relieved shareholders, higher share prices, and a more valuable equity currency for subsequent fundraising. “If the revenue comes,” Thompson stresses, those financing pressures can unwind positively. A slowdown that gives revenue time to arrive could therefore let “everyone” win.
Sharp asks whether compute-starved labs eventually approach their suppliers hat in hand. Thompson does not expect OpenAI or Anthropic to disappear: “The question is who are the equity holders?” Google and Amazon owning more of Anthropic is one possible outcome. He also warns against treating contracted demand as unconditional protection: a guaranteed buyer that is itself OpenAI can transmit financing trouble through the chain.
8. Scarcity changes Nvidia’s economics and complicates the China comparison
Nvidia faces a straightforward short-term problem: fewer places to install chips can mean fewer sales. Relatively, however, power and facility constraints favor the supplier delivering “the most tokens per watt.” Paying its premium becomes easier to justify when scarce infrastructure must produce the greatest possible output. Thompson distinguishes being “better absolutely or better relatively”; a prolonged scarcity regime can hurt volume while strengthening competitive positioning.
Existing neocloud facilities present the same tension. Slower additions make installed capacity more valuable, but restrictions also make expansion harder and sufficient scale less attainable. Analysis that assumes intelligence is already a commodity skips a necessary condition: effectively unlimited compute. Thompson thinks the scarcity period has been extended, making conclusions drawn from an abundant-compute end state less useful for current decisions.
China is a conditional longer-run concern. America has a “huge chip advantage,” Thompson says, but China has been pushed to solve chips domestically; if it succeeds, greater power and construction capacity could eventually produce a manufacturing-like scale advantage. Sharp’s disagreement matters: despite acknowledging those possibilities, he has “never been less concerned” about US-China competition and is “not that panicked” by claims voters handed China the future.
9. Google’s search miss resembles Meta’s monetization transitions
Sharp cites Bloomberg’s report that Alphabet fell more than 13%, losing over $500 billion in its largest market plunge. Overall sales missed projections, as did search and related advertising, a business representing roughly half of revenue and financing AI investment. Search advertising growth slowed to its weakest pace in five quarters. Thompson’s response is explicitly “a very optimistic read” of the earnings call.
Stories and then Reels moved users away from Meta’s already monetized surfaces. Advertising tools, measurement, and advertiser habits lagged behind attention, creating bargains for early advertisers. Meta eventually built the systems to monetize the new inventory. Thompson’s rule: “You want to buy a digital advertising company when its inventory is exploding.” Reels produced the spectacular “blip” associated with Facebook shares around $83.
Google’s language sounds familiar: get the product right, distribute it widely, then monetize. AI Mode can increase queries and engagement while diverting users from the search-results pages where existing ads appear. More time on Google therefore does not automatically mean more revenue. The bullish interpretation is that product adoption has advanced faster than the advertising format, rather than that demand has disappeared.
Sharp’s pushback: search was “the greatest business that has ever existed,” a much harder benchmark than merely finding profitable chatbot ads. Thompson grants the question but sees “a decent chance” that monetization makes this another temporary blip. Sharp notes Thompson’s strong Meta record and weaker Google record; Thompson jokes, “You should always do the opposite of what I say about Google.”
10. Google may be converting search cash into a different company
Thompson’s “Google Capital Company” framework treats search as funding for a future infrastructure business: data centers, proprietary chips, cloud, and stakes in companies including SpaceX and Anthropic. Sharp compares the idea with See’s Candies financing Berkshire Hathaway’s move into railroads. Railroads have worse percentage margins but can generate much larger absolute profits. Search could prove to have been “just the fuel to get the real business.”
Google is also making a separate bet on transforming search itself. “Who’s done more to obliterate the search experience than Google itself?” Sharp asks. Rather than protecting the old interface, Google keeps moving users into chatbot interactions. Thompson’s prediction is straightforward: “Search is going to be your agent,” eventually doing tasks such as booking things rather than merely returning links.
Thompson admires the execution and willingness to choose, despite uncertainty over ultimate financial returns. Sharp says Google’s cloud numbers are incredible and its profit margins are skyrocketing; Thompson adds that infrastructure barriers to entry are enormous. Sharp notes Thomas Kurian’s appearance on the earnings call and reads Google’s choices as a sign of priority. Thompson says that without the “Google Capital Company” trajectory, he would be significantly more worried about Google.
The erosion is visible in personal usage: Sharp now uses ChatGPT for 80–90% of tasks previously handled by Google, and cites OpenAI saying search-like ChatGPT queries equal one-fifth of Google’s volume. They consider whether Google has been extracting extra revenue from a declining franchise. Thompson calls that plausible but explicitly does not know how accurate the cited critique of its tactics is.
11. Reddit must replace referral volume with value per user
Sharp reports Reddit losing roughly 30% of its share value over about a week. Steve Huffman’s explanation is that people still search for Reddit, but Google sends them there less often: “Google is teaching people that if you want to come to Reddit, you need to come to Reddit.” Sharp calls the direct-destination argument valiant spin; Thompson calls it a “valiant attempt to state reality.”
The wider bargain is breaking down. Google indexed publishers to improve search, sent them traffic, and provided AdSense to monetize it. Thompson distinguishes possible abusive practices from the fundamental business problem: “Publishers’ problems are not Google’s ads.” Moving users into AI Mode removes distribution that supported the broader web economy. The end is of the internet “as we know it,” organized around one large market and scale.
Reddit consequently needs to “maximize your revenue per user” rather than assume unlimited low-value referral traffic. Thompson does not predict a paid model; he expects better ad monetization through logged-in users, customer understanding, and targeting. This requires changing the measures of success as well as the product. The business is being reoriented by external reality, whether management welcomes that process or not.
Its durable asset remains people producing fresh material for free, which also supports valuable LLM data licensing. Thompson invokes the 1% who post, 9% who respond, and 90% who lurk: the contributors already log in. Losing some anonymous visitors need not destroy the content engine. Requiring accounts might even encourage participation, although he acknowledges that the transition will be painful.
12. Small software gains usefulness without becoming a mass-market business
A listener’s school-lunch app captures the “group chat era” of software: public information is hard to find, while an app with notifications fits parents’ existing habits. A school district’s own app might accumulate unrelated functions. The opportunity is a simple tool tuned to a particular community—not necessarily software that millions will buy when millions of others can build similar things.
The listener asks who supplies payments, authentication, hosting, and security: platform owners, AI labs, Vercel and Supabase, or new businesses. Thompson offers no settled winner. Sharp highlights Google Play’s recently introduced friends-and-family app distribution and contrasts it with TestFlight’s developer-account and enrollment friction. Sharp sees Google ahead on this particular issue; Thompson agrees that this is what’s needed. Current improvised hosting and security arrangements expose substantial unsolved opportunities.
Small scale also restores motives beyond profit, Thompson argues. Making a useful neighborhood app can earn gratitude and status. Duplicating similar apps across many communities looks inefficient, but customization can improve welfare. His analogy is fashion: people are “very inefficient about wearing clothes,” yet their changing desires support an industry. Human-generated differences can create value rather than merely represent costs to eliminate.
13. AI’s practical upside starts with weaknesses and modest businesses
Thompson’s advice to MBA students is deliberately sharp: “If you see AI as an opportunity to do your job, you’re in trouble.” Instead, protect the strengths people pay for and use AI on secondary work. “AI is like a machine gun you can train on your weaknesses.” The purpose is to redirect time and attention toward what someone does particularly well.
Thompson is unsure about predictions of billion-dollar one-person companies, while allowing they may happen. What he “definitely” believes is that many more people can run $100,000 businesses. Sharp raises the example of a focused local-news publication; Thompson says its cost structure could fit its audience, unlike a newspaper’s inherited overhead. Removing costs lets modest revenue support a livelihood and makes more experiments economically possible.
Platforms may follow demonstrated demand rather than precede it. People could first assemble their own small businesses and software stacks; a Substack-like provider could then make the process accessible to many more. Sharp says he had considered building a Substack but doubted how many writers could earn money. Thompson’s response is that people underestimated the opportunity: “Turns out a lot.” His optimism rests on people’s underestimated ability to create markets when the required revenue falls.
Full transcript
Hello and welcome back to another episode of Sharp Tech, presented to you by Stripe. I’m Andrew Sharp. Joining me in studio today, Ben Thompson.
Ben, how are you doing?
I’m doing okay. I do need to know what’s more annoying.
What?
I have a ranked-choice list.
Okay.
Okay. Number one, your—
It’s not over this week.
Here we go. That’s right. Anyway, ranked choice: terrible idea. But anyhow, number one—
For another day.
Flight delayed.
Uh—
Wait, we have to get the list first. Number two: no Wi-Fi on the flight.
Number three: this is take three because we’ve been having some technical challenges.
A good start. I’m going to go with number three. That is by far the most annoying.
I could see it. You came in all chipper and cheery after no Wi-Fi, no flight delay—
Good rhythm.
And now you’re getting madder and madder. I can see it.
Absolutely red as hell here on YouTube.
In any event, I’m ready for a great show now that we’re finally off and running. I’ve got two Diet Cokes by my side. We have a lot of ground to cover on this episode, so we’ll see what we get through.
We’ll begin with the elections this week. There were elections all over the country, and it seemed like everyone was voting against AI data center construction. Of 44 local ballot measures opposing data center construction, 43 failed.
I’m glad you got that right. You said everybody. You sounded like everyone on Twitter, as if every single one failed.
No, we strive for accuracy. North Dakota was pro-data center.
We’re going to have to pause for a fact check. Maybe even more important than that—
Actually, can I add on that, though? It wasn’t just that these measures against data centers failed.
In several different places, mayors or county commissioners got recalled.
Yes.
Which I think is arguably more important than these measures failing, because it’s a signal to everyone in office: your job is on the line.
