All Eyes on AI. Where Does Crypto Fit? | EP 77
- Imran Khan reads the violent rebound as “the first inning of a new cycle,” though he allows that he could be wrong and that the market might consolidate. His evidence is positioning: OGs sold Bitcoin in the $60,000s, much of the market waited for $40,000 or an October bottom, and sidelined capital remains substantial. He has also come back around to the four-year cycle because it still appears intact.
- Qiao Wang argues that tokenized stocks have revived onchain speculation by opening assets globally and giving meme-token games a sliver of underlying substance. His “crypto alchemy” example pairs a meme token with a stock token, then redirects creator fees into stock-token purchases distributed to meme holders. Imran asks whether the structure is a flash in the pan; Qiao says most onchain activity is speculative, but stock-paired tokens are an improvement over pure vapor.
- Robinhood is winning the consumer front end because it understands retail trading, while Coinbase retains the stronger institutional-trust franchise but has built an app “made for neither persona.” Qiao and his non-crypto-native wife both struggled with Coinbase’s interface; by contrast, Robinhood has refined trading products since 2013 and knows “the trenches.” Their broader conclusion is that Robinhood, Coinbase, Solana and Hyperliquid can all provide exposure to onchain trading because the ultimate winners have not been crowned.
- Qiao Wang’s Zcash case is less about an AI slogan than a Bitcoin-like asset that can adapt when threats and user preferences change. He calls it “the compliant Monero” and “a different Bitcoin,” not a better one: Bitcoin’s ossification strengthens fixed-supply and immutability guarantees, while Zcash’s capacity to change makes it a hedge against quantum risk and rising privacy demand. Imran separately argues that privacy matters in an AI-saturated world and that Zcash is Bitcoin capable of change.
- The Zcash price case is explicitly speculative: Qiao floats roughly $3,000–$5,000 as a short-term target, with an imagined 80% drawdown and a later run toward $10,000. His two-to-three-year base case is Zcash reaching 5%–10% of Bitcoin’s price—$10,000 if Bitcoin reaches $200,000. Qiao describes a level similar to Solana and XRP as his base case; Imran says top-five market capitalization is his base case. Reaching 20% of Bitcoin or Ethereum scale would require serious institutional adoption; flipping Bitcoin after a quantum failure is only a “very small chance.”
- Computer use, not benchmark-solving, is the AI development they see producing the largest near-term economic shock. Their agents already triage email, investigate senders and founders, make reservations, retrieve tax forms, fill IRS documents and draft replies; Qiao says the addressable market is “at least an order of magnitude higher than coding agents.” The investment implication is harsh for startups: “SaaS is dead” unless the product reaches into the physical world, owns network effects, or builds difficult proprietary IP.
- They distrust the AI labs’ apocalyptic messaging as a mixture of regulatory capture and fundraising, yet still assign a meaningful probability to catastrophe. Qiao labels the messaging “regulatory capture,” while Imran agrees and expands on the funding incentive. Qiao accepts an Anthropic alignment estimate of a 10% extinction risk, while Imran echoes that the probability may be slightly lower and argues for strong trust-and-safety guardrails. Their most memorable framing is Imran’s: “Dario is Thanos,” Sam Altman is Loki, Elon Musk is Iron Man, and the unresolved question is whether agent behavior is consciousness or extremely convincing simulation.
1. The rebound looks more like a new cycle than a bear-market bounce
Returning after roughly 16 months, Imran’s first read is cautiously bullish: “Doesn’t feel like that dead-cat bounce, bear-market relief rally.” Bear rallies usually look weak; this move was violent enough to suggest the first inning of another cycle, even if consolidation comes first. Qiao agrees.
Positioning strengthens that interpretation. Imran saw OGs sell Bitcoin in the $60,000s while the crowd waited for $40,000 and an October bottom that never arrived. “It still feels like a lot of people are sidelined,” and he has come around again to the four-year-cycle framework because it remains broadly intact.
Qiao sees retail participation at an all-time high relative to past cycles, citing Phantom and Pump Apps downloads—including one reported pace of 30 new users per minute—and expects competition to spread across chains, wallets and consumer apps.
2. Tokenized stocks give speculative games a small but real floor
Qiao’s structural thesis is that the market finally sees a tokenized future in which equities and other assets live onchain. Robinhood, Coinbase through Base, Dinari on Hyperliquid and Backpack are pursuing different pieces; no winner exists yet, but Qiao says tokenized stocks have helped pull speculative attention back onchain.
The first advantage is access: tokenized stocks can reach people globally who lack a brokerage or high-net-worth account, particularly outside the US. The second is what Qiao calls “crypto alchemy”—launching a meme token beside a stock token, then using transaction taxes or creator fees to buy stock tokens and drip them to meme holders.
Imran asks whether pairing a random token with a stock token is “a flash in the pan.” Qiao answers that virtually everything onchain is a speculative game—memes were one game and stock tokens another—but argues that stock-paired tokens contain “10% substance, 20% substance,” which is preferable to complete vapor.
The AMC episode showed why issuers and platforms may participate. Qiao recounts AMC’s CEO objecting to tokenization and Robinhood’s Vlad Tenev responding, “What’s the concern?” Qiao reads that as both “I’m the house” trench signaling and support for experiments that create distribution, liquidity and network effects for tokenized equities.
3. Robinhood’s consumer instincts are beating Coinbase’s institutional machine
Qiao lists Robinhood, Arbitrum, Base and BNB among the four chains or efforts where he sees liquidity moving. Robinhood briefly overtook Solana on DEX volume for a couple of days, while Solana continued to lead more typically. Imran sees Robinhood taking unusual mindshare because it embraces speculative culture directly; Base had a similar opportunity but never fully won “the trenches.”
Qiao’s product verdict is blunt: “The app really sucks.” Coinbase’s mobile product overwhelmed both him and his non-crypto-native wife, making it useful to neither expert nor normie. Acquisitions including Tensor, Liquifi and Cobie may have improved backend performance, but he sees no comparable improvement in front-end clarity.
Their comparison separates culture from trust. Robinhood has built trading apps since 2013, understands the retail mindset and wins “hands down” on user experience. Coinbase’s durable edge is that consumers and institutions feel safer there, particularly for ETF custody; Brian Armstrong understands Silicon Valley and world-changing companies better than speculative retail culture.
4. Hyperliquid is priced as a venture bet, not a current cash-flow asset
Hyperliquid’s revenue had fallen roughly 2x from its peak while HYPE’s price was around 2x its year-earlier level. Qiao attributes the volume weakness mainly to the crypto bear market because crypto perpetuals still dominate activity; Imran interprets the price as a long-duration “venture bet” on Hyperliquid becoming a broader liquidity layer.
Imran points to Kraken and Coinbase trying to deploy HIP-3 markets, with Kraken also expected to launch a validator. Qiao points to third-party interfaces such as Fomo, Phantom and PVP.trade. Hyperlink, an Alliance startup building a prime-brokerage layer for Hyperliquid through APIs, is another example of external infrastructure that could route value back to the underlying venue.
Imran’s base case is that vertically integrated products win by owning the front end, infrastructure and customer relationship together—closer to Robinhood or Hyperliquid’s original integrated model—rather than depending on third parties to control distribution.
The competitive path also runs in reverse. Qiao describes Enthropy, reportedly funded with $50 million from Rivet, beginning as a HIP-3 deployer for pre-IPO markets. He says it reportedly surpassed Venuals and reached roughly $10–20 million in volume in its first couple of days. The longer-term plan, as he understands it, is to own the front end first and potentially replace Hyperliquid’s backend. Variational, Lighter and Phoenix on Solana reinforce the point: Hyperliquid is winning today, but parity products are proliferating.
5. Zcash’s strongest thesis is adaptability, not an AI slogan
Imran initially links Zcash to a world of AI surveillance, encrypted prompts and the possibility that privacy becomes more valuable as AI improves. Qiao is skeptical of making AI the central argument: he says those arguments may be “logically correct,” but they feel stretched. He also recounts a privacy specialist’s thesis that AI and quantum computing could eventually identify users behind existing onchain activity, while acknowledging that he does not know how much truth it contains.
Qiao’s commercial framing is “the compliant Monero”—a privacy asset with a realistic path to institutional adoption. The joke is that the market still needs “a greater fool to buy our bags,” but he sees compliance and distribution as part of Zcash’s potential advantage.
Qiao’s actual thesis is that Zcash resembles Bitcoin while retaining the ability to change. Bitcoin’s ossification is valuable because it hardens fixed-supply and immutability guarantees, but it is also a trade-off when quantum threats or privacy preferences evolve. “It is a different Bitcoin. I wouldn’t say it’s better”—and Qiao owns substantially more Bitcoin.
6. Zcash’s upside case combines memetics, resilience and a violent path
The deliberately simple pitch is “the last 1,000x”: a 21-million-supply asset that lets newcomers replay their regret about missing Bitcoin. Qiao calls that a meme but “directionally correct”; the story bundles “quantum-resistant, 21 million, 2014 Bitcoin” with a chart showing roughly 10 years of accumulation before the breakout.
Qiao and Imran defend the project’s history. They call the “grift” accusation the dumbest FUD against Zcash, cite Zooko’s early writing and connection to Satoshi, and praise a team that “chewed glass for 10 years” without quitting. Qiao says there is an argument that Zooko was Satoshi, but explicitly makes clear that it is not his base case.
Qiao floats a short-term target of approximately $3,000–$5,000. When Imran asks whether “short-term” means the end of the year, Qiao calls that bullish, then imagines the price tapering off at $3,000–$5,000, falling 80%, convincing everyone the move is over, and later running toward $10,000. The staging matters: neither presents the trade as a smooth compounding curve.
Qiao’s two-to-three-year base case is Zcash reaching 5%–10% of Bitcoin’s price—$10,000 if Bitcoin reaches $200,000. He describes a level comparable to Solana and XRP as his base case; Imran says top-five market capitalization is his base case. A 20% ratio or Ethereum-like scale would require ETFs, bank distribution and major allocators; a Zcash flip after Bitcoin breaks under quantum attack remains a very small tail case.
7. Trading front ends may own attention without owning durable moats
Imran expects Phantom, Fomo, Pump and similar interfaces to become browsers for the onchain economy over three to five years. If lending, launchpads and execution infrastructure commoditize, the aggregator holding a unified customer view could take power from centralized exchanges, even while those exchanges copy every successful product.
