The Worst Is Behind Us

Summary

• Crypto has bounced, but the desk stops short of calling a bottom. Yan says Bitcoin is back to $84.4K and that the majors are showing strength after a rough month; Jason’s tell is that the daily 10:30 a.m. seller may have run out of coins. A speaker sees Bitcoin at $120K or above next year, after the Saylor-liquidation fears faded.
• Kevin lays out the AI CapEx hurdle: roughly $4T through 2029 at a 12% return implies about $500B in free cash flow, or $1T of revenue at 50% gross margins. Another speaker argues there are plausible ways to reach the numbers: Anthropic grew from $5B in Q1 to $16B in Q3, while Nvidia is growing rapidly and trades at about 13x forward earnings. His conclusion: “the market is not pricing in AGI.”
• The near-term AI risk is reflexive demand, not simply debt. New labs trying to compete with Anthropic and OpenAI have raised $40B since 2024, and much of that money may go toward training and model use. A speaker who sold his neocloud holdings says they’re too hard to underwrite; the group expects possible overbuild and spending/revenue timing mismatches, not a perfectly smooth climb.
• Quantum is affecting allocations, but the risk remains hard to underwrite. Yan’s base case is that most holders will migrate or have coins they can migrate, and that a Chinese or U.S. lab would discover the capability. José says he sold some long-term Bitcoin for Zcash; Jason says he’s seriously reconsidering his own long-term Bitcoin position.
• Alternative stores of value have regained attention. Yan mentions Zcash, QANX and ORE; a speaker warns that Zcash could reach $5K and later fall below $1K. Pearl draws both interest for its compute-security thesis and skepticism over its roughly $3M in 24-hour volume.
• The HYPE ecosystem’s loyalty thesis is about giving holders somewhere to use their assets. Jason points to the button for moving funds from Hyperliquid to HyperEVM; Kinetiq’s proposed buybacks are split 50/50 between Kinetiq and HYPE. The discussion treats Elysium’s November 9 mainnet and subsequent user and TVL traction as signposts—not a guaranteed Kinetiq launch date or outcome.
• Unlocks are concentrated in projects the desk follows, not the whole industry. One speaker sees material weakness in Lighter as a potential buying opportunity, while Jason is watching for teams to cushion supply with buyers or restaking. The closing high-conviction list is Grass, Hyperliquid and probably Lighter; Venice is another possibility, pending an acquisition.
• The Year Ahead themes are money and better token designs. A speaker argues that value-accruing tokens can build loyalty, while crypto’s unlocks are further along than those in AI equities. Jason has no read on the midterms; another speaker sees a possible pre-event sell-off but doesn’t treat it as a major market factor.

Deep dive

1. “Imagine being a bear”: a bounce, not a confirmed bottom

• Yan says Bitcoin is back to $84.4K after a rough month. He says the speculative Robinhood/Solana “trenches” have been hammered, while some majors are showing strength. José objects to using “on-chain” as shorthand for the speculative end of the market.
• Jason had bids around $73K–$75K: Bitcoin could trade below $75K, $74K or $73K and still look fine in the bigger picture. A three-day rally came before those bids filled. His tell: “The 10:30 a.m. seller who’s been here every day for the last month, just nuking the charts, appears to have run out of coins, maybe.”
• On the Robinhood ecosystem-fund rumor, one speaker says even a small amount of business could move the needle, but he isn’t banking on it. He bought Pawns and Cash Cat after hearing the rumor.

2. Unlocks and midterms: risks people already see

• Jason is watching year-end unlocks in projects the group follows, including Lighter, where investors in private rounds are already well up. He wonders whether teams will try to pull unlocks forward or arrange restaking to get through the supply overhang.
• Another speaker cautions against treating all unlocks equally: the strongest, revenue-generating projects may be able to speak with investors and line up buyers. He says he’d consider material weakness in Lighter a buying opportunity. The reasoning is that widely anticipated events may not unfold as badly as the market expects—like people calling “Black Monday” on Sunday.
• Jason says he has no read on the midterms and won’t spend more time trying to game them. Another speaker notes that the incumbent party has often lost the House in midterms, and that a Democratic House could be a crypto headwind. But with crypto resilient despite failed legislation, trade concerns, AI worries and hot inflation, he sees the market climbing a wall of worry. He also notes that open interest in many alts has reset toward pre-move levels and that volume deltas show mostly perp selling—a healthier reset that favors concentration in the names most likely to become blue chips.