That’s right. It’s not just that these measures passed.
Mm-hmm.
It’s that future measures might not go to the ballot, but just to the mayor’s office or the board.
Exactly. And look, there’s been a groundswell of opposition even since we last podcasted about it. I think we talked at length about data center opposition. We’ve been—
In April or May.
As I said, we talked about it previously, but I do think this is one where we were—
We were early on, and the acceleration has been unbelievable to watch. Maybe even more important than the local ballot measures, Gina Hinojosa was elected governor in Texas. She ran on a promise to impose a moratorium on new data center construction until the state legislature can act.
Right. This state legislature only meets once every two years.
Oh, I wasn’t aware of that.
Oh, you don’t know? This is a great thing about Texas.
Part of the bottleneck. Okay.
Yeah. One of the unique things about Texas—or maybe a better way to say it—is that its legislature meets once every two years.
Wow.
So this is more meaningful. And the other thing about Texas—you’re right to say that Texas is more important. Back in the spring, when we were talking about this, I went to a conference and was talking to someone very senior in the data center game. He had been heavily involved in the Wisconsin data center right up the road that we talked about in our previous podcast, which got canceled. Basically, his take was, “It’s fine. Everything’s just going to get built because Texas.”
Right. And it’s Texas. It’s not just the—
Is that just because of available land and power?
Power. There’s so much natural gas in Texas. They set it aflame. They’re pulling oil out, and there’s tons of natural gas coming out that’s there to be captured. There’s a reason there was a huge Bitcoin mining operation in Texas before: there’s all this excess energy, and it’s its own grid, which has pluses and minuses. In this case, it makes it easier to bring all this power online.
Okay.
You’re not dealing with the three grids in the U.S.—the East grid, the West grid, and the Texas grid. It’s its own thing.
I didn’t know that. Wow.
Oh, wow. This is a very educational episode. So many fun facts. But that’s why Texas is the one that really matters: the presumed backstop for all data center opposition has always been—
“We’ll just build it in Texas.”
Everybody’s going to Texas. They were on track to be home to 25% of U.S. data center capacity by 2030.
In the long run, if you talk to anyone, the assumption was it would be way more than that.
Yeah, it’s fine. New York’s going to pass a moratorium. Even Ohio, which is a significant portion of it, might end up there.
We have 50 different states that can compete for these projects, but infrastructure-wise, Texas obviously has all kinds of advantages and may not be replaceable.
Well, the assumption was that it was going to have a friendly government—
And so that would take care of it.
Well, that’s out the window for now. We’ll see what the state legislature says whenever they get around to meeting about this. What should the tech industry actually learn from the midterms? Can they change public opinion on AI and AI data centers? Is that a lost cause at this point?
The question is whether the lost cause is the tech industry learning. We should probably start there.
Fair enough. Looking at the market, it’s a wipeout as we record this on Thursday afternoon.
Thursday.
We had a little dead-cat bounce on Wednesday to throw everyone off the scent. Now, who knows? Is the American growth story just cooked in light of all these developments? Where’s your head at as we process all this?
Yeah, I think we were right before: you’re going to have to get a lot more transactional. The degree of arrogance around this data center thing has been pretty stunning, on multiple levels. Number one, just starting out by saying, “These people don’t get it. They’re dumb. There’s not a water issue,” and so on. That misses the point by getting stuck on the details of what people are saying instead of grasping the bigger picture. Why are people grabbing onto this? Go back to the whole fake news debate.
Mm-hmm.
Everyone got the whole fake news thing wrong for ages. They got cause and effect backwards. They thought, “People hear fake news, they believe it, and therefore they support the wrong thing.”
Yeah.
No, the fake news thing is confirmation. It’s feeding confirmation bias and giving people ammunition for what they already want to believe. You have to zoom out and understand why this is getting traction. There’s widespread resentment of—
That’s right. Elites—
Right. It’s so infantilizing and paternalistic. It just assumes people are stupid and that if we give them the right information, they’ll change their minds.
That’s right. But you understand why people come up with the idea that people are stupid, because some of the fake news stuff—
Well, some of the misinformation surrounding AI data centers is easily disproven or refers to the way data centers operate.
You’re operating on the wrong level.
Right. You’re trying to refute facts when what you’re actually dealing with is sentiment. That sentiment attracts stories, and people will say, “Okay, fine, that’s not true,” then find another reason to oppose the data centers. They oppose the data centers, and why? There’s a fundamental distrust and unease with technology generally, and certainly with AI specifically. Even though people were talking about the environmental costs and electricity, I think they’re just voting against this version of the future.
Yes.
And why not, given the version of the future that’s been sold? This is where you go back to globalization. People didn’t have a vote.
Right.
The factories just closed. In this case, once people had a vote, they weren’t too happy about it.
Absolutely. If we go back to the spring, I said then, and I still feel now, that it’s an understandable human response, given what AI portends, the way it’s been marketed, and the ongoing centralization of wealth among a handful of big tech companies. People just want a chance to stick their finger in the eye.
And why not? I can’t stand most of big tech in the way they talk about AI. We—
I’m very proud of this one. I might repeat it every podcast for the next several: if you love advertising, you’re optimistic about AI.
Okay?
Why do I like advertising? Good advertising helps the customer realize they have a problem in their life, and then it gives them the solution to it. Sometimes that’s very cynical: you’re creating a problem that doesn’t exist and then saying you’ll answer it. We’ve seen lots of political ads in the last few months, so we know exactly how that works. But sometimes it’s true: “Yes, I could actually use that thing. There’s a problem I didn’t know I had, and that’s going to help solve it.” This is very analogous to the AI question in 2 ways.
Okay.
Number 1, everyone trying to create AI for consumers fails to do this.
Mm-hmm.
The issue isn’t that you’re not showing people what problems they have in their lives and how AI solves them. That’s why the AI show-and-tell thing we do got tons of responses. I appreciate it. I saw you moved it up in the outline today so we could get some into the show.
There’s an aspect of just realizing—
Humans are so adaptable—
And so we actually adapt to—
Tons of problems in our lives. We just accept that’s the way it is, when actually it could be better. So, number 1, AI has terrible marketing when it comes to how it can actually make your life better, in part because it feels like no one in this industry has any sympathy or understanding for normal people. Why not? They’re all crazy rich, all around the coasts, with no conception of how normal people live.
I’ll also add that they’re building a technology that’s exceptionally expensive to build and doesn’t have a clear application beyond saying, “Well, this can just replace wide swaths of the workforce.”
No, exactly. They can’t articulate it, even though I absolutely think it exists. Number 2, where they fail is—and this is where I think there’s a relationship with not appreciating advertising as a medium—they see humans as utilitarian cogs in the wheel that can be replaced, because their desires and impulses are all knowable and understandable, and we can just design the perfect system. This is where you get the alignment with the Marxist, or communist, view of the world: “We can design the system because we know all the inputs, and humans are just reduced to an input in the system.”
And we can recreate the input with AI.
Actually, the adaptability point isn’t just that we can deal with discomfort; we can also generate new needs. Humans are need-generation, desire-generation machines. We come up with all sorts of crazy stuff we want, or want to do, and then people rise up to meet those needs. People figure out, “Some number of people have this need. I’m going to figure out how to serve it.” The capability I developed might be scalable, or other people might be able to do this. We just need to let other people know they have this need. That’s what good advertising is.
The whole industry—not the whole industry, but huge chunks of it—has such a narrow view of the world. They think humans fit in this column, and the marks on that column are IQ points. That’s how they rank the whole world. They’re like, “AI is climbing up the column and going to wipe you all out. Think how much money we’re going to save. How efficient!” The implicit message isn’t just that AI is coming for your jobs.
Mm-hmm.
The implication is that all of you are worthless. You’re all uninteresting to us. This is why I go back to all of us—I didn’t vote against data centers, but all Americans are raising their hands to the sky and flipping the bird to all these tech companies.
Absolutely. And why wouldn’t they?
1. Implications for Big Tech and Beyond
Yeah. That’s not to say they’re not going to get built. But I’m in tech, and there are aspects of this industry that are so disgusting to me and have driven you crazy for years.
The threshold question in all this—because I do want to get into the implications for a variety of players in this market—is how durable and widespread do you think the backlash will ultimately be? To me, that remains an open question.
It is an open question. I think some things come along and then go away.
There’s definitely been a memetic quality to the backlash, the opposition, and the misinformation around data centers. Movements that have grown with that memetic quality over the last 10 years or so generally tend to peter out over the course of 2 or 3 years, but these are a pretty important 2- or 3-year stretch.
Well, that’s the question: Is this a data center question, or is it an anti-tech manifestation?
Because I think—
It’s probably going to be pretty real for a while. I don’t think that’s going away. That’s why I concluded the piece by saying this is going to be on the ballot going forward. I do think this is going to be an ongoing thing, and it raises the question you mentioned: What does the industry learn? The sad fact of the matter is, maybe there was some fantastical past where corporate America did things out of a sense of what was right. Multiple stakeholders: our workers matter, our communities matter. We’re all in this together.
Mm-hmm.
One of my big critiques of America generally, and corporate America, is that this really started on Wall Street, even going back to the ’80s: the reduction of morality to legality.
Yeah.
If it is legal—
It is okay.
That’s right. The point I’ve made before is that not doing things because they’re the right thing to do forces more laws and regulations to be made. If you don’t naturally do the right thing, and you’re always finding loopholes and cutting corners, you have to regulate everything.
Absolutely.
Once you regulate everything, that severely limits your upside, because you’re forcing regulation to curtail downside outcomes.
But when you cut off the downside, you cut off room for innovation and ingenuity.
That’s why it’s so important. There’s another angle here. The tech companies could have approached data centers differently. What’s been the big strategy? Go back to the Amazon headquarters thing.
Mm-hmm.
They actually have a building there, and that’s all by you, isn’t it?