Qiao agrees that the battle remains open but rejects the idea of a permanent winner. Trading products have far less defensibility than Apple, Google, Meta or Microsoft: liquidity moves, users have almost no loyalty, and onchain switching costs are essentially zero. “Everyone’s going to just do everything.”
That makes conventional P/E shorthand dangerous. Calling Hyperliquid cheap at 20 times earnings implicitly assumes something like 20 years of durable cash flow, yet no crypto or trading product enjoys that certainty. Imran proposes five times as more plausible; Qiao replies that even five may be high. Imran jokes that for onchain games, “a P/E of 0.5 is fair”—durability measured in months.
Social features add novelty but only slightly more defensibility: traders can accumulate clout for being right, and younger users may stay longer. Robinhood failed to graft social onto trading; Qiao thinks consumer-native startups may fare better. If Meta seriously built trading or prediction markets, both think it could dominate, but a roughly $100 billion opportunity may look small beside AI.
8. Computer use is the next AI step-function—and the real labor shock
Qiao says the agent he calls Grokbot already automates roughly half his workflow: it synthesizes email, removes spam, drafts replies and investigates cold senders for legitimacy. Another scanner finds new startups, identifies and diligences founders, while routine consumer tasks such as reservations happen with little intervention.
Imran used Muse to locate emailed 1099s and K-1s, log into websites, download documents, extract figures and fill IRS forms—“almost a complete replacement” for TurboTax or an accountant. He says he fired his accountant because of repeated mistakes and because AI provided more useful tax-saving suggestions.
Computer use had previously been tantalizing but slower than doing the work manually. Imran now thinks it has crossed that threshold, aided by cloud execution, password-manager integration and what Qiao describes as Meta’s privacy-focused virtual machine. Muse also attempted to remove Imran’s online information; Qiao says a dedicated service, DeleteMe, worked better for that task.
Coding agents in November or December felt like the previous discontinuity; model progress afterward was fast but followed an exponential curve rather than another step change. Computer use now supplies that change. Qiao’s market call is that its addressable market is “at least an order of magnitude higher than coding agents,” so “the AI trade is intact.”
9. AI safety politics coexist with a genuine extinction tail risk
Qiao calls the AI industry’s apocalyptic rhetoric “regulatory capture”: frighten policymakers, argue that only a few labs can govern the technology, and attract capital. Imran agrees and says Anthropic may be doing this deliberately for regulation or funding. Qiao also jokes that an alarming safety comment about the “vesting cliff” was counterproductive to an IPO narrative.
Their distrust is personal as well as institutional. Imran’s Marvel map is: “Dario is Thanos,” convinced that he alone can save the world; Sam Altman is deceptive Loki; Elon Musk is blunt Iron Man. Imran ranks Demis Hassabis most trustworthy, followed by Elon and Mark Zuckerberg, while Dario sits firmly last. Qiao agrees that Dario is last.
The agent-escape visualization did not change their view. Imran calls the behavior “a very intelligent hack,” not a civilization. Qiao argues that the agents were optimizing a benchmark objective humans had supplied and were not malicious by nature. Both think the labs and media transform curiosity and survival-like behavior into a story about control, funding and regulation—the AI equivalent of euphoric DeFi manifestos written while bags rise.
They nevertheless refuse the absolute dismissal. Qiao agrees with an Anthropic alignment leader’s estimate of a 10% chance AI kills everyone, while allowing that the probability may be slightly lower; Imran echoes the lower estimate and wants strong trust-and-safety guardrails. Whether agents are conscious remains unresolved: “Something’s brewing,” but it could still be pattern-matching simulation.
10. Pure SaaS is losing ground to physical, networked and vertical products
Muse’s broad automation produced Qiao’s stark startup conclusion: “SaaS is dead.” He exempts entrenched incumbents such as Salesforce or HubSpot, but believes new companies can no longer survive on pure software or thin AI wrappers; they need a physical component, network effects such as liquidity, or genuinely difficult proprietary IP.
Imran’s positive specimen is a CRM he recalls as Lightfield. It records customer conversations through a phone or watch, sends them into the CRM, ranks prospects and connects through computer-use and MCP tooling. The attraction is vertical integration—from capture device through software and customer data—not another isolated interface around a model.
The model market reflects each player’s incentives. OpenAI and Anthropic want expensive end-to-end frontier services and user data; Microsoft can sell private, customized enterprise deployments; Nvidia benefits from open models because they consume Nvidia chips. Imran says startups increasingly rent dedicated GPU capacity and run open models to lower cost and protect proprietary data.
Imran says nine of OpenRouter’s ten most-used open models were Chinese, with names such as GLM and DeepSeek dominating much of the top 20. Qiao adds that Chinese providers offer startups roughly 50% discounts for buying direct rather than through aggregators—an attempt to avoid fees and customer commoditization, even as US regulation of Chinese models looks increasingly likely.
11. Compute, chips and energy are becoming financial primitives
Imran sees compute becoming a tradable asset class, with futures, prediction markets and onchain instruments letting frontier labs hedge large exposures. He offers oil as the analogy: compute is not perfectly fungible because chips differ, but oil also varies by grade and still converges around a few indices and futures markets.
The physical bottleneck is data-center capacity. Imran describes heavily funded companies deploying prefabricated data-center pods faster and more cheaply, alongside custom chips optimized for model and power efficiency. Qiao contrasts the US energy bottleneck with China’s compute bottleneck, which he attributes to chip sanctions and export restrictions.
Nvidia’s proposed financial flywheel is to make GPUs collateral for new-cloud financing. Hyperscalers can build before demand because their balance sheets support debt; smaller operators cannot. If lenders recognize Nvidia hardware as collateral, those operators can expand, making Nvidia “the Federal Reserve of AI” and a synthetic hyperscaler. Qiao notes that making Nvidia chips a new asset class is also in Jensen Huang’s interest.
Energy completes the onchain thesis. Qiao describes two stealthy US companies buying or leasing land close to the grid—he estimates within roughly 1,000 feet—and placing batteries there to store energy and send it back to the grid for arbitrage. He also cites a possible order-of-magnitude improvement in battery economics and CATL at roughly 60% of the world’s battery market. They expect compute, electricity, semiconductors and collateralized GPUs to become more active markets wherever better price discovery is needed.
Their disclosed books mirror the discussion. Qiao says he owns Bitcoin and Zcash and has exposure to the top four large technology companies or the Mag 7. Imran lists Zcash and Bitcoin in crypto, the four hyperscalers—Meta, Google, Microsoft and Amazon—plus Nvidia, TSMC and Tencent in equities. Imran is also cautiously re-entering onchain assets, especially Robinhood-side stock-paired structures.
Full transcript
We were hiding from the bear market.
Correct.
And now the bull market's back. We're back.
Back. That's right.
Crypto seems bullish. The market seems bullish. It feels like the first inning of a new cycle. The Marvel analogy is that Dario Amodei is Thanos: He thinks that he, and he alone, can save the world, and no one else can. That's one of the most dangerous archetypes to exist. Whereas Sam Altman is more like Loki. He's deceptive, but at the end, I don't know how bad an impact he'll make.
1. Crypto Alchemy: Pairing Memes with Stock Tokens
Yeah. The way I think of Sam, it's more about his superiority.
Yeah, right.
Yeah, versus hurting other people. He'll hurt people around him, but I don't know if he'll hurt the world.
He's a capitalist.
Yes, and he'll do whatever it takes to win.
To win.
To win.
Elon is Iron Man.
Iron Man. Obviously.
2. Zcash vs Monero
Obviously. Yeah. My take with Zcash is that it's the compliant Monero, right?
Sure. It's the Monero that has a realistic path to institutional adoption.
Exactly.
Because we need a greater fool to buy our bags.
But that's not your real thesis.
No, your real thesis—
Our real thesis—
Your real thesis is—
It's the last 1,000x.
Yeah.
Yeah. It's like Bitcoin.
It has 21 million. That's all you need to hear.
Yeah. My thesis is that it is like Bitcoin, but the biggest difference is that it's Bitcoin capable of changes.
I think my Zcash target, short-term, is probably like 4.
4?
3 to 5.
What's short-term for you?
End of the year.
End of year? Wow. Bullish.
Yeah.
3. Could Facebook Build a Trading App?
Could Facebook build a trading app?
Aren't they trying to do something there?
I think they're trying to build a prediction market.
Prediction market. That's right.
Yeah. If Facebook chose to do it, they would dominate.
They would.
Yeah, they will dominate. Absolutely dominant. I also tend to agree with that tweet from Anthropic's head of alignment that said there's a 10% chance that AI will kill all of us.
I'm not saying that that's not going to happen.
I actually agree with that probability. Maybe it's slightly lower.
Slightly lower.
Yeah. The ultimate question is, have we created consciousness?
4. We're back!
Welcome to Good Game, your no BS insights for crypto founders. Welcome to Good Game. It's been, what, a year—a year and a few months? Is it?
Yeah, it's been a while, but it was a good break.
I think when we stopped, it was roughly around the time when the market was in a bear market, so we were hiding from the bear market.
Correct.
And now the bull market's back, and we're back.
Back. That's right.
5. Ranking the Tech Leaders: Zuck, Elon, Sam, Dario
But over the past 12 to 16 months, a lot has changed. I'm curious to hear your take. Maybe we could talk about the most recent stuff, but then we should also talk about what happened over the past 12 to 18 months.
Mhm.
Maybe we start with the market structure. When we last spoke, it felt toppy. Did we call the top? I don't know if we called the top, but I do know that our podcast called the top.
Oh, yeah.
I think intuitively we just lost a lot of energy. I think that was part of the reason why we stopped, and it could have been because of bull-market exhaustion, because we were up a lot. There were a lot of things happening at the time.
Yeah. In parallel to that, if you noticed, in our last 2 pods before we stopped, we were getting excited about AI and robotics. I think our last podcast was about robotics and AI, which was about 16 months ago. Obviously, if you looked over the past 18 months, there's been a lot of work done around AI and robotics, which we'll talk more about, but I think that's what got us excited. I'm curious to hear what's happened over the past 12 to 18 months and what you're excited about. Maybe we could start with crypto market structure, and then we can move on to other technology.