3. The $4 trillion AI CapEx question

• Kevin walks through a Venky-from-Menlo post: roughly $4T of AI CapEx through 2029, at a 12% return, would require about $500B of free cash flow by 2029. At 50% gross margins, that implies about $1T in revenue.
• One path is that the labs fall short, causing a painful unwind. Kevin notes that this is not necessarily a 2001-style debt crisis: the top four AI spenders account for about 70% of spending and fund most of it from operating cash flow, with strong credit ratings. The debt is more concentrated in Oracle, CoreWeave, Meta’s off-balance-sheet financing and private credit, which the post estimates at about $350B.
• Another path is that the revenues arrive. Kevin cites possible sources: 3% of roughly $15T in U.S. wages, or about $500B; roughly 1.6% of U.S. GDP; and a shift in global software spending toward AI compute. His takeaway is that there is a path, though the timing is uncertain.
• A participant flags a popular chart, also tweeted by Chamath, that says 1% of spenders account for 80% of spending, much of it AI and VC-backed. The caveat: “VC-backed” is a binary label; the share of the actual capital coming from venture versus credit matters, and both sources have risks.

4. The bull case: unprecedented growth and new demand

• One speaker argues that the labs need to grow about 70% a year from a roughly $100B base, which has never happened before—but points out that frontier-lab revenue was around $2B three years ago and is now about $100B. Anthropic grew from $5B in Q1 to $16B in Q3.
• Nvidia was up 106% year over year in the last quarter, with revenue this year projected to be 80% above last year on a roughly $200B base. Analysts had expected flat or lower results, while bullish analysts such as Altimeter had forecast about 20% growth. The stock was up only about 20%, at roughly 13x forward earnings. The speaker’s conclusion is that “the market is not pricing in AGI.” SpaceX was up 161% year over year and had a roughly $1.5B quarter.
• Demand indicators include spot GPU rental prices rising through October on the Oren Index; OpenRouter token spend up 36% in a month and 35x year over year, mostly on open models; and, according to SemiAnalysis, about $1T in binding data-center leases signed this year.
• A further source of demand is classical computing moving from CPUs to GPUs. Citing Jensen Huang on a podcast with Gerstner, the speaker says this is about a third of Nvidia’s revenue and growing 30%–40% a year. He also points to potential future demand from robotics, self-driving, surgical robots, medical and satellite imaging.

5. Reflexive VC demand and the difficulty of underwriting neoclouds

• A framework from Giovanni Cattani, whom the speakers say they are LPs in, divides AI tasks into bounded and unbounded. Bounded tasks—taxes, accounting or basic front ends—have limited token needs and may shift to cheaper models as they improve. Unbounded tasks, such as quant trading, AI R&D and building software, can absorb much more compute; the analogy is Formula 1, where users keep wanting the fastest car.
• The speaker says frontier revenue is mostly tied to unbounded tasks, but worries that much of the demand is VC-funded. New labs trying to compete with Anthropic and OpenAI have raised $40B since 2024, and much of that funding may go to training and to using existing models. If venture funding slows, demand could become reflexive.
• Another speaker notes that if Anthropic takes a material valuation markdown, that could affect the wider VC chain and tighten funding.
• One speaker says he sold his neocloud holdings and has them in the “too hard bucket” because of the businesses’ complexity and off-balance-sheet structures. He is less concerned about debt itself: the widely circulated 15% figure concerns unsecured debt, while secured borrowing is backed by Nvidia GPUs, including older generations that are reportedly selling for more than their original purchase price.

6. Overbuild, timing and the 2000 comparison

• One speaker uses a solar-system analogy: revenue may move steadily while spending oscillates around it. Spending and revenue won’t line up perfectly; periods of overspending can be offset by positive revenue surprises, such as a model breakthrough or new use case.
• Another speaker says an overbuild seems very likely. Jevons’ paradox could play out over a longer timeframe, with companies going bust in the meantime. The speaker distinguishes this from 2000: today’s companies have substantial revenue, though investors may not believe it will last. Nvidia and Meta trade below 20x forward earnings; Nvidia at 100x earnings would imply a roughly $40T valuation.
• Yan argues that bubbles usually have fewer bears and that buyers tend to keep capitulating into the rally; he jokes that Michael Burry would need to go full-port on CoreWeave before a top.
• The host emphasizes that revenues may be discounted because their sources are correlated: a problem affecting one part of the AI trade could affect the wider system. CoreWeave has said it would be massively cash-flow positive if it stopped building, though the host advises taking that claim with a grain of salt.
• The host also highlights the gap between spot GPU pricing and take-or-pay contracts. Lenders want longer commitments from creditworthy customers, while many startups may only be able to commit for a week, a few months or six months. If those shorter-term customers provide enough demand, or longer contracts roll over at higher spot prices, neocloud revenue could surprise positively. The host stresses that the compute-market mechanics are more complex than the headlines suggest.