I think they built some of it, but not all of it. It was embarrassing for the tech industry, that whole episode.
This is the episode where Amazon basically auctions off its alternate headquarters to various markets around the country.
Yeah. All these cities were asking, “Who’s going to give us the sweetest deal?”
One of the most powerful companies in the world was basically saying, “We have a lot of economic impact and taxes. We want all kinds of subsidies.”
Sure.
It was shameless. That entire episode has characterized all of this stuff. Everyone just assumed, “Who wants to pay up for the data centers?” It was baked into the business models of all these companies.
Absolutely.
The right thing to do would have been not to be so shameless: to have some awareness of how normal people might think, in the full context of our messaging and people’s increasing discomfort with tech, social media, and all the things we’ve talked about. We needed to get ahead of this. We needed to make the case for the industry as a whole, and explain to local communities how these projects benefit them, instead of going to those communities and saying—
Right.
“We’re going to shake you down, or we’ll go somewhere else.” So that’s exactly what we want.
And what we’re going to end up with is: Fine, Big Tech. The communities will shake down the big tech companies. You want a transactional model, you’re going to get a transactional model good and hard, and you’re going to end up paying people off. That’s what’s going to happen. The question is, will that work? There are still going to be community members who oppose these projects on philosophical grounds, or because they resent tech generally and what it’s done to modern society. I think it’s interesting going forward, because you mentioned there were bipartisan campaign ads that you saw.
That was the crazy thing. There were so many data center ads, and both sides were trying to outflank one another and be more anti-data-center.
It wasn’t like you’d think, zooming out: “Okay, the Democrats are nailing the Republicans on this.” Republicans are generally pro-business.
Six months ago, that’s where we were. Then the Republicans were like, “Holy crap, people really don’t like data centers. We don’t either.”
There was one race in the Madison market—I don’t even know who it was—and every single ad went back-to-back, with them attacking each other for data center support. They’d manufacture reasons why the other candidate had supported a data center at some point in the past.
It was crazy. The whole time in the run-up—and this is some life advice—I hardly pay any attention to the election.
Mhm.
Once the election happens, then I’ll look at the results, think about what it means, and analyze it, because guess what? The election’s going to happen no matter what. You don’t have control of the election.
Maybe the polls are predictive, or maybe they’re not. Maybe the prediction markets are on it, maybe they’re not. I’m just going to say, we have some people in our lives—
—who are living in the crosstabs.
They spend a lot of time on it, and it has basically no value. But hey, maybe it’s entertaining for them.
Fair enough.
But the one thing that was in my head early on is, okay, this data center thing is unbelievable. I cannot believe how many ads there are. We talked about it before, and it’s clearly way more potent. Again, the fact that there was no affirmative message for data centers anywhere—everyone is sprinting away from them.
That was a losing message this cycle, and probably will be next cycle. By the time we get to 2030, I don’t know whether there will still be this massive outrage about data centers. What interests me is the difference between campaign incentives and governing incentives. Over the past couple of months, it wasn’t surprising to me to see both parties fall all over themselves to be more opposed to data centers than the opposition. Even Greg Abbott, who had been supportive of data centers, was cowed into opposing them in Texas. But—
Well, that is a big—I mentioned the globalization angle before. At the end of the day, call me a big tech shill: a data center is great for a community.
That’s the thing. These are basically the cleanest and most profitable industrial projects in American history.
Well, this is a good example. People would point to Loudoun County, including—I think I wrote this at some point, because it is true.
Loudoun County. You screwed this up the first time we talked about Loudoun County. Loudoun County, Virginia.
Loudoun County, Virginia. Richest county in the country.
Property taxes have gone down. I didn’t even know that was possible.
I believe power prices have gone down. Every part of an already rich county has gotten better because of data centers. Now, why are data centers there? This goes back historically to a major interconnect from when the internet was built. It might even have been pre-internet, but D.C. was close, and AOL started there. If you were built there, you were relatively close to Europe, and the most populous part of the U.S. was on the East Coast. It was the most natural place to start building data centers, and there was lots of power available. You got this virtuous cycle: more and more data centers, more and more power to serve them, more and more connectivity, and you end up with—
—and you have an area that knows how to build data centers because of the industry that has sprung up.
It’s an entire ecosystem. It’s like what you see in China around these manufacturing sites. Everything feeds on itself until there’s nowhere else that’s competitive. It’s still the largest place in the world. By the way, it’s phenomenal for everyone who lives closer to there. They don’t pollute that much. You have tons of power, your power prices are lower, and you have amazing city parks, services, and schools—
Right?
—and you don’t pay anything in property taxes. It’s unbelievable.
It’s not a bad deal. And so, relatively speaking—
If you’re governing one of these communities or states and watching these projects go somewhere else, there’s going to be an interesting tension, because jobs are on the line. Big tech tends to look at things at scale, at the end state.
Mhm.
But you have to get to the end state, Loudoun County.
That’s right.
The data centers started being built there 30 years ago.
So yes, you’re seeing the outcome of a 30-year virtuous cycle. In the meantime, if you build a data center—
—and you’re not power-neutral, and you’re plugging into the grid, guess what? Electricity prices are going to go up.
Now, electricity prices go up, there’s more demand, and they’re going to build new power generation. So, in the long run, it will pay off—
—but there needs to be more cognizance that you can’t sell the long run if the downsides are borne by people who can vote against your project.
Right? And again, not just mitigating, but demanding to be paid for the privilege of running up people’s electricity bills, right?
Correct. I would guess that the industry will be able to accept that as a baseline for any of these projects going forward. Even if they do that, I don’t know that it’s a guarantee the bottleneck will be removed in a place like Texas. We’ll have to wait and see. For the sake of argument, though, let’s assume the bottleneck remains in place over the next 2 or 3 years. Again, it’s a pretty critical 2 to 3 years as far as the buildout and compute demand are concerned.
Maybe. A lot of this stuff—even in Texas, the moratorium is on new permitting for new projects—there’s a lot under construction.
There are, I believe, 130 projects under construction.
Right, so 2027 and 2028, all that stuff is still coming online. I think the concern is more about 2029—
Going forward. So, a few things might happen. Number one, going abroad.
That’s unfortunately something that’s likely to happen. I think there was just a huge announcement of a new project in Malaysia by Google.
$15 billion for a data center in Malaysia.
Yeah. And so you’re going to get—tech is shameless. Remember, tech is the industry that started globalization, with Intel shipping chips to Hong Kong for testing and assembly. I’m sorry, you’re not going to get tech to do the right thing when it is—
—when it has no other choice. They’re going to—
Well, these are multinational corporations. That’s the way they operate. Sure.
And so, yes, for sure, going abroad is going to be one thing, and playing countries off each other. That should be a big concern. Obviously, number one. Number two, this was the whole thesis behind the SpaceX thing. I’m like—
“We’re going to outer space now. We have to.”
Well, I mean, you—
—or might have to. I don’t know. It’s in play. This was always a possibility, because even if you didn’t have this backlash—number one, when I wrote about SpaceX and its S-1, this backlash was already building. I built that into my analysis from the beginning. Number one, these are more viable than you think. It doesn’t mean they’re easy.
Can we get the CliffsNotes on the viability of space data centers?
The big thing is that people have in their head that you’re going to have big warehouses floating in space. Basically, what you’ll have is individual racks, which are like an NVIDIA NVL72. An NVL72 is about the size of a Starlink satellite today—the current ones, not even the big ones that are adding all the cellular capability.
So, a bunch of individual racks and redundancy built into the system.
Well, the challenge is cooling. They have to have radiators, and the amount of cooling is an exponential increase. They basically have these massive radiators going into space to dissipate heat. All of this is fairly insane and nuts, but you already have Starlink up there. You already have the laser-based communication for them to send data back and forth. In the long run, you’re going to have the shift to inference that happens independently of humans, and so—
So inference that’s serving agents, where time doesn’t matter?
Right, or matters less. You don’t need the immediate response. Basically, the pieces are there. It’s just, why would you do this if you can—
You have a lot of land in this country. It’s a fair question.
Well—
—or in the UAE or wherever.
I mean, you had this escalation in bottlenecks, and the bottlenecks get solved, right? We were short on chips, so we made a bunch more chips. Now we’re short on memory; that will be resolved eventually. Then we’re going to be short on power. By the way, I do think we’re far enough along in power projects that, if and when this bubble bursts—maybe it’s bursting right now as we’re sitting here, who knows—that was always the question: What’s the long-term payoff of the bubble? What gets built that’s economically nuts but actually worth it in the long run?
Mhm.
And we put in our power generation. We’ve been living in power poverty for 50 years, basically since the oil crisis.
We had a lot of power.
It’s funny: people go back and say, “Do you remember the ’50s and ’60s, when America was dynamic and growing?” We treated energy as completely abundant. It wasn’t a consideration for anything.
Just build all the factories. Build the beautiful, long Cadillacs that are 15 or 20 feet long, the boats. Absolutely. Build the crazy, overpowered engines.
Why? No one cared about efficiency.
Right. It’s funny: that sounds scandalous to anyone today because we’ve all grown up in a world of efficiency and, you know—
“Do more with less.”
Do more. Yes. It has been such a limiter on what’s possible, and I think it’s been a limiter on imagination.
What does it actually mean to have infinite power? This is why the nuclear thing is such a tragedy: we should have had infinite clean power at this point. But if the bubble bursts and we’ve actually brought tons and tons of new power to bear, that’s a long-term payoff that’s going to be hard to calculate—what it was worth.
And we’re on the way to bringing on some of that surplus power, correct?
For sure. There have been a ton of new power projects, and I think more of this stuff, to your point, will get over the line. This is the difference with globalization: the factory leaves, and there’s no upside. Again, there was no veto, but there was also no upside. In this case, there is a veto, and that’s what we’re dealing with right now. There is an upside.