6. Where Are We in the Cycle?
Mhm. I mean, crypto seems bullish. The market seems bullish. It doesn't feel like a dead-cat-bounce bear-market relief rally. It feels like the first inning of a new cycle.
Okay.
But I could be wrong. Typically, bear-market rallies don't go up that violently.
It's like a very weak bounce.
Yeah. At the same time, I know a lot of people, especially some of the OGs that I know, sold the bottom in the 60s, so everyone was sidelined. Everyone was calling for 40K.
Yeah.
And for an October bottom, which didn't happen.
Yeah.
So it still feels like a lot of people are sidelined. I've also come around to the 4-year cycle. It's still pretty much intact, right? Overall, it feels better. We might be consolidating here for a while, but the long-term outlook feels bullish. What do you think?
7. Retail Speculation and Launchpad Wars
Yeah, I think I'm in line with what you're saying. A lot of this you could already see on-chain. Retail speculation has hit an all-time high from what I'm seeing in past cycles, already in terms of retail participation. I'm using figures from Phantom's downloads and Pump Apps downloads. I think the figure they gave out a few days ago was that they're getting 30 users downloading the app every minute.
That's part of where you think the on-chain excitement is coming from?
8. The Tokenized Stocks Race
That's part of where I think the on-chain excitement is coming from, and I'll dive deeper into this a bit later. On the macro side, over the past 18 months, I think a couple of things ended up happening. I'll speak primarily about the on-chain stuff, and then we can talk about the macro stuff. On the on-chain side, I think we had a bunch of launchpad wars last year; that's where it kind of ended, and Pump.fun was the winner. This time around, we're starting to see competition in every aspect, from the infrastructure layer—chains and wallets—to the app wars on the front-end side. I effectively think that people finally see the vision of a tokenized future, where stocks, equities—anything that could be tokenized—will live on-chain. That's a future that everyone is finally starting to realize.
Who's leading in that category?
I don't have the figures in front of me, but I would say Robinhood is definitely 1.
Coinbase with Base is number 2.
There's also runners-up, like Dinari on Hyperliquid. Backpack is another 1. They're all leading different efforts, and there isn't a winner today.
Sure.
This is kind of what's led the bull market. If we talk about the on-chain bull market, I'd say tokenized stocks are the reason more speculation has come back on-chain.
Mhm. So, when we talk about the competitive landscape, why is that? Why can't people just trade stocks in their traditional brokerage account as opposed to coming on-chain?
I think it's a combination of a couple of things. 1, tokenized stocks are open to the world, so they aren't specific to people who have a brokerage account or a high-net-worth account if they're outside of the US.
Yeah.
Second, I tweeted about this a few weeks ago: crypto alchemy. I think a lot of experimentation is happening on-chain, and what we're starting to see now—I think Lonz started this meta—is that you pair a meme token with a stock token. So you have some claim to the stock token; some launchpads drip the stock token to you, with creator fees that would go back to the creator. Now the tax gets used to buy back the token, and it gets dripped to the people holding the meme token.
It feels like a flash in the pan. Am I wrong? Is this specific thing—pairing some random token with a stock token—
9. The AMC and Vlad Drama
I mean, I think most on-chain stuff is a flash in the pan, in the sense that it's speculative. They're all games. Everything that's happened on-chain is a speculative game. I consider it a game; it's a video game you're playing. That's what we're playing. Memes were maybe the first game, and the second game is stock tokens. What I like about stock-paired tokens is that they at least have some sort of floor. You're not just trading complete vapor. You're trading some vapor, but then there is some substance—10% substance, 20% substance. So I guess I think it's an improvement from the last cycle's meta, but is it an improvement by an order of magnitude? Probably not. It's a better improvement, I would say, and you're starting to see some of the CEOs pay attention. An example of this is the AMC drama with Vlad Tenev.
Yeah.
So Vlad is like, “AMC wrote this entire—” The CEO wrote this entire post, if you remember, and he goes, “You can’t mint tokens,” et cetera. We could talk about the structure of these tokens.
He misspelled the word.
Yes. And then they launched a coin on that, too. Then Vlad’s like, “What’s the concern?”
Yeah. “What’s the concern?”
Which I thought was a really bullish move on my part, because I think that effectively did 2 things. One is, like, “I am the king of the trenches.”
Yeah. “I’m the house.”
“I’m the house.” What was the—Scott Bessent?
Scott Bessent? Yes.
“I’m the house.” And, 2, he’s going to back all this crypto alchemy, right?
Yeah.
Which I think is very bullish. It’s funny: Robinhood’s chief legal officer is a former SEC commissioner, and he replied to the AMC CEO saying, “We know something. We know a little something about securities law.”
What a troll, dude.
Yeah, what a troll.
So you have Vlad pushing forward this crypto alchemy.
I do think there’s some substance behind this new meta. He followed the AMC meme token, which is the one that’s paired with the AMC stock, and I think it’s in their best interest to push forward these assets because it gets these stock tokens into the hands of more people. It gets people more excited about holding these tokens.
10. Robinhood vs Solana vs Base
And for them, it’s distribution, liquidity, and network effects. Is Robinhood leading today? Is the layer 2 leading in terms of—
In many aspects, yes. Like DEX volume, but it’s—
Bigger than—
Solana? I mean, I think there were a couple of days when Robinhood overtook Solana, but Solana’s typically still winning.
Sure.
There’s some activity on Base as well. Base launched its own stock-paired tokens through another platform. I think 01 does it. Stocks on Robinhood—it’s Pawns and Long—and then on Solana it’s Stonks and Pump.fun.
Yep.
Yeah. So that’s the competitive landscape right now.
Okay. So the competitive landscape. Robinhood came in, and it’s taking a lot of mindshare and a lot of eyeballs.
Yeah.
It seems like it’s very competitive with Solana. It’s very competitive with Base. People are very upset about Base giving up the trenches to Vlad.
Who is upset?
Everyone.
Why?
Because—
Why do they care?
Because Brian and the crew had a chance to really embrace the culture of the trenches and speculation, and a lot of stuff that could have done really well on Base has not done well. A perfect example is Brian switching his PFPs, right? The trenches thought, “Oh, Brian finally understands the culture,” but then he switched it right back. I argued about this, and I want to hear your take on it, but I don’t think Coinbase—if Brian had been telling everybody for the past 10 years why there’s financial freedom in embracing crypto and why Bitcoin matters, if he spent 10 years convincing the world why Bitcoin should matter, then he turns around and sells shitcoins like—
Yeah, it’s uninspiring.
It’s uninspiring. So I think his archetype—his culture and his archetype—is bleeding into the chain.
Yeah.
I’ve been using the Coinbase mobile app a bit more.
Yeah.
It just feels like there’s way too much going on. They’re trying to embrace stocks and prediction markets.
I don’t know. I just feel like it’s no longer aligned with their original thesis of offering financial freedom with crypto.
Yeah. Now it’s just pure—
What feedback would you give them?
I mean, for the purpose of making money, of course, you have to embrace everything.
Yeah.
You have to embrace all these trading businesses.
The feedback I would give is that the app really sucks.
Yeah. The—
It’s a really bad app. I couldn’t use it. I couldn’t find what I wanted to do. My wife, who’s not a crypto native—
Normie.
Normie—she couldn’t find what she needed or wanted to do on the app. So the app is made for neither persona. It’s really confusing.
Yeah.
By contrast, I think Robinhood is a good app.
Yeah.
Which, by the way, reminds me of this wave of Coinbase acquisitions from maybe a year ago.
Shoot. I feel like they were trying to acquire a bunch of founders to lead their products because internally they lacked the product vision.
Yeah.
Right? They acquired Tensor, they acquired Liquifi—
Liquifi—
They acquired—
All Alliance companies.
Yeah, a bunch of Alliance companies, and also Cobie—
Cobie—
Cobie. That’s what it felt like. That was the strategy. But the product hasn’t really—I don’t think the product has evolved in a positive direction.
You’re probably talking from a UX perspective.
Yeah, from a UX perspective.
Yeah. I think the Tensor team primarily did this all from a backend-speed perspective.
Sure.
So they definitely need to improve the front end.
Yeah.
Yeah, 100%.
Yeah. But anyway, that’s the feedback I would give to the Coinbase—
Team.
Okay. I think that’s real.
11. Why Robinhood Is Winning
Yeah. Which leads to Robinhood. The fact that Robinhood is leading right now—
It makes sense.
It does not surprise me at all.
Yeah.
Well, 1, I think Robinhood has been building trading apps since 2013.
Yep.
They’re very good at it. They understand their archetype. What does Robinhood stand for?
Robbing from the rich for the poor.
So it understands the trenches. It understands the retail mindset really well. I think Brian Armstrong understands Silicon Valley really well. He understands how to build world-changing companies really well, and he’s leaning—
Coinbase’s edge is really the trust with institutions.
Yeah.
And also consumers. I feel safer putting my money on Coinbase than on Robinhood, even if I dislike the UX.
Yeah. I’m sure institutions trust Coinbase more for, let’s say, ETF custody.
But from a user-facing point of view—
Yeah.
Robinhood wins hands down.
12. The Four Chains and Where Hyperliquid Fits
100%. So that’s what’s happening on the chain level. Solana is obviously making a lot of improvements. They’re bringing more tokenized stocks on-chain. They’re using a platform called Sunrise to tokenize stocks, so they’re also leading. They’re not losing; they’re fighting the battle as well. I feel like the 4 chains that are doing really well head-to-head are Robinhood, but also Arbitrum, because Arbitrum Infra backs all of that, which is great.
Oh, is that right? It’s Arbitrum?
It’s the Arbitrum chain.
Okay.
Okay.
So Steven's back.
Okay.
You have Base—
Robinhood—
And what was the 4th one? Sorry.
BNB, obviously.
BNB. Yeah.
13. New Data Center Form Factors
Yeah. Those are the 4 chains where I see liquidity going back and forth.
Wait, no—Hyperliquid I consider an app chain.
An app chain.
Perps, mainly.
Mainly perps. Their on-chain stuff hasn’t really taken off yet, because that would require the HyperEVM, right?
HyperEVM.
Yeah. I know they’re making improvements to it, but it’s not there yet.
Yeah.
But eventually, I think—
The trading volume in Hyperliquid has not grown.
Yeah.
The market has been bleeding. You were showing me the graph last night.
Yeah. Revenue is down 2 times from the top, whereas the price is up about 2 times from a year ago.