7. Goldilocks for crypto—and a Bitcoin outlook

• One speaker’s Goldilocks case for crypto is that AI continues grinding higher with enough money to sustain it, but without repeating the biggest public-market outliers. He says speculative capital has shifted after the large moves in neoclouds, memory and other AI names.
• The same speaker sees a good setup for Bitcoin next year: many people sold, fears around Saylor’s potential liquidation eased, and Saylor is no longer seen as a major market force. He says Bitcoin could return to $120K or above.
• The caveat is that Bitcoin would likely struggle if the AI trade rolled over sharply and the frontier labs were badly damaged. A joking reply is that quantum won’t crack Bitcoin, so the group will be fine.

8. Quantum risk: migration, old coins and uncertainty

• Yan’s new base case is that by the time quantum risk becomes actionable, most holders will have migrated or will hold coins in unspent UTXOs that they can migrate. He thinks a Chinese or U.S. lab is most likely to achieve the capability, and says it may not matter which country gets Satoshi’s and other old coins.
• Yan says those coins are too large a supply to treat as permanently lost: Bitcoin should be priced with the assumption that all coins could become active. He allows that a government might sell some over time. He estimates a 90% chance—possibly as high as 99%—that the breakthrough comes from a Chinese or U.S. lab. If it were a Lazarus/North Korea doomsday scenario, he thinks people would fork their coins out.
• Another participant ranks “Red US” best, China in the middle and “Blue US” worst, arguing that the latter might use the coins to fund a short-term political agenda. Yan says he agrees with that ranking.
• Yan says quantum concern has helped renew interest in other stores of value, even though he doesn’t think the risk is disproven. If it does not happen for 10 years, people may start to think it will never happen—but until then, it can remain a headwind for Bitcoin.

9. Alternative stores of value—and real portfolio changes

• Yan says the industry had largely settled on Bitcoin as the store of value, with little room for alternatives, but that quantum discussions have changed that. He mentions Zcash, QANX and ORE on Solana, which he says is still small at around $50M.
• José says he sold some long-term Bitcoin holdings and put the money into Zcash, because he is less confident holding Bitcoin for a 10-year horizon if it cannot adapt to a cryptographic threat. Yan says his Zcash position changed partly because of how he expects other people to allocate, not because quantum made him personally afraid of Bitcoin.
• Jason says he has not sold his own long-term Bitcoin holdings but is seriously considering it: he cannot accurately underwrite the quantum risk and would not “mindlessly shill” Bitcoin to family without acknowledging that uncertainty.
• A participant recounts being asked about quantum risk on a date; another says his uncle, who owns Bitcoin and ETH for his grandchildren, asked him the same question. The discussion illustrates that the concern has reached beyond crypto specialists.
• One speaker warns that Zcash could reach $5K but later fall below $1K, as these store-of-value assets can be volatile.

10. Pearl: a small-volume trade and a larger thesis

• After Yan says people are discussing Pearl, another speaker calls it “a tool to identify who doesn’t have size,” pointing to about $3M in 24-hour volume and the aggressive promotion around it. The host notes that OTC brokers with miner connections are facilitating some trades. Yan is skeptical of buying ahead of a possible Coinbase listing, calling it a potential “classic top.”
• Another speaker says Pearl has a compelling narrative and that people outside crypto are asking about it, though some misunderstand it as a decentralized inference network. The initial thesis is that compute power could secure the network; if it succeeds, the supply of GPUs might also support decentralized inference.
• A speaker says he bought a bit but found it hard to buy, and wishes the asset were more liquid.