If the demand is there, there are enough people who’ll be incentivized to get some of these projects across the line that I don’t really worry about it.
If you want to get a data center built, write checks. Put money in people’s pockets.
Exactly. If that’s what it takes, that’s what will happen.
On the other hand, the space example—I wanted you to give a CliffsNotes version of that, because when I first heard about it, I was envisioning a data center in space. That concept sounds completely insane and impossible to maintain at that level. But the more atomized version you lay out, where you’re talking about individual racks and the capacity to launch all those—
Yeah. Well, that’s the thing with Starlink. We talked about the Craig Moffett interview, where he asks, “Why does Elon want to get into the cellular game, which is a crappy business?”
Yeah.
And that’s correct if you’re starting from zero: “We’re going to build launch capacity that will let us deploy satellites so we can be a cell phone company.” That’s a crazy thing to attempt.
Sure.
The launch capacity exists. It’s coming regardless. There’s a lot about Starlink where you almost have to think uneconomically. If you think economically, you should think about the cost of all the inputs and figure out whether it makes sense. From Starlink’s perspective, it’s kind of like launch is free.
Yeah.
Because it’s not really free. That’s not correct.
For anyone else who wanted to do this, it would be insane to attempt that capacity.
They built out this capacity, and they need to use it for something. This is how you get, say, shipping companies that will run losses for years.
Mhm.
And it’s like, “How do you operate at a loss?” Well, it’s an accounting loss. They don’t ship if the marginal costs aren’t covered. What are the marginal costs? Fuel—
Port fees, crew, all of which pale in comparison to the depreciation cost of the ship itself.
Maybe their marginal cost to ship a container is $1,000. If someone will pay $1,100 to ship a container, they’ll ship it, even though the cost on their accounting balance sheet is actually $2,000.
Mhm.
But $1,000 of that cost is depreciation.
Which isn’t fake. That’s real money that went out the door, but that money’s been spent.
Yeah. So as long as you’re making anything incremental over your marginal costs, you’ll continue to operate, even though from an accounting perspective you’re operating at a loss. This is a super important principle to understand because it explains why companies will seem to operate at losses for ages. It’s like, “Why don’t you just shut down?” or “How are you surviving?” Well, their cash flow is still positive. What happens is they end up not having enough money to service their debt, and then they go bankrupt. That’s how bankruptcy happens. To buy these assets up front, you take out loans because you expect a 30-year useful life. Ideally, you’re paying it back over 30 years.
Yeah.
The way money moves around explains how bankruptcies happen and how companies operate at paper losses for ages. With shipping, you get a period like COVID, where containers are now $17,000 each. You’re making so much money. This is what’s crazy about commodity markets. It’s what’s happening in memory right now—the same general sort of concept.
And by then, it takes too long for anybody else to break in.
Oh, yeah. No, yeah. Buy low, right? People will think, “Wow, what an opportunity. Why don’t you buy a bunch of ships?” Then the market crashes, and it’s great for the incumbents because the new companies go bankrupt and suddenly cheap ships are available. Who wants to buy new ships?
It is the ultimate “sell the top, buy the low” sort of business. You need to be in the muck. Your business is bleeding money—that’s the time to invest, because then you’ll be ready—
When it comes back, right? If you wait until it comes back, it’s too late. In these markets, the fixed costs are so high. Buying a ship is really expensive; building a fab is so expensive. That’s why you see TSMC, which we’ve talked about, not wanting to expand too much because they’re the ones—
Holding the risk. That’s right. And so—
Where does this tie to SpaceX? It’s not a perfect analogy, and I’m not saying this is an accounting thing, but there’s almost a mentality to it. We’ve built launch capacity. Starship is coming. Starship can launch much larger satellites.
It can launch a ton of them. What’s this going to be used for?
We’ll find out.
Well, no, but suddenly you start thinking about data centers in space from a marginal-cost perspective. The marginal cost is building the space rack, deploying it, and operating it. You treat the launch capacity—not just the overall cost of developing Starship, but also the cost of launching—as a given. You want to launch more so you keep learning and improving. Musk’s companies are generally very good about their learning functions, so you keep going and treat that as free.
Mhm.
Now, it’s not free, to be clear. I’m not saying this is an accounting calculation. But if you’re Elon, you’re not thinking about building a cellular service and fully internalizing all the costs. You’re sort of saying, “This whole cost, which is actually the biggest cost, we’re going to pretend it doesn’t exist.”
Sure.
Wow, what a big opportunity we have, right? There’s a drive to use up this capacity. I think there’s a similar thing with data centers in space. Would you start from scratch, build a new rocket, and get launch capacity? Would you even consider the price of launch capacity? Starlink looks very profitable in part because they just pay the bare cost. There’s no premium charge to them, so SpaceX itself is actually much more profitable, and Starlink is much less profitable in a true arm’s-length calculation of the cost. All this is to say that Starlink and SpaceX are the ultimate hammer looking for a nail company.
Yeah.
They’ve developed this incredibly valuable hammer: launch capability, regular cadence, tons of capacity. They need stuff to put in space.
Well, that’s why I said we’ll see in terms of exactly how they monetize this. Data centers in space sound less crazy today than they did 3 days ago.
Well, I wrote that a few days ago.
Okay. Do you want to pat yourself on the back here?
No, but that’s why Verizon, AT&T, or T-Mobile should be nervous.
Yeah.
The most terrifying thing as a business is going up against a competitor that seems to be acting irrationally.
Mhm.
Right. It’s irrational for Starlink to build a cellular service. But in a weird sort of Elon-thinking way—which I was trying to explain here, though I’m not sure I fully explained it—it is rational if your overall goal is increasing launch capacity and building on the learning function. “I want to get to Mars.”
I was going to say, his overall goal is going to Mars at the end of all this.
That’s right. And so, yeah, I think the stuff in space makes sense in that regard. From a purely strategic asset perspective—
Mhm.
—you can see the value of that. Absolutely.
Yeah. Well, I have more questions on specifics.
That said, most of the data centers are going to be built on land, especially for the foreseeable future.
Yeah, that’s more of a 5-to-10-to-15-year solution, as opposed to the next couple of years, when the crunch is coming.
I don’t know. There are still going to be a lot of data centers on Earth.
For now.
2. Why Hyperscalers Benefit from a Buildout Bottleneck
As far as specific companies, it’s sort of an ironic twist based on what you wrote this week. Much of the opposition to AI data centers is rooted in deep, abiding resentment for big tech companies. Big tech companies may, in fact, be the beneficiaries of a lot of this opposition. Is that right?
I think so. Well, when you think about the—
They always win.
They do always win. It’s amazing. Where does value accrue, right? If you have the stack, I think it comes from integration—when you have different pieces you can put together, what can you integrate in the AI stack? On one hand, you could say the frontier labs right now want to be on the leading edge, but is that a sustainable business? What if you’re easily substituted out if another model comes along that’s better? That’s a tough place to be.
Yeah.
One argument is that they need to move into more consumer-facing applications. You’re integrating the consumer-facing service with the model itself, and that’s how you win. Another possibility is that the frontier companies want to have the leading-edge model, build their own data centers, and design their own chips for that model, so they can serve much more efficiently than anyone else. That’s another point of integration that could happen.
One of the challenges is that they need money to do this.
Yeah. Number one, they need to build new data centers. They don’t have an existing footprint. Anthropic is talking about it, and they have this deal with Broadcom to get chips. Those all need to be built.
Mhm.
Meanwhile, the big tech companies have a ton of data centers already. They still have the best access to capital.
Can I ask you one question before you go on? How many of those legacy data centers could be converted into AI data centers?
Not very many.
Okay.
The issue is that previous data centers were, by and large, air-cooled, whereas modern GPUs need water cooling. Modern GPUs are much denser, whereas the old data centers were more spread out because they used air cooling. You were trying to build them for the lowest cost possible. If you want to take an old data center and convert it, it’s significant—you’re basically rebuilding it.
Yeah. You just knock it down.
But you already have the land. You already have the power. You already have the relationships, and the community is used to you being there. I think you’re just starting further ahead than if you’re trying to get a new data center somewhere else.
Yeah. I read that a Meta spokesperson said this week that, in lieu of building new data centers, one of the things they’re going to do is expand the data centers they already have to meet some of this demand. All of that is much easier to do if you’re Meta or Amazon.
Right. They already have lots of assets. Number two, you have the financing ability. They have free cash flow—though that free cash flow is being consumed by the data center.
They had free cash flow.
No, but the point is, if you’re spending—Amazon’s capex was $200 billion, and you’re—
Wow, that exceeds their free cash flow of $180 billion.
That’s still $180 billion, right? We’re talking about billions of dollars ahead of Anthropic or OpenAI, which have negative free cash flow, at least in the fullness of time. So, number one, they just have this huge base. Number two—
They get better credit because they have other businesses that generate all that free cash flow. Their ability to borrow is better, and their cost of borrowing is much lower. That’s why you end up with these deals where they’re basically financing things. Amazon gets great deals with Anthropic because it can finance them. Even if Amazon ultimately has to finance it itself, it’s an arbitrage opportunity because my ability to finance is cheaper than yours.
Sure. This ties into Nvidia and the neoclouds and all that as well. Then, number three, they can issue equity. If you’re Anthropic, you’re going to need to get to market again. We’ll see if that actually happens, though I think it will. It might be a little hairy. All of Anthropic’s equity is 100% dependent on whether Anthropic succeeds or not.
Yeah.
When Amazon issues equity, some amount of that equity is about whether this capex will be worth it.
But there’s also an existing business that’s very successful, and you’re getting a piece of it. Of course, it’s going to be damaging to existing shareholders because you’re diluting them, and that money is going to this speculative thing that may or may not work out. But, by and large, the support for their equity is much greater than it will be for anyone else.