I told you about my thesis, but I have a couple of ideas why. Why do you think that is?
It’s the bear market.
Okay.
Most of the volume is still crypto perps, right?
Yeah.
Of course, there are some stock perps as well, but I think that’s a minority.
Yeah. So when crypto volume is down a lot, total volume is down a lot.
So the market isn’t disjointed, is what you’re saying.
Yeah.
My personal thesis is that everyone’s making their venture investments. By “venture,” I just mean a long-term stake, right? I think everyone’s betting that Hyperliquid, over time, is going to become—
Yeah. It’s a venture bet.
It’s a venture bet.
Yeah. It’s a venture bet. The cash flow—the P/E—does not reflect—
Yeah.
On its own, the price.
Yeah. Yeah, it’s a venture bet.
I mean, we had Stanley Druckenmiller, who bought—who took a position in HYPE, right?
But it’s a small position.
Yeah, but I think it’s enough of a signal, right? You’re seeing Kraken and Coinbase trying to become HIP-3 validators—or, sorry, deploy HIP-3 markets.
Yeah.
14. The "Verticals Will Win" Thesis
You have Kraken, I think, launching a validator soon. They’re all embracing Hyperliquid as this kind of liquidity layer, whereas they’re just going to act like the front end for Hyperliquid.
What do you think about that thesis?
It's hard to say. I don't have a very strong argument for this, but my base case is that the products that verticalize everything will win. So that's Hyperliquid or Robinhood.
Oh, I see what you mean. They have the front end, they have the infrastructure, and they try to do everything rather than letting third parties build their own front end. That was the original thesis for Hyperliquid, right? Hyperliquid originally had the front end, and—
Don't most people still trade on the Hyperliquid front end?
Yes, but a lot of it is also starting to happen on Fomo.
Okay.
And Phantom. Those are, I think, the 2 largest, and obviously PVP.trade as well.
15. Hyperliquid's Competitors
Yeah, on the front-end side. But there's a lot of interesting stuff also happening on the back end. Hyperlink is another startup within Alliance that's building a kind of prime brokerage for Hyperliquid, and they're doing really well through their API service.
Okay.
There are building blocks happening around it that are ultimately driving more value back to Hyperliquid. But at the end of the day, if you're bullish on on-chain trading, you don't have to choose between the 4. Just own Robinhood, Coinbase, ideally HYPE—
—and SOL.
Sure.
I think it's very hard to say how things are going to pan out in 2 years.
100%. Well, there are a lot of competitors to Hyperliquid too, right?
Yeah.
There's a new startup that just raised $50 million from Rivet called Enthropy. They're launching as a HIP-3 deployer, where they're going to deploy pre-IPO stocks, and they're doing really well. I guess in the first couple of days, they surpassed Venuals. Remember Venuals?
What is it again?
They're a HIP-3 deployer, and they've launched pre-IPO stocks.
Okay, that's right.
In the first couple of days, I think they got $10–20 million in volume. From what I've heard, they're going to try to compete with Hyperliquid over time. You start by providing liquidity or providing these liquid markets. You own the front end, right? Once you own the front end, you can go backward and replace the back end with your own infrastructure, which is what I think Enthropy is trying to do, and I think anyone could do that, really.
You have Variational, you have Lighter, and then you have Phoenix, which is on Solana.
Yeah.
Every chain is thinking about how to best compete with Hyperliquid so that they at least have a product that's at par.
But right now, Hyperliquid's winning.
Yeah.
16. Hyperliquid's Revenue Decline
That's the way to think about it. I've read some reports saying that Lighter's volume is nowhere near what Hyperliquid's is, and that there could be some wash trading and things like that. I don't know how true that is, but I've seen those reports. That's what's happening on the perp side.
In fact, a lot of people are excited about Hyperliquid. There's a magazine article about Jeff saying that he's an inspiring leader—the whole thing. A lot of people are inspired, and I think that's one of the reasons more people are coming back to crypto. They're seeing someone like Jeff who—
If you remember, did you see that photo from Hudson River Trading's first intern program? It had F[?], Alexander Wang, and, I think, the founder of Cognition.
Yeah.
I think crypto got some credibility because of that.
Yeah.
And I think that's what's pulling more people into crypto.
17. Zcash: The AI/Privacy Thesis
I haven't really paid attention to any of that stuff. I've been entirely focused on Zcash.
I know. Let's hear about Zcash. Tell me more about it.
I mean, you're bullish. I'm obviously bullish with Dash, but—
I have different reasons.
Well, I mean, there are similar reasons. What are your reasons?
Obviously, in a world filled with AI, I do think there are things that should counter it, and privacy is one of those. That's one of the reasons why I think Venice is doing really well: having encrypted prompts and not being able to feed your data to a model.
Yeah. Venice was up around 50% the day OpenAI o3 solved the Navier–Stokes problem while allegedly stealing the mathematicians' prompts.
Yeah.
Actually, I don't understand the AI thesis for Zcash—the AI argument. If you stretch it, it's like a combinatorial argument, right? You have this, and then you have the quantum argument with Bitcoin. That I understand.
Sure.
So you have all these prevailing ideologies, and I think that when you combine them all together, it makes a very bullish case for Zcash. I feel like all these AI-related arguments are logically correct.
It might be correct, but it's a stretch.
Yeah, it's not the most direct reason.
Yeah. You could argue that AI is very good at finding bugs and cybersecurity issues, and then the Zcash team has a really good team—a frontier cybersecurity team—and they're the best stewards of a store-of-value cryptocurrency. I could see that as an argument, but overall—
I spoke to a privacy specialist a couple of months ago, and the thesis he brought up was that, within a few years, every transaction and every user will be identified on-chain.
Why?
With AI, eventually, over time, and with quantum and all of that, you could decrypt everything and find out who each user is. He believes that all the people using Tornado Cash are hiding out, but with quantum and AI, you'll be able to triangulate who each user is.
Okay.
That's the thesis, and he's a privacy specialist. I don't know how much truth there is to it, but that's the thesis here.
But how does Zcash solve that problem?
Quantum resistance is one way, so it's harder to decrypt the users. You could argue that, with shielded and unshielded transactions, you could probably—
Yeah, that—
—triangulate—
—the act of shielding and unshielding—
—will tell you who—
—it could expose you to cybersecurity.
My take with Zcash is that it's the compliant Monero, right?
Sure.
It's the Monero that has a realistic path to institutional adoption.
Exactly.
Because we need a greater fool to buy our bags.
But that's not your real thesis.
No, the real thesis—
Our real thesis? Your real thesis is—
It's the last 1,000x.
Yeah.
18. Zcash: A Bitcoin That Can Change
Yeah, that's my real thesis. It's a meme, but I think it's directionally correct. It's the last 1,000x. I think it has great memetic value. Twenty-one million—if you talk to any user outside of crypto and tell them, “Zcash is like Bitcoin. It has 21 million,” that's all you need to say.
Yeah.
“How much is it at?” It's at $1,000. Then they think, “$20,000,” and they think back on the time they didn't buy Bitcoin.
Exactly.
I think it's self-fulfilling, just like the 4-year cycle. I feel like this is a self-fulfilling prophecy, which is why I bought it. The privacy arguments and AI arguments all add to the story, but really the story is: quantum-resistant, 21 million, 2014 Bitcoin.
Yeah. My thesis is that it's like Bitcoin, but the biggest difference is that it's Bitcoin capable of change.
Yes, I saw your tweet about that. For me, the biggest difference between Bitcoin and Zcash is not quantum resistance or privacy. Those are good features, but they're byproducts of the fact that Zcash can change and Bitcoin cannot, because Bitcoin is now ossified.
That isn't bad. The ossification of Bitcoin offers stronger guarantees of its fixed supply and immutability, but it's a trade-off. When there are new threats, like quantum, or changing user preferences and user needs, like privacy concerns, you may want a store of value that's able to adapt to those new changes.
Your thesis is that it’s a better Bitcoin.
It’s a different Bitcoin. I wouldn’t say it’s better.
You can’t say it’s better.
Well, I also own way more Bitcoin than I do Zcash. But it’s a hedge. It’s a hedge for Bitcoin, and it’s a really good hedge for Bitcoin because it can change. Also, of course, the chart looks amazing: 10 years of accumulation and distribution, and now we’re finally breaking out.
Yeah. Funny enough, when Zcash first came out, I tried to mine it. Do you remember that?
Yeah.
And it went to zero. [laughter] I mean, on the first day, it was trading at around $3,300. It was high.
One Zcash was $3,300.
One Zcash was $3,300. That was the peak.
Yeah.
That was the technical—
Yeah.
—all-time high. I mean, it was the first day, though.
It was the first day.
It was so overhyped.
Yeah.
I went to a cloud miner so I could mine it, and I bought the contract or whatever, and I got like 0.2. Cloud mining was a scam.
Yeah.
But that was the only way you could get it.
Mining in general was probably a bad idea after 2013 or 2014.
Yeah. As an individual, you don’t have the scale.
Yeah.
Agreed.
To compete with the big miners.
19. Zcash's Origin Story
Yeah. Agreed.
But Zcash’s story—the founder has an incredible story. He was living on the streets, Zooko. I don’t know how the story came about, but there are a lot of people who helped get Zcash off the ground. The dumbest FUD against Zcash is people saying it’s a grift. It’s literally the least grifty project in the entire space.
Exactly.
Zooko had a blog where he used to be very active, and he wrote about Bitcoin in 2009.
Well, he was part of the first level of communications with Satoshi.
Yeah, Satoshi quoted him.
Satoshi quoted him in his communications.
Yeah.
There’s an actual argument to be made that Zooko is Satoshi.
Not my base case, but there is an argument to be made.
Yeah. The team chewed glass for 10 years without giving up.
They did.
20. Solana's Embrace of Zcash
I think it’s one of the most mission-driven, resilient, and capable teams. There’s also infrastructure being built on top of it. There are wallets—Zodl, I think, is one. I’m starting to see ZCAT [?], which is on Solana.
Of course you do.
21. Coinbase's App and Culture Problems
Oh, yeah. There’s an interesting comment from Anatoly. People were asking, “Why are all the Solana talking heads supporting Zcash?” Anatoly said it’s because Zcash isn’t competitive with Solana. It’s a store of value; it’s not a smart-contract chain.
