Delphi Digital’s Hivemind argues that Anthropic’s IPO, with insider shares unlocking fast, could pull money out of an already risk-off crypto market. In this 86-minute episode, published November 19, 2026, hosts Kevin Kelly, Jason Pagoulatos, José Maria Macedo, Yan Liberman and Ceteris also cover the weekend washout that took Bitcoin to the mid-60s, Jensen Huang’s exit from Nvidia, why AI-era hacks make crypto a honeypot, Bitcoin’s security budget, Leopold Aschenbrenner’s fund and the one asset each would buy for the next 18 months.
TLDW
Pagoulatos says Delphi’s macro regime model flipped to risk-off at the end of October and has only worsened, with real yields heading higher and talk of a 50 basis point hike, so he is mostly in cash and HYPE. Ceteris is still buying the index, and Liberman says crypto cannot bottom while equities fall. Macedo wants his money where it can be frozen while AI-assisted hacks hit crypto, and argues subscale chains should shut down. On Anthropic, he says the unlock schedule is more aggressive than any low-float, high-FDV token. The panel splits on Nvidia, where Kelly sees value at about four times 2028 sales and others say the business depends too much on Huang to bet on without him.
Thoughts
Crypto investors are the right people to price Anthropic’s IPO, because they have seen this structure before. Macedo’s chart of the unlock schedule, which he says is more aggressive than SpaceX’s and any low-float token, is the episode’s best contribution. The detail matters: Anthropic’s November filing does set a 180-day lockup on paper, but reports on it say non-founder employees can sell 10% from day one and more after the fiscal 2026 results, while other holders can sell 6% to 7% tranches on days 27, 48, 62 and 76. The hosts make the human version of the point: employees who went from a salary to nine figures, with Anthropic as 90% of their net worth, should sell, and will. The difference from a crypto unlock is that there is a real business behind the supply, but the flow mechanics are the same, and in a risk-off tape, new supply competes with crypto for the same marginal dollar.
The Nvidia debate shows why a cheap stock can be a bad trade. Kelly’s case is the numbers: Nvidia beat revenue estimates by about 6% on $115 billion, raised guidance to about 75% growth next year, and trades near four times his estimate of 2028 sales. The counterargument, made by another host who holds the stock and shorted some on Hyperliquid after the news, is that Nvidia’s supply chain runs on Huang personally. He says Huang talked TSMC and memory makers into raising supply when they did not want to, gets special pricing, and has about 60 direct reports. Another host adds that competition is arriving just when a founder matters most. They end up describing an asymmetric bet: perhaps 20% upside if the investigations clear him, 30% downside if they do not. That is why “I’m not touching it” is a reasonable answer even from people who think it is undervalued.
Macedo’s line that he wants his money “as much as possible in places where it can be frozen” is a striking thing for a crypto investor to say, and he means it. Decentralization trades censorship and intervention risk for smart contract risk, and when AI makes exploits cheap, the second risk dominates. So he wants vanilla ETH and Bitcoin, Tether because it freezes stolen funds, and centralized exchanges that make users whole, and no exposure to any team without the scale to defend itself. Later in the episode a host goes further: subscale chains and L2s that have not found traction should shut down, because running infrastructure with no users is “return free risk.” The panel expects crypto to emerge in six to twelve months as one of the most hardened places to keep money. In the meantime, the asset list shrinks toward whatever has the least code and the most recourse.
The Leopold Aschenbrenner discussion is about the gap between a correct thesis and good execution. The hosts admire Situational Awareness, one calling it a life-changing read on par with the Bitcoin white paper, and agree its early returns came from a view nobody on Wall Street shared. But the fund later took in money near the top and bought calls expiring in three to six months, which a host calls hard to square with a decades-long thesis, and illiquid positions made redemptions reflexive. One host’s conclusion is the useful one for allocators: the edge is in choosing when to invest with a manager like this, since he will do best in markets that reward aggression, rather than trusting him through turbulence.
Pagoulatos’s macro read is the frame for everything else, and it is gloomier than the price action. His point is not that crypto is oversold but that the regime has changed: a cross-asset model that took Bitcoin to zero allocation at the end of October usually stays there for months, and this looks to him more like 2021 than 2018, with inflation above target, real yields rising and growth slowing. Ceteris’s answer, buying the Nasdaq after a roughly 15% drawdown, is the right counterweight, since a regime model is a guide to position size, not a timing signal. The live market made the point for both of them, retracing the day’s losses during the recording.
Key Takeaways
- Kelly lists the damage: weekend volumes two to three times 2026 averages, about $2.5 billion of ETF outflows in two weeks, a negative Coinbase premium, and stablecoin supply shrinking at its fastest pace since June 2022.
- Delphi’s macro regime model, which Pagoulatos says will come to the Delphi portal, flipped to risk-off at the end of October, and he expects three to six months before conditions recover.
- Liberman says crypto cannot find its own bottom while equities fall, and that perp selling ran seven to ten times spot selling in the run-up to the weekend washout.
- The hosts think Huang could have survived the spy allegations or the affair revelations, but not both, and that new founders will all now insist on super-voting shares.
- Macedo is still long AI for the long term, says valuations look fair to cheap on earnings and revenue, and is buying gradually rather than trading macro.
- The hosts argue that most outages are not AI hacks, but that the public now blames AI for every one, calling it “internet COVID.”