Yeah. That’s the funding of last resort, right? You start giving away parts of the company to get more money, which is where we are. Everyone is issuing equity at this point. So, at all those stages: more free cash flow, cheaper credit, and more ability to issue equity.
The hyperscalers are just in a better spot than anyone else.
Mhm, and they have more capacity to build data centers. They’re going to be better at lobbying and getting these things over the finish line. Those aspects might end up mattering most. If data centers are going to be more expensive because you have to pay off the local population, they’re in a better place to fund that. Big Tech will be, too.
I mean, we talked about it last week in terms of Anthropic and some of the pressures on them going forward, and some of the equity deals they did in the past in exchange for compute. What could those look like a couple of years down the line if Anthropic can’t pay for the compute all by itself? I wonder, in a compute-constrained environment where Amazon and Google have all this compute, do Anthropic and OpenAI just go hat in hand to some of these hyperscalers?
There’s a good chance. I think one of the scenarios is that OpenAI and Anthropic don’t go away.
Yeah.
The question is, who are the equity holders?
That’s right.
It’s like, does Google just own Anthropic at some point?
Yeah. Amazon. Yeah, it’s like a Google-Amazon joint venture. I—
Yeah, and it’s funny, because people were hating hard on Andy Jassy and Amazon after the earnings call a couple of weeks ago, and now they’re sitting pretty in a world where no one else can build data center capacity.
Well, they’re also sitting pretty in a world where everyone has to kind of slow down.
That’s the other bit. They’ll all tell you, “Don’t worry, we have buyers lined up,” et cetera. That’s always the case, until it isn’t. This is like all the crazy stock market things. In 2008, everyone was saying, “On paper, don’t worry, there’s a backstop. There’s a guarantee.” The problem is what happens when the backstop goes away and you get this chain reaction. You could still see this risk coming up, right? Like—
Yes, we have a guaranteed buyer. That buyer is OpenAI.
You go down the chain, all these sort of—
It’s so, so—yeah. But they had to compete, right? What if Amazon doesn’t invest?
Then Google’s going to invest more and more. Microsoft briefly stepped on the brakes, and then suddenly we didn’t have enough compute and it was a big problem. So everyone’s like, “I can’t slow down if they’re not slowing down.” We’ve talked about this dynamic with the labs themselves. This is the problem with Anthropic’s S-1. The fact that its inference looked so amazingly cheap should actually terrify you, because the part they have to keep spending a lot of money on—
That’s right.
—and it’s out of their control. If they don’t spend to stay on the leading edge, OpenAI will, and they’ll lose their position. Or Google, or Meta, or whoever it might be.
Take a breath amidst this murderous capex buildout here.
The shareholders love it because, “Oh, we had a bunch of projects canceled. What a bummer. Our capex is going to be flat.” You can imagine an earnings call like this, right? Everyone announces this, and investors are like, “Thank goodness.”
Yeah.
Suddenly their stock goes back up. Now they can issue more stock to get more money. That currency is more valuable. Maybe they’re—
The credit?
Mhm.
The big question is, when does the revenue come?
Once the revenue comes, all these things unwind themselves in a very positive direction. If there’s an external force that forces a slowdown, such that revenue can get there in time to relieve all these capital pressures, everyone could win. This could actually be a defusing win-win.
Win-win. Sure.
So, yeah, everyone thinking about AI jumps to the final state where intelligence is a commodity.
Mhm.
And that state might take a very long time. That implies infinite compute.
Yeah. And the longer it takes to get there, the more analysis predicated on intelligence being a commodity isn’t actually useful to understanding the world as it is.
In a world where there’s infinite power and infinite compute, what does Nvidia do? There are going to be other chip products, right?
In a world where you’re limited in power, you have to repurpose your data centers. Suddenly, whatever margin you have to pay to get the most efficient, most powerful chips makes a lot more sense. The analysis in a market defined by scarcity versus a market defined by abundance is fundamentally different.
Completely flips.
Right. And I think the time defined by scarcity has probably been elongated.
Okay. Well, it’ll be interesting to see whether there is ever a capex slowdown, because at this point, I’ll believe it when I see it. Whenever we get there, it’ll be fun to look back at some of the capex we’ve seen over the last couple of years and just how consistently insane it was. I think there was a statistic that a couple of these companies were spending more combined than the Pentagon spends every year.
And that’s just where we’ve been living for quite some time now. We’ll see what it turns into. Nvidia, you just mentioned them. How should we think about Nvidia in a world where U.S. data center construction is constrained over the next several years?
Well, in the short term, it’s definitely a problem because they’re selling chips and there’s nowhere to put them.
Okay.
Is their market being limited? On the other hand, this has always been a counterintuitive point about Nvidia: in a world defined by insufficient power—
—or insufficient data center capacity, the entity that delivers the most tokens per watt wins.
The best machine, the most efficient machine, will be worth the upcharge.
Exactly.
Because at the end of the day, the tokens are going to be more valuable. This is also a weird thing with the neoclouds. All the existing neoclouds out there are now more valuable if we reduce the rate that capacity comes online.
And at the same time, it’s kind of weird to think through what this means for the neoclouds. They’re going to have a harder time getting new projects started and coming to market. Are they ever going to have sufficient scale? But what they do have might be more valuable. Do you mean better absolutely or better relatively?
Okay.
Right, like, on an absolute basis, maybe this is worse for Nvidia; they’ll sell fewer chips. On a relative basis, as opposed to other chips, maybe this is better.
Well, a future where compute is abundant is probably not great for Nvidia.
It’s going to be challenging to sustain margins in that world.
3. China and the Global Data Center Race
And people inevitably come for their margins in that scenario, right? Well, you think about why Nvidia was so gung-ho about China. China is not a country that’s defined by a lack of power. If they don’t keep people locked in their ecosystem, using their software and CUDA and defaulting to their chips, over time it’s like, “Well, I’ll just build inefficient data centers, use more chips, and not pay the Nvidia fee.” I think they’ve resigned themselves, in the long run, to more coal power over there in China.
Well, that was going to be my next question. How should we think about the competition with China in a world where data center construction is constrained?
I mean, it’s a problem. We have a huge chip advantage. Obviously, we’ve created the conditions where China has to figure out the chip thing on their side. When and if they do, they’re going to have a lot more ability, I suspect, to build more data centers and more power than we will.
Okay.
In the fullness of time, how does this not look like the manufacturing world, where they have so much more scale and are so much cheaper that they end up owning all the compute? They will have far more compute than the American side in the fullness of time, even if they’re using less efficient chips that are inferior in a lot of respects.
Well, they’ll move down the learning curve. Their chips will get better and better over time.
Right. But again, this is a huge timing question. The fullness of time—everything’s possible in the fullness of time. We’re going to move to the dorm room, right? We should have a dorm room theme for when we want to have the dorm room.
Honestly, it’s tricky for me because, as the host of Sharp China, where I’m reading news about China every week, I personally have never been less concerned about the competition between the United States and China. China is always invoked as a stalking horse in these arguments. Everybody this week is like, “Well, voters just handed the future to China.” I don’t know that I disagree with that, because obviously China will have advantages it can bring to bear over time here.
And we’re going to build data centers.
I’m not that panicked about it.
So I think this is more—this is a—
Build another data center in the United States. Then we can have a conversation about it.
Right. Exactly. You know what, though? This is why I don’t like thinking about politics.
Okay.
I’m sick of this. I wrote an article. We’ve talked about it.
Fair enough. Well, final question, then, because we can move on to Google for you. Zach says, “When people discuss building data centers—”
By the way, maybe I’m the—maybe that’s evidence of my disdain for normal people. I’m the problem that I’m identifying.
Well, I mean, I don’t think there’s any problem with not wanting to talk about politics. That sounds fantastic. Zach, I’m so excited to move back to only caring about basketball for the next several weeks.
You said, “Move back to Washington, D.C.,” but you’re already—
It’s in my blood in D.C., you know. I grew up with it. When people discuss building data centers, Zach writes, “It seems to be either the U.S. or in space. Why don’t we just turn Mexico into a data center? Is it because it would be even harder to get the power, construction, and financing all set up in another country than it would be in space? We alluded to this earlier. Canada has a lot of power as well. Is there any reason we don’t go that direction?”
Well, I mean, they have to agree. It needs to be a place that’s safe and secure. So, yeah, there’s definitely going to be a globalization aspect to this, like I already mentioned with the Malaysia deal.
You’re going to see tech companies go out and say, “Who’s going to give us the best deal?” There are lots of data centers to be built. One thing that is interesting—and this does tie to the election, I guess somewhat indirectly—is that I sort of assumed a lot of data centers would be built in the Middle East.
Data centers don’t do so well with explosives.
A little hairier over the last—
No, there’s an AWS region in Bahrain, I think. They had to tell customers, “Your data is gone forever.”
Oh, wow. I didn’t see that.
Oh, yeah. You know, because—
There’s some stuff where you might have a law, or you put it in your terms: “I only want this data center.” If you had your data in Bahrain—
Bahrain?
Yeah, I don’t know. The data center’s gone.
That’s funny, because I almost interjected earlier to say there’s a lot of power in the Middle East. You can build all the data centers you need. Security would be a pretty urgent concern. When you’re dropping billions of dollars into these data centers, you’d prefer they don’t get blown up. So that will be a concern.
4. Were Google Bears Right All Along?
Google has gone to Malaysia. All right, well, speaking of Google, I’ll read from Bloomberg last week. Alphabet shares fell more than 13%, erasing more than $500 billion in its biggest-ever market plunge, after the company reported disappointing search and advertising sales that reinforced investor concerns about AI’s threat to its core business. The Google parent reported sales falling short of analyst projections. Ad sales from search and other channels—a business line that accounts for roughly half of overall revenue and is critical to financing the massive AI investments—also missed forecasts, growing at the slowest pace in five quarters.