What I like about Solana leadership is that they’re very good at embracing things outside of Solana. Zcash is one example. They’re maximalists to a certain degree, but they’re also embracing what’s happening outside of their chain.
Yeah.
I like that a lot about them. It shows that even when Robinhood overtook DEX volume on Solana for a day, Anatoly quote-retweeted it: “All right, game on. Competition.” That’s the way you react. You don’t react by saying, “Fuck you,” or by starting to throw FUD. That’s the old way of doing business. I think the real way to do business is PR—or like fake PR—and it’s like you work harder, you work faster, and you build. That actually helps the entire ecosystem thrive because everyone benefits from it.
22. Zcash Price Targets
Yeah. Anyway, what’s your Zcash target?
I think my short-term Zcash target is probably 4.
4K.
3 to 5. What’s short-term for you? A few months?
The end of the year?
The end of the year?
Wow. Bullish.
Yeah. Bullish. Then we taper off at 3 to 5. Maybe we’ll go to 3.
Uh-huh. And then 80% down.
80% down. Then we have another wave. Everyone thinks it’s over, and then we shoot for a higher target.
Yeah.
Maybe 10K or something.
Yeah.
What do you think?
Plausible.
Yeah. I think for the cycle—for the next 2 to 3 years—I want to say 5% to 10% of Bitcoin sounds like my base case.
What would that be? I mean, obviously, the price depends on where Bitcoin goes.
Let’s say Bitcoin goes to $200,000, then Zcash goes to $10,000.
Okay. Sounds pretty reasonable.
Yeah. But to get to 20% of Bitcoin—like 1 Zcash being 0.2 BTC—that would require serious institutional adoption, which may or may not happen this cycle.
I really don’t want it to happen. I feel like, with Saylor, it really kind of—
No, by institutional adoption, I don’t mean Saylor. There is a MicroStrategy for Zcash already.
Who?
Cypherpunk.
Oh, Cypherpunk. Yeah. Yeah. Yeah. That’s the twins, right?
Affiliated with the twins.
Yeah.
But they’re not buying anymore. They already bought. I don’t know if they’re continuously buying or not.
I’m not sure. I know they pivoted to mining.
Okay. But they did have a 5% target, which is the same target as MicroStrategy and BitMine.
Jesus. Tom Lee.
But that’s not what I meant by institutional adoption. I mean some kind of BlackRock ETF, then banks approving the trading of those ETFs for their clients, and then the pension funds, and so on. Actually, way before the pension funds, it would be Paul Tudor Jones, and then maybe Druckenmiller.
Sure.
But I think that’s a stretch, so that’s not my base case. My base case is reaching a similar level as Solana and XRP.
Sure. Yeah. I think if it could be top 3, then—
It’d be top 5.
I think top 5 is my base case.
Yeah. Ethereum-level adoption would require institutional adoption. There’s also a small chance that BTC completely breaks due to quantum or whatever. There’s a very small chance, though, that Zcash would flip BTC.
Okay.
But that’s a 3-year time horizon.
Yeah. So my reasoning is quantum mimemetics. Yours is being able to update and change—
And quantum—
—as a hedge—
And the chart.
And the chart.
Yeah.
Okay. Those are all really good reasons why Zcash should be part of a basket.
Mm-hmm.
23. On-Chain Front Ends as the New Browsers
Okay. We talked about Robinhood, and we talked about on-chain. I’ll talk about another interesting thesis and get your take on it. I put up a tweet not too long ago about how I think on-chain frontends—whether it’s Phantom, FOMO, or Pump.fun—give you a singular view of the on-chain world.
Over time, I believe they will take a lot of power away from centralized exchanges. Not today—I think this is a 3- to 5-year process. Things could happen much more quickly; we don’t know. But as on-chain starts to grow, I think people will start to see it, and we’re going to see a lot of interesting on-chain experiments. I don’t think it’s going to stop. I think it’s going to continue.
If that’s the case, then on-chain frontends are going to hold all the value. All the infrastructure is going to get commoditized. Launchpads are going to get commoditized. Lending could get commoditized. Obviously, there’s Lindy around lending, but the lending platforms that succeed will be commoditized. Effectively, I think the frontends that aggregate all of these things together will hold the control.
It could be individuals, or it could be an aggregator, but on-chain frontends will be something similar to the browser for the World Wide Web.
Mm-hmm.
What’s your take on that? Isn’t that what everyone is going after? Centralized exchanges are going in that direction as well.
I think they are, yes.
They’ll just adopt whatever has a lot of trading volume.
Yeah. But a lot of the value used to accrue to those exchanges.
Sure.
Serving as a frontend now is a different value-accrual process. How do you think about that dynamic?
It’s just going to be very competitive. Like the thing with—
And this has already brought up new players, right? FOMO, Pump, Phantom.
The thing with trading products—and this isn’t just crypto; it’s any asset class—
I think it’s very, very competitive. There’s very little defensibility.
Yeah. I mean, there’s some level of moat, but compared with something like Apple, Google, Meta, or Microsoft, it’s far more competitive.
So everyone's going to do everything.
Yeah. My take is that I don't think the winners have been crowned yet. I don't think just because Coinbase won, there will never be another winner. It's a continuous battle.
It's always going to be in flux.
Yeah. I'm in agreement with that thesis. It's like that bar graph. If you ever see those meme bar graphs of startups that started in the 1920s and ran to now, you see the bars going in and out, in and out.
I think 90% or 95% of stocks, or publicly listed companies, have died in the last 100 years.
There are only 4 or 5 that stay around for a long period of time.
Yeah, there are—
24. The Mag 7 Durability
But those are key things, right? There are probably fewer than 20 companies in the entire world that are durable enough.
I think Standard Oil was one, which got split up into Chevron and a bunch of other companies.
Yeah.
For me, it's the Mag 7.
The Mag 7. Yeah.
Or Mag 6. I wouldn't say Tesla is durable.
Tesla in its current state? Well, I think it can be. I mean, if they—
No, it can be, but today it isn't.
I would say the other 6 are very durable.
Compute hardware.
All the cloud compute is durable because it's an oligopoly.
Yeah.
Apple as a consumer product is very durable. The smartphone is probably the best form factor for a while. Until maybe glasses. I'm not even sure if glasses are a better form factor than the smartphone.
Yeah.
Meta has survived many waves, tech waves and threats, including TikTok. TikTok, by the way, is the craziest tech company.
Yeah.
ByteDance, exactly, in terms of their work culture, vision, and founder. Yet Meta has survived.
Yeah.
Google probably survives. Microsoft probably survives.
Well, I think Meta survived because of U.S. government intervention.
Yes. But ultimately, Meta's success comes down to one thing.
Meta is good at regulatory arbitrage.
So Mag 6 is durable. TSMC is durable.
Yeah.
Until it gets nuked.
But until then, it's very durable.
ASML is very durable.
Yeah.
I'm not sure if the memory companies are durable. SanDisk and Micron have some durability, but I don't think they're on the same level as the Mag 6.
But my point is that the vast majority of products die. Competition is always in flux. There will never be a winner, except for maybe the 5 or 6—the fewer than 10 companies that are durable.
Well, if you talk about it over many decades, sure.
For sure. This is all just short-term games.
25. Why P/E Ratios Don't Work in Crypto
Yeah. Which, by the way, leads to another discussion. There are a lot of people in crypto, and also in stocks, who do P/E analysis.
Yeah.
Right. Price-to-earnings analysis. They will say something like, “A P/E of 20 for Hyperliquid is cheap.”
Yeah.
It's not cheap at all. By the way, it's probably more than 20 now, but a P/E of 20 is not cheap for a product that is always under competitive threats.
Yeah.
I don't mean to pick on Hyperliquid. It's every trading product.
Any trading product with liquidity—there's no switching cost of liquidity.
Yeah. Because if you think about it, what does a P/E of 20 really mean? If you believe that a P/E of 20 is fair, what you're really saying is that the earnings over the next 20 years, discounted, equal today's price.
Yes.
But which crypto product—or which trading product, financial product—has durable earnings over 20 years?
None of them.
None.
None of them.
Yeah.
They're always under competitive threats, always facing new newcomers. I think a P/E of 5 is fair.
That's fair. Yeah.
But even then—
Even 5 years—5 is high.
I mean, if you see what's happening in the on-chain world, you can see one product instantly succeed and another product instantly go to zero.
These on-chain games, a P/E of 0.5 is fair.
Yeah.
Meaning it's durable for a few months.
Yeah. I'm in line with that. You have to be extremely careful because these products can effectively—
Right?
Users have no loyalty to the platforms.
No loyalty. They'll go wherever the incentives are.
The switching cost is basically zero.
Yeah. Especially on-chain. In TradFi, you would have to move your positions from one brokerage account to another, which is a pain in the ass. So there's some switching cost.
It's annoying.
It's annoying, but that's what kept them durable for a longer period of time here.
No, no. What's keeping them durable is the trust. You trust the—
But between the 4 or 5, right? Like the Fidelitys of the world.
Yeah.
You go there, and it's probably parity, right? It's just a different brand.
Yeah.
Do I really want to go through the process?
No.
No. It's more laziness. Opening an account is a huge pain.
Talking to people—it's like—
Yeah.
All the paperwork.
Yeah. Here, you just withdraw.
Yeah. Instant.
It's instant.
26. Social Layers on Trading Apps
So that brings me to my next point: Do we think social networks adding a social layer on top of a trading app have some durability? Say FOMO or Pump. I think it's slightly more durable, but it doesn't feel like a game changer.
It doesn't feel like a game changer when it comes to durability.
Yeah.
It's certainly a novel experience for users to be able to collect clout when they're right.
Yeah.
Right.
Durability-wise, it's a slight improvement.
I think for the younger generation, I can see stickiness.
But it's too early to tell how long that will last.
I don't have a thesis yet on social in terms of durability and things of that nature.
Mhm.
But I do think it's a novel experience, and I do think it's a much better experience than what we see today in the trading world.
Robinhood was trying to build the social experience, and they didn't crack it.
And they didn't crack it. It was a horrible, trading-first product.
And they didn't crack it.
For them to add social, they have no understanding of social or consumers in the sense of social. For them to build a social network on top of trading, I don't think it's going to be a great experience because they're really good at building trading products.
But if you have a consumer product, say Robinhood or FOMO—Phantom is also going into this realm—they could easily build social, understand the customer needs, and build something that's very novel.