- Store-of-value crypto is being bid partly because it has a smaller attack surface, with Zcash cited for its formally verified pool after the Orchard bug.
- On Bitcoin’s shrinking security budget, one host expects it to drift toward a de facto proof of authority run by exchanges and custodians, and another puts the odds of Bitcoin being flipped by 2030 above 10%.
- The hosts want rate limits on bridges, including Hyperliquid’s, and centralized security monitoring across on-chain projects for the next year.
- Anthropic perps on Hyperliquid were trading well above the IPO price, with one wallet $100 million short, according to the hosts.
- Pagoulatos’s cash is about 35% of his portfolio, and one host would start buying HYPE in the low 60s and add more in the 50s and 40s.
- For one asset over 18 months, picks included an MSCI World index fund, Meta, SpaceX, Grass, Hypercall and Nvidia.
Chapters
0:43 Are We Near a Market Bottom?
Kelly runs through two painful weeks: the weekend crash, ETF outflows, whales sending coins to exchanges, miners selling reserves to fund AI data centers, real yields at cycle highs and a stronger dollar. Pagoulatos says his model has been degrading for months and sees no bottom into year end. Ceteris is fine buying indexes with the Nasdaq down about 15% and the S&P down less than 10%.
10:45 Was the Weekend Washout the Bottom?
Liberman says Bitcoin has retraced its whole move since August and sits near major support, but the systematic, price-insensitive weekend selling and a liquidation cluster below suggest the low 60s. A proper bottom usually needs a high-volume flush followed by a low-volume retest, which he has not seen.
17:28 Jensen’s Exit and the AI Trade
The hosts discuss whether Huang can return, with a prediction market at 34% by the end of 2027 on thin volume. They debate whether his China policy made sense, noting Beijing banned H20 imports to protect its own chip industry, and speculate about the timing of the leaks while acknowledging they have little information. Macedo says nothing in his AI positions has changed.
29:23 The AI Cyberattack Wake-Up Call
One host says the hacks have crossed into mainstream conversation, even if many outages are not AI-driven. Macedo says people he knows at AI labs put high odds on current encryption being broken within two years, and that crypto will take the brunt of the transition before becoming one of the most hardened places to hold money.
38:17 Dip-Buying Opportunity or Trap?
A host argues Zcash, at about $20 billion with a roughly $5 billion shielded pool, is still an experiment rather than a store of value, and flags Pearl as a risk to it. The CFTC’s first license went to Backpack rather than a Hyperliquid deployer or Lighter, and the hosts discuss Grass, Hypercall and on-chain equities.
44:43 Bitcoin’s Security Budget and Whether It Gets Flipped
With the halving approaching and fees not filling the gap, the hosts say Bitcoin’s ethos will have to change through the fee budget or quantum resistance. They disagree on whether anything flips it, and one host explains why ETH would need roughly a sixfold move against Bitcoin to do it.
49:44 Why Chains Should Go Bunker Mode
The hosts argue over whether recent exploits were social engineering or genuinely sophisticated, citing a fake Homebrew install, front-end attacks on about 20 websites, and an attack on a Delphi portfolio company that reached hyperscaler infrastructure. They agree that exchanges and custodians will win users who want to swap from a safer sandbox.
56:12 Anthropic’s IPO and Its Unlocks
Macedo presents his unlock chart, and the hosts expect heavy selling from newly rich employees, though one prefers fast unlocks to a four-year drip. Pagoulatos says the timing is perfect for the company but not for the market.
1:01:57 Would You Back Leopold Again?
The hosts debate whether Aschenbrenner’s early alpha was skill or a one-time non-consensus view, comparing him to Multicoin’s 90% drawdown and later recovery and to launching a crypto fund in 2022. They note 13F filings overstate recent drawdowns by measuring from all-time highs.
1:12:04 One Asset for 18 Months, and Is Nvidia a Buy?
Each host names a pick, and Kelly makes the case for Nvidia. The others answer with its size, new competition and Huang’s role in supply deals. The episode ends as markets bounce during the recording.
Notable Quotes
“I don’t have any optimism to give you.”
Jason Pagoulatos, on whether crypto has bottomed
“Trying to find unique reasons for crypto to bottom is kind of impossible because gravity will do its thing.”
Yan Liberman, on crypto’s link to equities
“Geopolitical analysts have only ever cost me money.”
José Maria Macedo, on staying long AI
“I want my money to be as much as possible in places where it can be frozen.”
José Maria Macedo, on AI-era crypto hacks
“The unlock schedule is like significantly more aggressive than any of the crypto low float, high FDV.”
José Maria Macedo, on Anthropic’s IPO
“The IPO is not for the market. It’s for the company and the, you know, raising money and tapping capital markets.”
Jason Pagoulatos, on Anthropic’s timing
“They beat top line revenue by 6% on $115 billion in revenue.”
Kevin Kelly, on why he would buy Nvidia
Watch the full conversation here.
Related Reading
- Anthropic IPO lockup explained (RoboForex) the staged early-release schedule from Anthropic’s filing.
- Situational Awareness Leopold Aschenbrenner’s essay series behind his fund’s thesis.
- Lock-up period (Wikipedia) how IPO lockups normally work.
- Delphi Digital the research firm behind the Hivemind podcast.