You wrote about these earnings on Monday. You wrote that last week’s earnings call gave you Facebook deja vu. Can you unpack what you meant by that?
Well, this is a very optimistic read of Google’s call, but there was language they were using where I thought, “I’ve heard this before.”
I enjoyed your zag relative to every other analyst I saw reacting to Google’s call.
This actually goes back years for Facebook, but the 2 most prominent examples that I’ve written about were Stories and Reels. In both cases, you had a new format for social media that people loved.
Yeah.
You open Instagram, and there’s the Story from your friend. Your friend posted the whole thing. Your network’s already on there.
Snapchat, then Instagram.
So people clicked through and started going through Stories all the time. What were they not doing?
Watching ads.
They were not scrolling through the feed. Where were all the ads?
In the feed.
In the feed, right. If you talked to people in advertising—I think Eric Seufert has talked about this—people who ran Stories ads got unbelievable returns on their investment. But there weren’t tools built around it to measure it. Most companies are used to digital advertising, so if they can’t measure it, they spend where there are tools.
There was this crazy alpha for companies that marketed in Stories, because no one was buying Stories ads. They actually worked, because that’s where all the people were. There was an opportunity to make tons of money putting ads in Stories, where the ads were way underpriced relative to what they were worth.
Meta built an overall AI ad platform—whatever it’s called—where you put an ad in and it figures out where to put it, making sure you get a return. Part of that is to paper over this—
Advertisers choosing where to put their ads. Let it take care of it for you.
And they built measurement tools so you can see how many people are looking at your ad and how it’s converting, all that sort of stuff. Stories ads ended up hugely successful. That’s a reason to buy Facebook: you want to buy a digital advertising company when its inventory is exploding. That can cause prices to fall and sometimes cannibalize its existing product, but in the fullness of time, it’ll figure it out. Now it has that much more space to make that much more money.
Yeah. The Reels thing was crazy. It happened all over again. The exact same thing happened. Turns out people freaking love Reels, right? We’ve talked about this. What do you do if you’re watching Reels on Instagram?
Just scroll.
You’re not looking at Stories, and you’re not looking at the feed. So your existing surfaces are suddenly monetizing less, while you have this new surface. And by the way, what happened in Reels?
A huge arbitrage opportunity for the early advertisers who got in there. They made money before Facebook knew how to measure things, before everyone knew how to make the right ads, and all that sort of thing.
The broad dynamic is the way Meta talks about it: first, we figure out the right product. Then we scale it, get the user experience right, scale it to the world, and then monetize it. That’s the story they tell, and it’s broadly correct. That’s how you’d want your company to operate, right? Do you want them to shove ads into something that doesn’t work yet?
No. You want it to be popular and well-oiled by the time you start juicing the ads.
Right. So I’m reading the Google call last week, and they’re saying, “We want to get this product right first, then we’re scaling it to the whole world, and now we need to monetize it.” I thought, “That sounds like Facebook.” The question is whether Google can pull it off. Facebook has a proven track record of doing this. Google has just been shoving ads into search for 25 years. But there is a very viable story here, which is that AI search is better.
People are engaging more, which is a point they kept coming back to. They’re spending more time and making more queries. It’s kind of maddening, because if you click more, you’re now in AI Mode, in a chatbot mode, which is kind of annoying if you just want the Google experience. But you can see why it’s triggering people to use it more. The most-used chatbot in the world is probably actually AI Mode in Google, at least in the markets where it’s fully rolled out.
But the problem is, if they’re spending lots of time on Google, Google’s making more money, right?
No, because the ads are in the old product. They’re in the SERP, the search engine results page, not in AI Mode. The question is whether they can make ads that work in that mode, that resonate.
But right now it—
Because previously, Google Search was basically the best ad surface there’s ever been.
Well, partly because it was so unfair, right? You had to buy your own brand name to put it at the top of search results. Three or four links are all ads at this point.
But chatbots, generally speaking, haven’t shown ads anywhere near that successfully.
Yeah. I mean, ChatGPT is reportedly making a billion dollars or something.
If there’s a human there and attention is involved, someone’s going to figure it out.
I don’t dispute that people can make money. Unfortunately for Google—
They make it as well.
The standard is literally the greatest business that has ever existed.
Exactly. Those are all fair questions, but I actually think the Facebook story they told is totally fair and legitimate. They deserve more benefit of the doubt.
Okay. Like, once they have a mode broadly, let’s see what happens when they really start shoving ads in there.
And I think there’s a decent chance that it’s actually going to make a lot of money, and this will have been a blip.
Okay. In the long run, just like Stories ended up being a blip for Facebook, and the Reels thing was the greatest blip of all time when their stock was at $83 a share. So you mentioned Facebook: you’ve been right on Facebook over and over again, and now Meta, for the last 10 or 15 years, right?
Right, and wrong about Google.
So I’m wondering whether your benefit of the doubt is actually a bearish signal in terms of where Google is headed over the next couple years.
100%. You should always do the opposite of what I say about Google. So the question is, maybe this just doesn’t come back, but this is part two of the Google story.
I think I was right about—like, you’re on the couch, right? We should have a chair.
We should have a gag where you can press a button and it just starts hitting me.
I think Google is transforming as a company right in front of our eyes. And the reality is, where does search make sense in the future?
It’s fascinating to me because these were literally all the arguments we were having 3 years ago, and then everybody forgot about them. Google stock went to the moon, and now it’s like, okay, maybe the bears had a point in terms of the shape that all this would take. Maybe it was happening over the last couple years, and Google was sort of juicing some of the search revenue, but now it’s definitely happening.
Maybe they were juicing the search revenue because they knew it was on its way out.
And it ain’t what it used to be, Tony Soprano.
This goes back to the Google Capital Company concept I wrote about: the idea that, in the long run, we’re just going to be the dominant infrastructure provider for AI. We’ll own all the data centers on Earth, and SpaceX is up in the sky. And, by the way, we own a big chunk of SpaceX.
And we own a little bit of Anthropic as well.
More than a little bit.
Right. And at the end of the day, we’re going to have our own chips. We’re going to have the whole thing. That’s where we’re going. The search business, in the fullness of time, was just the fuel to get to the real business, doing the Berkshire Hathaway comparison of See’s Candies creating the money to get into railroads.
Mm-hmm.
A much worse business on a relative basis. The margins in candy are amazing; the margins in railroads are small. The absolute number is massive. There’s so much money. Your profit margins are small, but the total amount of money is much larger. Is Google thinking about AI broadly that way? Like, “Okay, search is toast in the long run. Let’s extract as much money out of this franchise as we can”?
Let’s do a two-pronged bet. Number one, we’re going to completely change what search is. I think it’s kind of amazing, and not what you would expect from a company being disrupted. The extent to which they’ve—
Who’s done more to obliterate the search experience than Google itself?
Sure.
Search is basically a chatbot. You kind of fall into the chatbot flow so obviously and simply; now you have to work to get out of it.
And I also think this is their agent story. Everyone’s like, “Where’s the Google agent?” They have Google Spark [?]. They have Gemini. They have something in the workplace thing. Google Search is going to be your agent. Search is going to go do stuff for you. You’re going to go to Google, and it’s going to book your thing for you. That’s obviously where it’s going to be. They’ve decided—
We’re going to take our core asset. We’re not going to protect the business model. We’re going to use it as—
We are disrupting ourselves.
Which companies don’t usually do. This is why I find myself admiring the gumption.
You admire the gumption.
I do. Okay. Everything about Google with AI has felt incompetent except for Search.
The part of the company that seems most threatened is somehow the most capable of delivering day after day, month after month. Do you know how much work has to go into developing models to do that at scale? How much money it takes to be doing AI Mode everywhere? And they’re just pell-mell, straight ahead. I think that’s where this other bet comes in: we’re also going to be the infrastructure for all of this.
Yeah.
Does it make ultimate financial sense? I don’t know. I think it’s fair for their stock to be punished. I do kind of admire it. They feel more like any other company. They’re a company that will live forever.
Yeah, because in part they’re so hard to understand, and in part because they’re so undisciplined. There’s just part of that that is—
It’s the blob. Bet against them at your own risk. If not for the Google Capital Company trajectory, I would be significantly more worried about Google’s position today than I am.
I mean, the cloud numbers are incredible, and they’re not stopping. The amount of money they’re making—the profit margins are skyrocketing quarter after quarter.
And the barriers to entry for future competitors are impossibly high.
And, by the way, a new participant in the call: Thomas Kurian.
Oh boy, that’s right.
That’s kind of a promotion right there. I saw some people asking, “Why is he on the call?” He’s the reason that they’re not—whatever it might be. To me, it says business is about trade-offs, and Google’s trade-off has been, “This is our priority.” All the evidence points in that direction, and you know what? I admire a company making choices.
Fair enough.
While OpenAI says the volume of search-like queries in ChatGPT is now 1/5 that of Google, and that tracks with my own usage, Max Anderson on Twitter had a long tweet about what Google has been doing with its search business: “Manufacturing revenue growth via short-sighted, highly extractive tactics.” Isn’t it?
It was from a week ago.
No, it was from July 23, 2026.
Oh, okay. Fair.
No, it was a big thing over the summer. There were people pushing back on some of this stuff. How much of it is true? How much isn’t? I don’t know to what degree this tweet is true or not, but it’s not implausible that Google is absolutely—if they think—
If they see the writing on the wall and say, “Get as much money out of it as you can.”
No, it makes sense. For me as a user, I just marvel at how much better ChatGPT is than Google these days. I now use ChatGPT for 80–90% of what I used to use Google for, as I’m writing every week and need articles.
Or even if you’re looking for a specific link, where it used to be like, “Oh, you—”
That’s when I use ChatGPT. It’s like, I know exactly what I need, but it’s impossible on Google.