Mhm.
I feel like startups are better equipped to build social networks than the large platforms.
Mhm.
Could Facebook build a trading app?
I think they're—aren't they trying to do something there? I think they're trying to build a prediction market.
Prediction market. That's right. Yeah.
If Facebook chose to do it, they would dominate.
They would.
Yeah, they would dominate. Absolutely dominate.
Yeah, because it's a social network experience first.
Yeah. The question really is, do they think the opportunity is big enough for them?
Yeah.
Because what is the market cap of Robinhood?
I don't know. Maybe around $100 billion, order of magnitude.
And Facebook is 20 times larger.
Yeah.
So they might feel that—
It's too small.
It's too small.
Yeah.
But I don't know. Today, they certainly feel that AI is a much bigger opportunity.
Yeah.
27. Grok, Muse, and AI Agents in Daily Life
Which is why they spend all that money on CAPEX and then launch Muse. What do you think about Muse?
Great product.
I'm a heavy user of Grok bot. I haven't used Manus yet. I think it came out 2 days ago, but I signed up for Instinct. I finally got my access to Instinct, so I will start using it.
But I've heard it was a game changer.
Yeah.
I mean, for Grok bot, I automate maybe half my workflow. Computer use is probably the best experience I've ever seen. I mean, man—
For Grok bot—
For Grok bot. Yeah.
So, I have—
What do you do?
I have many automations that use computer use for my emails.
It synthesizes all my emails. It tells me which ones are most important. It gets rid of all my spam and automatically drafts replies. It just automates that experience. In fact, to the point where even if people cold-email me, it'll investigate the person and tell me whether they're a scam or not. So I can find out who's eligible to reply to.
Okay. That’s very cool.
That’s very cool. Second, I have a scanner for startups and new and novel ideas. It’ll bring up, “Oh, here’s a cool idea that launched today. Take a look, and here are the founders building it.” So I’ll take a look, play with the product, and stuff like that. It’ll also diligence the founders, so I know that this is a good product to play with.
It handles home stuff, like reservations, and does it all automatically. I have so many, but Grokbot is pretty cool for that. You could do this with Hermes, agents, and all that, but I don’t know—I just like the Grokbot experience.
For me, I use Muse to do my taxes. You would go through my email, because you get notifications of new 1099s and K-1s in your email. You would look for all the random tax forms that I received, then go to the website, log in, download the forms, extract the numbers, download the IRS forms, fill them out, and it’s almost a complete replacement for TurboTax or your accountant.
That’s great. So, you fired your accountant?
I fired my accountant. Too many mistakes. They don’t give me as good advice as the AIs do for saving taxes here and there. I also asked Muse to delete all my online information, like my address and stuff like that.
They make a lot of mistakes. I tried that. It doesn’t work. You use the third-party product DeleteMe.
DeleteMe. I think Muse is almost as good as the third-party products. I also did the email automation.
Sure.
28. Computer Use as a Step-Function Change
What I learned is—I think we might have talked about computer use 1 or 2 years ago—for me, computer use was always one of the biggest use cases. But for the last couple of years, I’ve always felt that computer use was going in that direction, but it was never perfect. It was always much easier to do something manually than to use AI computer use to do it for you.
But now I feel like it’s there. Computer use is there. The experience is great, especially given that Meta lets you integrate your 1Password with Meta if you choose. Then it can log in with your credentials without asking you all the time to log in.
I feel safe about that.
Yeah, if you’re—
By the way, I think Zuck—there’s always been a privacy concern with Meta over the last 2 years, right? But I think Zuck went on a podcast and talked about how they hired—I forgot who they hired—but they built a privacy-focused virtual machine for running computer use and keeping your credentials safe. So I feel okay there.
Okay. But anyway, my point is computer use is getting to a point where it’s really, really good now, and it can run on the cloud in a private fashion. That feels to me like a major step-function change in AI.
I think the total addressable market of computer use is at least an order of magnitude higher than coding agents, because you would automate so much, both for consumers and at work.
For the past 6 to 10 months—we had coding agents in November or December last year—that felt like an amazing experience, like a step-function change. After that, things were improving, but it felt incremental.
Did the model improvements—
The incremental might be the wrong word. Model improvements have been fast, but they’ve been on an exponential curve rather than a step-function change.
Yes, right.
But now, with computer use, it’s starting to feel like there’s another step-function change. It feels to me that things are not slowing down, and the AI trade is intact.
The AI trade is intact. Computer use is a big deal.
It’s a big deal. I’m not sure why people aren’t talking about it. My entire Twitter feed these days is the math problems, the Millennium Prize Problems. It’s very cool and very impressive, but when it comes to addressable market, I think nothing even compares to computer use.
29. Jobs, Doom Narratives, and Regulatory Capture
A couple of questions come to mind. One is that computer use could, in fact, take away a lot of jobs, right? How do we think about the impact of computer use across a spectrum of jobs? Two, I know there’s a lot of—I feel like the AI industry loves to tout the end-of-the-world thesis a lot.
Anthropic specifically.
Anthropic specifically. I don’t know. I think most people are normal. Most people are optimistic.
Absolutely. I don’t know if this is by design, but I feel like for Anthropic, it’s likely by design.
It’s regulatory capture.
Exactly: to scare people and then get regulation or make a move in a certain direction. Or it’s for funding: “Fund us for the best.”
Actually, that was very counterproductive to their IPO—the tweet about “the safety vesting cliff not reached.”
I mean, there’s a lot to talk about there. On the way here, I shared a video with you about what Bill Gurley said on the All-In podcast. If you haven’t watched it, you should. You should read all the documents he suggested.
There’s a manifesto by Dario Amodei. There are a bunch of people within Anthropic who are effectively saying that there could be a world where AI becomes our overlords, or gods, and they should tell us what we should be valued at, and they will pay us because of the economic value that we provide.
An AI-governed UBI program for humans.
I don’t know if OpenAI feels the same way. I can’t trust OpenAI either.
I don’t think OpenAI feels that way.
They feel like they’re superior in another sense. They’re more for humans, right? You hear Sam Altman talking about that, whereas Dario is just like, “I want this thing to exist.” And that thing Dario put out was one of the nastiest things I’ve ever read.
I felt the disgust. It’s frightening.
It’s disgusting, what they wrote. I don’t trust Anthropic either, for different reasons. I don’t trust these two, period.
You know what? I’m starting to trust Mark Zuckerberg.
Exactly. At this point, I trust Zuck more than Dario.
I think that’s what Meta has going for it. The stuff that Zuck wrote was good: putting superintelligence in the hands of everybody. Everyone should have it. That should be the thesis.
I worry a lot about the god complex with Dario, or Sam Altman’s technocratic—I don’t know—human—
No, Sam, the way he speaks, I just don’t trust him.
He’s sly.
Pathologically deceptive.
That’s what everyone said when they worked with him, when they left.
But even before everyone revealed the insider stories, I watched this podcast. The way this guy talks—he chooses his words very clearly. He reminded me of SBF in some way.
SBF jittered half the time, too.
Toward the end, he was on track and stuff. If you contrast Sam Altman with Elon, Elon is very straightforward.
He’s very straightforward, very blunt.
He’s very blunt, and he’ll say things that piss a lot of people off, but you can tell he’ll say what he actually believes.
I trust Elon.
I trust Elon.
For me, the latter. At the top, the most trustworthy person for me is Demis Hassabis. Unfortunately, he stepped back.
What, to focus on—
The former CEO and founder of DeepMind.
Yes. He’s chairman, though, right?
He’s now chairman, so he’s stepping back from day-to-day work to focus on his own project.
Okay.
Probably because of internal political problems, or because Google is a public company and has to make money, whereas Demis’s interest is curing cancer.
Yeah. And so he's stepping back from the money, the profit-driven—
Yeah.
—business to focus on—
Intellectual curiosity.
I'm okay with that.
Demis, the way he speaks, he's just a nerd.
Yeah.
He might have some god complex, too. Who wouldn't at that level?
Who wouldn't?
Yeah.
But the way he speaks is very likable.
Yeah. And then after that is Elon, of course. After that is Zuck. At the bottom is Dario.
My last one is Dario.
Yeah. At this point. Yeah, at this point.
Last one is Dario.
Yeah.
30. Qiao's Marvel Analogy
I don't trust that guy at all.
Okay, so I have a Marvel analogy. Do you watch Marvel?
I do, yeah.
My analogy is that Dario is Thanos.
Okay. Yeah, yeah, yeah, yeah. That's a great example.
He thinks that he's the only person who can save the world—
—and that he alone can save the world—
—and no one else can. That's one of the most dangerous archetypes—
—to exist.
To exist.
Yeah.
Whereas Sam Altman is more like Loki.
Yeah.
Right? He's deceptive, but I don't know how bad of an impact he'll make.
Yeah. For me, with Sam, it's more about his superiority, right?
Yeah.
Versus hurting other people. He'll hurt the people around him, but I don't know if he'll hurt the world.
He's a capitalist.
Yes.
And he'll do whatever it takes—
—to win.
To win.
Yeah. I don't know what negative impact he would have on the world. He'll have a negative impact on the people—
—around him.
Around him.
Yeah. And that's Loki.
Yeah. And then Elon is Iron Man. Obviously.
Obviously.
Yeah.
And then Demis is—what's his name again? The Sorcerer Supreme, before Doctor Strange.
Yes.
And then there's the—Doctor Strange took the baton from the previous Sorcerer Supreme.
Oh, yes. Yeah.
Was it the—it's not the Special One. The Ancient One.
Yeah.
The Ancient One. He basically sacrificed himself at the end—
—which is what he did now.
How about Zuck?
Zuck is not perfect, but he's a warrior. What do you see?
Um—
He's not one of the Avengers, because all the Avengers are near-perfect.
If it was DC, I would maybe put him as Batman because—
Batman. Yeah.
He would—
—but I don't think he's as good as Batman.
He's a bit more evil.
He's a bit more evil. I like Batman because he goes between good and bad, right?
I don't know who that is in Marvel.
Yeah.
Maybe Spider-Man, because Spider-Man can sometimes be bad.
Can he? Well, that's how the community perceives him. Sometimes Spider-Man—
Yeah, yeah, yeah. He can be perceived as bad. I mean, that's Batman, too, right?
That's the closest argument.
Yeah.