11. Paper Trade: transferability and attack risk

• The discussion asks whether making Paper Trade’s token transferable and listing perps would expose the protocol to an oracle-style attack. A participant says he did not see an obvious solution. Paper Trade’s thread gives a rough example: a $10M–$15M Bitcoin short paired with a $10M Paper perp short could be enough to threaten the protocol.
• Another speaker says exposure should be limited on assets that are cheap to manipulate, while noting that existing risk metrics already try to address manipulation. He cites an ETH exploit on the first night that made roughly $1.5M, with some uncertainty about the amount. A shortable native token adds another profit opportunity if an attack succeeds, but could also increase insolvency risk or force more restrictive risk settings.
• One participant compares the problem to stablecoins, where shorting can be attractive because the worst case is a return to $1; here, the risk is built into the mechanism.
• Another speaker explains that Rollbit hedged positions by identifying sharp users and hedging winners, which is harder when users can create wallets freely. Keeping the book more delta-neutral can allow more open interest, but might constrain scaling. Possible risk controls include tranching or automatically closing positions far into profit; these could reduce usability. The speaker thinks it may be solvable with risk controls, but the effect on UX is uncertain.
• A participant says he had owned some PaperStrategy. He says the strategy controls about 2% of Paper’s supply, stakes it and has buybacks that currently outpace inflation. If the issue blows over in a couple of months, he thinks the situation could turn out well for the strategy. Another speaker jokes about the “farm the Hyperliquid airdrop by getting liquidated” strategy.

12. Kinetiq, HYPE and ecosystem loyalty

• Jason points to the button that lets users move funds from Hyperliquid to HyperEVM: there are billions of dollars on Hyperliquid, and some users may want to transfer funds and explore on-chain. José says HyperEVM can offer a more performant experience, and that app support could bring more users into its ecosystem.
• Jason frames the loyalty argument: people tend to spend time in ecosystems where they made money. A speaker calls this “patrons of the chain” and argues that the best way to build loyalty is to make people rich on a protocol. Hyperliquid holders have seen large gains and may welcome more ways to use the ecosystem.
• Kinetiq’s buybacks are split 50/50 between Kinetiq and HYPE. José says that is more aligned than the Ethereum rollups he is comparing it with, which paid ETH data-availability costs. A participant estimates roughly $250K a year in HYPE buybacks based on testnet activity; others caution that testnet activity is not the same as a live product. Solana is cited as an example of an ecosystem where users did want to speculate in native-chain assets.
• A participant who says he bought Kinetiq also cautions that “first alt on a chain” theses sometimes fail. The host asks about Elysium’s November 9 mainnet; Jason says the useful signposts are actual users and TVL, and that signs of traction should become visible within 30–60 days. Hypercall is said to be launching there on day one.
• José criticizes the idea that many rollups offer meaningful technical differentiation, citing an L2BEAT founder’s “zero-proof system”/Gnosis Safe troll and the argument that many rollups remain Stage 0. Another speaker praises Solana’s Alpenglow update and its developer and user community, while saying he is unsure how SOL the asset will perform.

13. Year Ahead themes: money and token design

• A speaker says the two big themes for the Year Ahead are money and protocols with better token designs. Hyperliquid showed that tokens designed to return value to holders can attract interest. The speaker expects more thoughtful token designs and argues that launching a token avoids some of the regulatory concerns associated with equity.
• The speaker is bullish on tokens being “unfucked,” noting that many crypto unlocks are behind the industry while AI-equity unlocks are starting. He also points to improving breadth, with Zcash, Fennec, hooks activity and Solana among the examples, after Paper Trade took about $140M of liquidity out of the market.

14. Pump, social trading and the closing conviction list

• Participants debate Pump’s valuation and revenue. One says he sold the rest of his position but keeps changing his mind about whether it could be worth $2B or $10B. Kevin says the Pump revenue chart befuddles him; Jason agrees. They discuss Robinhood activity, Pump’s relatively stable metrics during that period, and whether Pump’s mobile-app revenue may now exceed FOMO’s.
• The group discusses a video of traders describing how they created a meme coin, bought 50% of its supply and made money as its market cap rose. Jason calls the behavior social trading that incentivizes bad conduct. One participant argues strong KYC and one-account limits could help make social trading more organic; Jason says KYC can help, while noting that people may still find ways around it.
• The host recommends Jason’s speculation report, highlighting its “antisocial trading” framing: if people are going to participate, owning the house may be the better strategy.
• In the closing token discussion, one speaker names Grass, Hyperliquid at its current valuation and probably Lighter as three tokens he thinks could be higher a year from now. He thinks options may become relevant for Lighter, with binary products first and event contracts a year out—not three months out—and says to revisit in October 2027. Venice is another possibility, pending an acquisition.