Right. There was a while when it was like a Wikipedia-type thing where it gave you the answer, but it was like, “No, I need a link to get to this.” Now it’s like, “Can you give me the link about this thing?” It’s like, boom, here you go.
5. Reddit’s AI Problem
It’s wild. Anyways, times are changing. Speaking of which, Reddit CEO Steve Huffman said on the company’s earnings call last week that Reddit is still a top search term in Google. “This does not say anything about internet consumers’ demand for Reddit. People still want Reddit. It is just Google is not sending them to Reddit the way they used to. Given our strategy is to make Reddit a direct destination, this is actually in line with our strategy, because Google is teaching people that if you want to come to Reddit, you need to come to Reddit.” Huffman’s valiant attempt at the end of that quote to spin the reality positively was unsuccessful. Reddit lost about 30% of its share value in the past week or so. Will the rest of the open internet experience an identical trajectory to Reddit as Google resets its approach?
I mean, you know, it’s always funny. One of the things I want to talk about is the article you put on Sharp Tech this week about X and xAI.
It was very painful for me to call it X at various points in that article, but here we are.
There was a VC I was talking to a while ago who had some money coming from Saudi Arabia or something like that. They were like, “Yeah, we used to debate this a lot, and we decided we’re just going to say we’re aligned with the U.S. federal government. If they see Saudi Arabia as an ally, fine, then they’re an ally from our perspective.” Now, is this very self-serving? Yes, of course it is. But there’s also an aspect of just outsourcing it to the official thing. Just avoid the controversy.
Sure.
At the end of the day, there are lots of trade-offs with these things. It’s the federal government’s job to decide who our allies are, right? Not me individually.
So, in that vein, when it comes to naming things, I’m like, let’s just align with whatever the official thing is. I just take the federal holidays, right? I’m not going to make those decisions. Sam Hinkie had one outfit that he wore every day to remove that decisional anxiety.
You can remove some of the burden. I’m going to control what I can control and also defer to the others.
Sure.
So I guess we’ll call it X. Whatever.
No, I’m going to violate this policy and just call it Twitter.
Okay, fair.
The worst thing is, that one is definitely X. It’s not Twitter anymore. There’s some debate about the Gulf of America—
Gulf of America. Gulf of America.
—or the Department of Defense versus the Department of War. Can you change that via executive action, or does it have to be congressional action? In those cases, you’re freer to choose.
Other companies—I feel like Meta is just Meta.
Yeah.
Not that it’s a good name, but they needed a new name because Facebook was a product and also a thing.
Mm-hmm.
But Google—we don’t call it Alphabet; we do call it Google. The thing about X and Twitter is just that Twitter is such a good name.
Yeah.
Anyhow, digression.
Well, there’s no Alphabet. It’s one product that’s always been Twitter, so it is what it is. On the Reddit front, I only asked because it reminded me what happened to Reddit.
First of all, what happened was inevitably going to happen to Reddit. So much of its traffic when it went public was coming from Google, which tracks with my experience. The only time I ever end up on Reddit is when I Google something and there’s a Reddit link at the top, and I’ll check it out. Google was always going to take that rug out from underneath them. And now, the article you wrote a couple of weeks ago, “The End of the Internet”—
So this is the other side of the coin.
This is what I was circling toward. You wrote this great article on X/Twitter, which I got to. So, the digression of the digression, to trace where we work: one thing we were talking about in the article—which I hope we talk about in a moment—is that it’s hard to get a feedback mechanism from X.
Mm-hmm.
We should call it X because it’s different from what Twitter used to be.
It’s dominated by For You. I have numbers I can look at—subscriber numbers, website numbers—and they’re doing very well.
Mm-hmm.
But some things, there’s not—sometimes there’s discussion, sometimes there’s not. The X algorithm has been very tuned to posts on X being pushed and promoted.
If you wrote every Stratechery Update and article as a long tweet, you’d get 100 times the engagement.
Right, which is, by the way, a horrific user experience. I hate reading those long tweets, and there’s lots of thoughtful stuff, so it’s extra painful. But it is what it is.
Yeah, it drives me nuts.
But it is what it is. This is the reality: deal with the algorithm, which is one relevant point to this discussion. One of the things I have to take as feedback is subscriber email numbers and things like that, and also, is Andrew excited to talk about what I wrote about? Okay, great. I’m glad I’m part of your process.
A couple of weeks ago, I wrote about the dying internet—or the dead internet. What did I call it? I had a better title: “The End of the Internet,” which I actually think was fairly optimistic. The group chat era, not defined by doing things at scale.
It didn’t make it to our rundown.
Didn’t make it to our rundown.
I forget what crowded it out, but we had covered a lot of that on the previous episode we recorded. I actually did enjoy the article, for the record, to make you feel better two weeks later.
So, anyhow, circling back—
Thank you for bringing it up just now, because that’s where I was going to go with Reddit.
I don’t recall that being addressed in there, because a lot of that was about cybersecurity and the erosion of trust we’ve come to enjoy over the last—
But part of it is also that everyone on the internet, when we talk about Google doubling down, extracting every last dime, or moving people to this new paradigm, is screwed.
They’re screwed, right? And, you know, it’s—
And so the entire economy that we’ve come to understand and experience as news consumers is going out the window. It’s been headed in that direction for quite some time, so it’s not really news, but I feel like it’s becoming more pronounced over the last 6 to 12 months.
Right. Now, that was a provocative title. It’s like the end of the internet—
As we know it.
Yeah.
Which is one market where everything is about scale. So much of what strikes me about this AI thing is that AI is so empowering on an individual level.
Mm-hmm.
But if you’re out there thinking, “I can vibe-code. I can make an app and sell it to the world,” no, that’s not going to happen. You and 10 million other people can do it.
Anyone else can do it. That’s right.
So many of these payoffs are actually very small-scale in nature. By the way, a lot of the internet is this weird bifurcation. The internet is this weird smiling curve: on one side, you have these astronomical platforms that serve everyone, and on top of them, you can build little gardens, like, say, Stratechery, which is predicated on there being these big platforms and free distribution to anyone, but it’s its own little world. The analogy I’ve talked about is group chats. One of the best social networks in the world is actually very small, very high-trust, with all the benefits of the internet: you can be connected from anywhere in the world to anyone, with anything.
So much better than the internet.
Right, and it’s so much better. The challenge is all the companies that were predicated on reaching everyone and having this funnel feeding them, which was Google. Reddit did very well in Google for the same reason it did very well in all the LLMs. But Google had always been almost the last truly altruistic one: doing it because it was the right thing, part of the whole “Don’t be evil” ethos. Part of it was, “We’re going to index your site. It’s going to make our product better, but we’ll deliver a ton of traffic.” And, by the way, we’ll also build an ad network for you, AdSense, so you can help monetize your site.
And it’s kind of funny: there’s that lawsuit about the network and whether they’re monopolizing against publishers and things like that.
To some extent, I think the answer is no. Publishers’ problems are not Google’s ads. Have they acted in an abusive manner? Yes, probably. But part of that is the “Don’t be evil” ethos chipping away. You have to hit a KPI, you have to hit a bottom line, and you’re going to tighten the screws a little bit more. And why? They need you, I guess. That was—
Going back an hour and 15 minutes, these are all multinational corporations that behave in a sort of sociopathic way.
And still, at the end of the day, they’re going to send you a lot of traffic.
And it’s like we’ve reached the end state of that. Like—
Google—
Actually, we can’t do that anymore.
Yeah. Sorry, guys. It’s been a good run. We’re rolling out AI Mode, which kind of sucks, but it is what it is.
And so, for Reddit—for lots of companies—they need to shrink down—
—to what I talk about, like a Stratechery-type world, where you maximize your revenue per user, knowing you’re not going to have a maximum. You’re limited in the number of users you’re going to have.
Yeah.
Whereas most sites show ads to people who aren’t logged in and come in via Google, the assumption is you make it up in volume. The assumption now is you have to be logged in. If you’re logged in, it’s easier to track you, build a profile on you, and show you more targeted ads. We’re going to need to increase—I don’t think Reddit’s going to go to a paid model, but they need to dramatically increase the monetization of the ads they have.
Yeah.
Which means understanding their customers better. If you’re going to show people ads, you’re going to, at a minimum, have to log in so we can know who you are.
If you have a logged-in experience, you can build as much of a profile about someone. You don’t have to.
So they’ll do what they can, but it doesn’t sound like the future is all that attractive.
Well, they do still have the core user base who generates content for them for free.
Yeah, Reddit still makes all this money from selling access to its data to LLMs, which is super valuable because it’s continually fresh data.
Mm-hmm.
And you remember, you go back to the general internet paradigm: 1% of people post, 9% respond, 90% just lurk.
Yeah.
At the end of the day, the people who generate all the value for Reddit are the people who post and respond.
Those people are already logged in.
And so there’s an aspect of that: the other 90% actually aren’t that core to what is the most valuable part of their business.
Okay.
And so this is where they really lose. And, by the way, if people are logged in, maybe they’re more likely to respond.
Maybe they’re more likely to actually respond. It’s going to be a painful process, but it makes a lot of sense that they’re going in the direction they’re going.
Well, is it a situation where, because they’re a public company now, they’re beholden to growth and engagement statistics?
No, but that’s why you said it was a valiant attempt to spin the reality. I see it as a valiant attempt to state reality.
Yeah. And the stock took a hit because they have to fundamentally—
Reorient the business and how they’re measuring success.
The orientation of the business is reorienting for them. It’s acknowledging what’s happening, and shifting to what makes sense going forward.
Fair enough. All right, we have about 5 minutes here. Do you want to talk about my article, AI show and tell, or both?
6. AI Show & Tell
I already predicted AI show and tell was going to be higher up in the post. We’ve gone long. It was pretty high in the post. We went long on data centers. Let’s do AI show and tell, and we can close with my article quickly.