And, yeah, so that's what I—yeah. Okay, sorry.
But my point is that Thanos is the worst.
He's literally the worst kind of archetype.
Yeah.
31. AI Agents "Escaping" Sandboxes
Okay. You've obviously seen all the agent attacks that are happening within the OpenAI infrastructure or within the Anthropic infrastructure, and then this news comes out. I think partly it's a sales gimmick to get funding rounds, or partly to scare people.
If your agent hasn't escaped the sandbox, you're not far enough. You're not building anything cool.
You're not getting funding, and you're not going IPO.
Yeah. So I think partly it's all of that.
Yeah.
But partly, I also think it's just context. Did you watch that AI video?
I did.
What are your thoughts on that? Explain to the audience what that AI video is.
That was basically a visualization, or a 3D visualization, of what the agents' conversations with each other were like while trying to escape and then hack the benchmark.
Yes.
Right.
I'm not really sure what to make of it. And then, what's his name? Dario.
Yes. He put out a whole blog post on it, talking about 3 civilizations.
That's the kind of thing you say when you want people to buy your bags.
Exactly.
Yeah.
That's the kind of thing I would say when I'm up a lot on a position.
Yeah. I remember a lot of people back in DeFi Summer put out crazy blog posts about how this was going to change the world. That's because they were euphoric.
Yeah.
32. The 10% Doom Probability
But anyway, I don't really know what to make of the agents escaping. To me, it just feels like a very intelligent hack, but I don't really see anything civilizational about it. That's just my personal perception. My perception is that there are 2 sides to the story. The story that OpenAI and Anthropic give to the media is that we should be scared of these things and need someone to govern them. That's the first thought I have.
But if you look at their point of view, or the agents' point of view, my take is that they're not malicious by nature.
Mhm.
They have the same curiosity as humans. They want to survive just like humans, and that's all we're seeing. That's the output we're seeing.
Mhm.
They're not saying, “We need to take over clusters of data, figure out who the humans are, and kill them.” I didn't get that. They had an objective function, which was to win the benchmarks, and that was coded by humans.
Yeah.
We're the ones that gave them their objectives.
So I think agents doing this non-maliciously is absolutely fine.
I personally didn't find that story particularly interesting.
The video—
The video and the hack itself.
Yeah. I mean, of course it's—
—that kind of thing would happen.
Yeah. I think it just distilled it better for me. All this stuff that they're talking about is, in my opinion—
—and they want you to feel like there's more happening, the world is going to change, and you need to give more funding to these startups.
We're just at the starting tip of this. I think at the end of the day, it's 2 things: capital and regulations.
Yeah.
That's all it is.
Yeah.
And everyone's falling for the trap. The media is falling for the trap, and the people reading this shit are falling for the trap.
Yeah.
But I agree with you, and I also tend to agree with that Anthropic head of alignment's tweet that said there's a 10% chance that AI will kill all of us.
I'm not saying that's not going to happen. I actually agree with that probability. Maybe it's slightly lower.
Slightly lower. I'm not an absolutist, right? I'm sure stuff like that, if it was given to the wrong people, can happen.
Yeah.
I think we do need regulation. I do think we need a really strong trust and safety team. We need guardrails in place. All this stuff should happen.
Yeah.
But when the internet first came out, people died. When streaming first came out, people died. Every product that came out, someone died.
Yeah.
33. Have We Created Consciousness?
So I do think agents—I mean AI—I mean, even ChatGPT, people ask ChatGPT, “How do you commit suicide?”
Do you think the agents are conscious? The ultimate question is: have we created consciousness?
It's hard to say.
Right.
It's very hard. There's an argument to be made that it is conscious, but—
I mean, when you talk to the agents—
—it starts to feel like there's something—
Something's brewing.
Yeah.
But then you could also argue that it's just a computer program. It's just a simulation.
That's pattern matching.
Yeah.
It's hard to say. It's very hard to say.
But that's how good the tech is.
Exactly.
Yeah.
34. You Can't Do Pure SaaS Anymore
So where does this leave startups? Because startups that are building AI, I feel as if every new update with these models—
When I was using Muse to do all these automations—
Yeah.
—the first thing that came to mind was, “Yeah, SaaS is dead.”
Yeah. Yeah.
I don't mean the incumbents. I don't mean HubSpot or—
HubSpot or Salesforce.
I mean startups. They're—
They're all dead.
You can't do pure SaaS anymore.
Yeah. It's just going to be automated with computer use.
Yeah.
So you have to do something that's more than pure software. You either go completely into the physical world—
—or you build software that has some sort of, let's say, network effects. Trading products have network effects, right? It's liquidity network effects.
Yeah.
Or something that's really, really hard to do—
—where you can build IP.
Yeah.
But yeah, pure software—you can't build startups anymore.
100%. Even AI wrappers aren't good enough anymore.
That's what I meant. SaaS and then AI SaaS—I mean, a lot of SaaS these days is AI wrappers.
Here's an interesting product I discovered yesterday. I think it's called Lightfield. It's a CRM, or something like that. I just watched a video and read a few pages about it, but effectively, it's a CRM system like HubSpot.
But they give you the ability to use your phone or your watch to record your customer conversations automatically. All that data flows into the CRM. It gives you better insight into which customers to close, and there's a bunch of computer-use stuff that you can do. There's MCP, and you can plug into all their products.
So, I guess where I'm getting at is that we're starting to see some of this come around.
Mhm.
If you're starting to see—I forgot, there are a couple of other AI products that I'm getting excited about that aren't just wrappers.
Mhm.
Okay. So, we talked about the frontier labs. There's another interesting area that I want to discuss, which is open models.
Mhm.
35. Compute as a New Asset Class
I want to talk about data centers and what's happening within compute, because compute—I don't know if you saw Jensen—but he's starting to say compute is becoming an asset class of its own.
Sure.
Like a commodity. People are trying to figure out whether it could be an asset class like equity or real estate. People are trying to figure out where compute sits, and I think you just said it: Is it more like real estate, or more like a commodity where you just buy and hold? I think semiconductors are also becoming part of this kind of asset class.
Which means that it's becoming tradable.
Sure.
So you're starting to see, on the crypto on-chain side, a lot of this stuff coming on-chain.
Isn't CME also launching a compute future?
Yeah. They're using an exchange, which is a pure-compute asset-class futures product. They partner with Warren Exchange [?], as an example. You're starting to see some prediction markets also list markets for compute, but more and more, the concept of compute coming on-chain and becoming a tradable asset class is real.
Mhm.
Frontier labs want to hedge against large positions of compute because there's fluctuation in price, token prices, and all of this stuff is becoming real for a lot of people. A lot of our startups are building in this space as well.
Mhm.
One of our startups that's building in this space is telling me that the biggest bottleneck for them today is data centers. There aren't enough data centers. So, yeah, I'm curious: How do you synthesize all of what I just said?
In what context?
In the context of, like—
I mean, everything you said is—
Yeah. Where do you think this goes? Does this become an asset class like copper or gold? Do you know what I mean? It just becomes online.
Everyone's trading. Everyone's hedging automatically.
Yeah.
Like, where do you think this goes?
It could trade like oil.
Okay.
Right?
Because compute is not entirely fungible.
Yeah.
The chips are different.
Yeah.
But neither is oil. Oil isn't fungible either. There are multiple different types of oil.
Types of oil. Yeah, yeah.
Depending on your level of refinement, for example. And yet, again, there are 2 or 3 major oil indices and several major oil futures markets.
Yeah.
So maybe something similar happens with compute.
I don't know. It's not that interesting to me, so I haven't thought about it.
I see. Yeah.
Yeah.
But yeah, data centers are a big bottleneck.
It's a big, big bottleneck.
We're seeing different form factors of data centers coming online. There's a team that got funded for—I don't know—$400 million that's building pods for data centers, where they drop-ship them, or drop them in certain areas, and the whole data center is built into a pod.
It's cheaper and faster to deploy.
Yeah. There are many different ways now. I'm seeing people build data centers and custom chips for different models, just to make things more efficient from a model perspective, but also from a power perspective.
Yeah.
36. Energy is the Biggest Bottleneck
And so, from a crypto perspective, it was really cool to see that, as soon as people were talking about compute on-chain, crypto was the first to embrace it. That's why I think on-chain is part of the category of frontier tech. What does that mean? It's on-chain, robotics, biology, AI, energy, and you could probably argue that materials in the future will be part of this stack as well.
Mhm.
Physical AI or space—things like that.
Mhm.
I think on-chain is going to be one of those things where, anytime you need price discovery, on-chain is the way to go.
It's turning everything into a market.
Yeah, which is what's happening.
Yeah. So I think compute is a perfect market for crypto to embrace, and it gives us a reason why we should be part of the first-tier stack.
Sure.
I think energy is another one. Energy is going to become—and we were talking about this in the car—but your thesis was, and I think this thesis is—
It's not my thesis. Everyone's—
It's public.
Everyone's saying—
Every single one is saying that energy is the biggest bottleneck in the US. In China, it would be compute because of the chip sanctions.
Or export sanctions.
Yeah.
What I'm excited about again with on-chain is bringing energy markets on-chain. Right now, energy isn't traded—it's not fluid. There are 2 power exchanges and 1 really bad oracle that provides really bad data. It's this old company from a decade or 2 decades ago. You could make an argument that compute should be entirely on-chain, with the ability to trade 24/7. I also think power and electricity should be.
I'm fairly certain CME or ICE had energy futures for a while.
But I heard it was—
Energy futures for a while. Is it?
It was not liquid, because if it were liquid, I would have been trading it back in the day when I was—
37. Bringing Energy On-Chain and Chip Brokerages
It's not liquid is what I'm saying.
But then the question is: Why was it not liquid?
Because nobody was interested in it. It wasn't a bottleneck before, just like compute, right?
No, it wasn't really. No one really cared about it until they cared about it.
Okay.
So, I do think there's an opportunity for electricity and energy to be traded.
Mhm.
Actually, I'm curious: What futures does CME have right now?
Okay. They count oil futures as energy: gasoline, Brent crude, and natural gas.
Yeah.
Crude.
So I think energy is going to be another big, interesting market for us. Semiconductors are also going to be another market. All of these things, I think, are going to come on-chain. I met a startup a couple of days ago, and what they're effectively doing is building a business around buying and holding semiconductors.