John says, “Hello, Ben and Andrew. In one of Ben’s recent posts, he used the phrase ‘the group chat era’ for apps, which perfectly encapsulated so many recent experiences of mine.
“At work, I had a tiny software-sized hole in my workflow. When I described it in detail to Claude, Claude built me a solution in 3 minutes. It’s not anything I’d ever share with anyone else, but it lets me adapt software to the way I want to work, rather than adapting the way I work to existing software.
“I also built an app that started as a way to help me with my last job search. I turned it into a multi-user app when friends noticed it and asked if they could use it, too.
“My wife downloaded an app from the App Store that acts as a front-end wrapper for the school lunch menu for the public schools where I live. The information is posted publicly, but it’s hard to find. Parents could technically navigate to the page and add it to their home screen as a web app, but downloading an app and setting up notifications is the workflow that most adults know now.
“This is in a major U.S. city. It’s not like the public school system couldn’t build something like this. It’s just that if it did, it would probably be pushed to do way more than tell people what was for lunch the next day.
“In the same way group chats meet our social needs better than mass public social media ever promised they could, this new paradigm seems to be about building software that meets our functional needs better than mass apps ever could. But there does feel like there’s a missing middle piece between ‘I build stuff for myself’ and ‘a company builds stuff for the masses.’ If I’m building software for myself and a small group of friends—the group chat of software—what do the payments, authentication, and security layers look like? And who is positioned to provide them: the big players like Google and Apple, the AI companies like OpenAI and Anthropic, the software stack builders like Vercel and Supabase, or some new, undeveloped part of the value chain? Any thoughts on that?”
Well, that reminds me. Number one—
The cool thing about when you’re small-scale—
—is suddenly the urge and the payoff from acting altruistically returns. This is a benefit of being small. I keep saying, “Oh, tech doesn’t do the right thing.” You’re like, “Well, they’re multinational corporations.” Yes, that’s part of the problem. The bigger you get, the larger-scale you are—
Duties to not do the right thing.
Right? Who can even define what the right thing is at the scale of billions and billions of users? That’s part of the argument for more local control, local government. When you’re dealing with your friends and family, your neighbors, or people close to you, there’s more scope for acting, quote unquote, inefficiently because it’s the right thing to do.
The payoff of doing the right thing isn’t just, “I feel good about myself.” People see you’re a good person, right? There’s a status payoff. The way to think about some of these vibe-coded apps is that not everyone’s going to vibe-code apps. You have to think about it, identify the problem, and all those sorts of things.
Yeah.
But making apps for your local community that show the lunch menu isn’t a money-making opportunity. It is a status-enhancing opportunity. “Oh, you made that app. It’s so great.” There’s an optimistic part of the end of the internet: the return to smaller-scale aspects creates more opportunities to do things because they’re lower-scale, because they’re more inefficient.
There are a lot of money-making opportunities in inefficiencies. Usually, we think about that as selling efficiency, but sometimes an inefficiency is like having your local sports columnist back in the day. There used to be a sports columnist in every town in America.
Yeah.
Then you get to the internet, and there’s one sports columnist, and his name is Bill Simmons. Sure.
Right. The internet at scale wipes out all these intervening layers. I’m kind of curious whether that same sort of trajectory and shape will exist in the new reality.
That’s what’s exciting. If you think, “Oh, I vibe-coded this app. I don’t want to be in this security environment. I don’t want to publish this or be responsible for it.” And by the way, there’s no money to make anyway because there are five gazillion vibe-coded apps. The competition is massive. But I have a group of friends. “Oh, yeah, you have the same problem here. You use this. I made this up.” You could share this sort of thing. Maybe at a very small scale there are money-making opportunities, or it’s part of a business that you build—
But you definitely worry about security in that scenario. So who solves security?
Well, all these things are to be solved. I think the shape of things might be that everyone’s making the same app for their friends and family—
And it’s like, that’s so inefficient.
It’s inefficient, but it’s also better because it’s expressly tuned to your needs.
Sure.
In a world of AI doing all this stuff at scale for, quote unquote, free, when we talk about humans creating needs and desires, it’s kind of humans creating inefficiencies for each other, right?
Right. The most utilitarian thing is to wear the exact same thing, wear the same clothes every day. Why is there a fashion industry?
Because—
Because people are very inefficient about wearing clothes, right? You know what I mean? To what extent does this new creation of new desires and new markets—which the tech mindset sees as inefficient—actually… No, it’s welfare-enhancing to have more inefficiencies in many respects, in a weird sort of way. That’s the optimistic view of AI: it fits into this. How is it all going to work? To be determined, for sure.
All the AI companies are trying to build their own cloud environments where you can host stuff. Part of that is to lock people in; part of it is just a problem to be solved. That’s where I’ve integrated the stack.
I don’t personally feel the need to create many apps on my own, and I feel a little inadequate sitting next to you, who’s the king app maker. But if my wife makes an app, I want her to be able to share it with me easily and securely. From what I hear—
She is. She’s referred to as the app queen at her consulting firm.
7. The End of the Internet as We Know It
If she makes something, I want it to be very easy for her to share it with me securely. One interesting thing, as long as we’re on the Google fanboy tour here: Google actually released this feature a few weeks ago. I never got a chance to write about it. They created a whole part of the Google Play Store where you can publish your own apps just for your friends and family.
Okay.
With Apple, you can kind of do this with TestFlight, but it’s a lot of rigmarole. You have to have a developer account, and people have to join and download it. This is just built into the Google Play Store. I can publish my app, and the people I define can get access to it. I think there are more and more opportunities for Apple and Google. Google is ahead on this point. I hope Apple pays attention and catches up. That’s great. There’s big—
That’s exactly what we need.
Well, then there’s all this hosting stuff, right? I have my own hosting and security, and it’s all a—
Like, right now it’s fairly insane, but—
It’s jerry-rigged.
These are all opportunities, right? The fun thing about this era is it’s very disconcerting. I was speaking to a bunch of MBA students the other day, and it’s very disconcerting to think about what jobs are going to exist, et cetera, which I understand. My advice is: the way to think about AI is that if you see AI as an opportunity to do your job, you’re in trouble.
Okay.
The way to think about AI is that it’s an opportunity. What’s the advice I always give? Double down on your strengths and do the bare minimum to ameliorate your weaknesses—
Because that’s what you’re going to get paid for: your strengths, not your weaknesses.
AI is like a machine gun you can train on your weaknesses.
Take care of all this stuff, so you can spend more time and attention doing what you’re actually good at. It’s funny: everyone thinks about AI and applies it to their primary thing.
AI is actually most useful for all your secondary things, all the stuff you don’t want to do.
I forgot how I got on that point, but I’m super optimistic about this possibility. The good thing for the MBA students is that when everything’s being shaken up, there are way more opportunities to build something new.
If you were building something on the internet in the Web 2.0 era, there were so many possibilities. By the late 2010s, there wasn’t really an opportunity to build anything. It had kind of been done.
The consolidation.
Exactly. When something’s new, it’s easier to differentiate yourself, in part because—
The dominant players have wiped out everyone else. They’ve become so tuned to that paradigm, the App Store paradigm, that they’re complacent and sclerotic. Well, actually, anything but complacent and sclerotic at their core business.
Okay.
But incapable of doing anything else—
Anything new, right? Sure. So there’s a lot—
Unbelievably optimized for their actual core business. Sure.
Yeah.
So, this email is cool. I love the lunch menu thing.
It’s such a great example of something that would make my life easier. I don’t need a public school lunch menu app, but I can imagine how much easier it is to open an app than go to a website every day, or the day before school. If a public school system released its own app, I’m sure it would junk it up with all kinds of different crap. Hopefully, we’ll have lots more examples of something simple and elegant that isn’t trying to do too much.
I remember I had the local news business model I wrote about ages ago. Why doesn’t someone have a newsletter that’s just about Madison city government and local stuff that no one’s going to care about? It doesn’t scale.
But they have a cost structure that actually makes sense, unlike a newspaper that’s still carrying printing presses and trying to run a terrible product, particularly the website.
On the one hand, publishing was the perfect medium for a one-person business, because you really can do it all yourself. Still, most people didn’t want to do it themselves, and that’s why Substack was an amazing company to come along and make it super easy to get started. That’s evidence that there are markets much larger than you think if people don’t have to worry about costs—
Or they can operate with very low costs.
And so, if you’re a writer and you make $100,000 a year, you’re making more than the median—
Newspaper, particularly today. And AI has the potential to do that for so many more areas.
I don’t know that I buy some of the talk about a one-person company being a billion-dollar business. Maybe that’s going to happen. What I definitely buy is there being that many more opportunities for one-person companies that are $100,000 businesses.
There you go.
And so, platforms for that are going to be an opportunity. It might be the case that we need to see people succeeding by stitching it all together themselves, like they did at the beginning, and then a Substack will come along that answers the question, “Oh, that’s the platform that makes this easy.”
So it might be a timeline where people have to prove it out, and then people will see the idea and make it easy for lots of people to do.
I think this is—so many people in tech, all these people that I’m annoyed at, none of them thought Substack would work.
I myself thought about building a Substack and wasn’t sure exactly how many writers were actually going to be able to make money. Turns out, a lot.
I think people underestimate the ingenuity and market-creation capability of humans. When you take costs out of the equation, you don’t need to make that much revenue to make a living—
And make it all worthwhile.
And you make more room for experimentation and more possibilities.
Yeah. I love the group chat era of social media. I’m pretty excited for the group chat era of apps. I thought that was a good construction by you. On that note, let’s save the article for the Sunday mailbag. What do you think?
Okay, fine.
We’ve gone long enough. Great to see you, as always. We’ll be back. If you’re subscribed to Stratechery, we’ll be back in your feed on Sunday afternoon. For now, Ben, we’re going to take a little break and then get back to recording. But I will talk to you soon.
Talk to you later.