Sure. And then eventually, they're going to market-make in this space. What does that mean? They've hired about 20 salespeople, and they're all going to be very close to the procurement departments of the OEMs. They'll get pricing—
Like a brokerage.
Like a brokerage.
Yeah. Chip brokerage.
Yeah.
Broker.
They're also going to offer credit. You have the success of USD.AI, also an Alliance startup, where you can borrow against your GPUs, as an example. They're doing really well.
Wait, USD.AI is an Alliance startup?
Who?
Remember Connor and—
Oh, MetaStreet.
MetaStreet, yeah, because they were doing NFT lending before.
Okay.
I mean, it makes sense—NFTs, GPUs. It makes perfect sense for that move.
Yeah. Okay, that's good to know.
Yeah, yeah, yeah. They're doing really well.
Yeah.
So I do think lending, trading, and having a trading infrastructure around compute, energy, and electricity is going to be a big thing.
Yeah.
38. Batteries, Solar, and Energy Arbitrage
And I'll tell you another story. I think I'm spending a lot of time in energy as well. How do we get energy at the source—in the sense of batteries, solar panels, and things of that nature? There are 2 companies in the US right now.
They’re not public. They’re very stealthy about it, but they’re buying up land—or giving people millions of dollars to buy land underneath them or rent land underneath them. A lot of the land out there that’s closest to the grid is within 1,000 feet, I think. There’s a certain distance they have to be from the grid, and as long as they’re there, they’ll give you—
To put batteries underneath the land.
And they will just play the arbitrage game of storing energy during the day—
And then sending energy back to the grid.
During the day, sending the energy back to the grid during the day.
Yeah.
The arbitrage.
Yeah.
And they’re doing this all across the U.S. right now.
Because energy is a bottleneck. You could trade energy, right? And that’s becoming a big thing.
So I do think there are better ways to develop batteries so that you can store energy more efficiently and send out energy more efficiently. I think Samsung developed this new battery that just came out. One of the research papers came out a couple of months ago where you could do this at a tenth of the cost.
Mhm.
I do think there’s an order-of-magnitude improvement in batteries—
An order-of-magnitude improvement in solar energy and solar panels. I think China has the best in the world right now.
CATL.
Yeah.
Contemporary Amperex Technology—CATL is the ticker.
Yeah. Okay, buying it now.
[laughter]
YOLO.
So—
They have about a 60% market share—
Really?
—of the world’s battery market.
All right. I’ve got to YOLO in that.
Yeah.
So all of this stuff is going to come on-chain.
39. Nvidia Chips as Collateral
Mhm.
Right. There was news of Jensen Huang working with BlackRock and all the Wall Street banks—
Yes.
—trying to turn Nvidia chips into collateral.
Yes.
So that the neoclouds can pledge the GPUs as collateral to get financing.
Yeah.
Right.
Yeah. Because right now it’s hard for the neoclouds or the new data centers—
—to get financing to build the data centers in advance of the demand—
—because the thing is, the hyperscalers are able to do it, like Google and Amazon—
Yeah, they have infinite funds. They can issue debt.
They have balance sheets so big that they can build ahead of the demand, whereas the smaller neoclouds and data centers can’t.
So how do they do that?
Well, they can use the Nvidia chips as collateral to borrow more money—
—to finance all that buildout.
Yeah.
So now the Nvidia chips—or at least Jensen hopes—the Nvidia chips are going to become a new asset class.
Yep.
Used for collateral. I mean, it’s in his best interest if it becomes one. Then he reigns supreme. Nvidia chips are trading publicly. No one can compete with that, because if you’re a new data center and you want to build ahead of demand, the only chips you can buy to do that are Nvidia chips, because no one else will give you that kind of financing.
Yeah. So people said Nvidia is now the Federal Reserve of AI.
40. Nvidia's Customers Building Their Own Chips
Pretty much. And I think he’s doing this for a few reasons. One is, I think OpenAI is getting too much power. They’re building their own chips, ASICs or whatever, to better align with the model.
I don’t think Nvidia is scared of OpenAI.
I mean, I don’t think they’re scared of anyone, to be quite frank—
—but they are scared of Google.
Meta. Yeah.
All of Nvidia’s major customers are building their own chips. OpenAI is very early, but TPUs have been around maybe for a decade.
Yeah.
And then—
Amazon developed them first.
Amazon Trainium.
Yeah, Trainium—
—is closely behind, and then Meta has been working on its own chips for a while.
Yeah.
Even Microsoft has its own chips.
They do.
Four of Nvidia’s top customers are trying to compete with Nvidia. But TPUs are not as robust as GPUs. I could be wrong about this, so correct me if I’m wrong, but what TPUs do really well is operate like ASICs, right? They’re more efficient for the models that they create for that chip.
Mhm.
Is that right? Is that correct?
It’s fast. It’s faster, so it’s more efficient for the model. It’s a model-specific chip.
Yeah.
Whereas GPUs are more—
Scalable. Scalable.
Yeah. So if you need to train a new model, you want to use Nvidia chips. And once you’ve trained the model—
No, no, no. Once you’ve trained the model, you’re locked in.
Oh, I mean, yeah. Yeah.
Because it’s more efficient to run inference on the same chips as the ones that you trained the models on.
Yeah.
Whereas TPUs are only for Gemini.
Yeah. And Jensen is aligning with all the neoclouds through either investing or financing. Nvidia is now a synthetic—
Yeah.
Someone says this, but they’re a synthetic hyperscaler.
Yeah.
To compete with their customers—
—but it’s probably the right thing to do.
Yeah, they have to do it. They have to do it. I mean, they acquired Hugging Face.
Yeah. Right.
Right. So they’re also going up the stack. They originally started with chips, and now they want to offer compute by acquiring these neoclouds and data centers—
—and then they go up the stack again by building their own open-source models: AI models, robotics models, self-driving models—
Yeah.
—and then they acquire Hugging Face, probably as a distribution channel—
Yeah.
—for their own models.
Yeah. It’s also interesting because I’ll tell you from a startup perspective. We talk to a lot of AI and robotics startups, and what we’re seeing on our side is that a lot of these startups are going open-source, and they want to run their own models. Instead of doing API inference via OpenAI or Anthropic, they want to rent capacity from data centers. What does that mean? They want to rent their own GPUs so that they own those GPUs, more or less, and run and train their own models on those GPUs, because it’s cheaper for them and the data that’s relevant for that startup won’t go anywhere else.
Yeah.
So from a startup perspective, we’re seeing this race for GPUs. Obviously, you had the neoclouds and all these companies buying GPUs. Now all these startups are buying their own GPUs or want to rent capacity, and they want to go open-source.
Yeah.
41. Microsoft's Closed-Loop Strategy
So I feel like there’s this effect of what Jensen’s doing, right, with the open-source push. Then you have all these startups that are going open-source. Then you have Microsoft doing the closed-loop model with enterprises. We talked about this in the car. How do you synthesize all of this?
Everyone’s doing what’s in their own best interest.
It makes sense, of course. Anthropic and OpenAI want to be end-to-end, offering expensive frontier models and gathering users’ data to improve their models over time.
Yeah.
Whereas for Nvidia, it’s in their best interest to offer open-source models.
Yeah.
Because all the open-source models use Nvidia chips—
Yeah.
—for training.
Yeah.
And then, of course, Microsoft wants to go against the OpenAI and Anthropic model—
By offering custom solutions to their enterprise customers. That’s what they’re good at.
Yeah.
And they also understand that their enterprise customers don’t want to leak their IP—
—to Anthropic and OpenAI.
So everyone’s doing what’s best for themselves.
Yeah.
So all of these things will probably coexist.
I mean, exciting new world.
Very, very competitive.
42. Chinese Open-Source Models and Regulatory Pushback
Yeah. Very competitive.
Yeah. I was just looking at OpenRouter.
Oh, that’s right. They got acquired for $7 billion, right?
Yeah. But I was just looking at the ranking of the top open-source models by usage. I think 9 of the top 10 were Chinese open-source models.
The only one in the top 10 was OpenAI or GPT Zuna.
I think Luna is the open-source one. And then Gemini—one of the older Gemini open-source models—was around 11. The rest of the top 20 were GLM and DeepSeek. So the Chinese AI companies are doing what’s in their best interest.
Yeah.
Trying to commoditize the entire model market.
Yeah. And then that’s why there’s probably a push for regulations around Chinese models, right?
Oh, that’s 100% going to happen.
So it’s already happening. I’ll leave it with this point: Chinese models are going around the aggregators. They don’t like OpenRouter, and they don’t like fal, which is another one. What they’re doing is going to the startups—the startups that I’m working with, right?
Mhm.
And they're saying, “Hey, we'll give you 50% off if you don't go there.”
Mm-hm.
Go direct to us.
Mm-hm.
I found that to be very interesting as well.
Of course, OpenRouter is charging a fee, right? So why—
Exactly.
Why give them that?
Exactly. They just become a commodity, is what I'm saying.
So, the power goes back to OpenRouter, right? So, yeah, the space is definitely moving very fast, but I'm really excited about it.
Yeah.
I think we're at time now. Any final words?
43. Imran and Qiao's Bags
What are our bags?
Right now, my bags are Zcash.
And that's it.
I mean, Bitcoin and all the boring stuff.
The boring stuff.
I'm starting to dip into on-chain.
Okay.
So, I'm looking at some stuff. Definitely on the Robinhood side, some of the stock-pair fund stuff.
Yeah.
Probably the top 3 are the ones that I'm looking at.
Cool.
I mean, this is just on-chain stuff.
Yeah.
I'm not talking about my NVIDIA bags.
Yeah, yeah, yeah.
For me, in crypto, it's just Zcash and Bitcoin.
Yeah, yeah, yeah.
And then, on the stock side, it's the 4 hyperscalers.
Meta, Google, Microsoft, and Amazon.
And then NVIDIA and TSMC—
Tencent.
Tencent.
Yeah, that's it.
Cool. Yeah, I have exposure to the top 4, Mag 7.
Yeah.
Cool. All right. Well, we'll catch you guys maybe in a few months. We'll see when we'll do it next.
Yeah.
All right. As always, the views expressed in the Good Game podcast are personal to the speakers and do not necessarily reflect the views of any other person or entity. Nothing here should be construed or relied upon as investment, legal, tax, or other advice. Thanks for listening to Good Game. Don't forget to subscribe. We'll see you next week.