CoreWeave Crash, AI Trade Cracks, Anthropic IPO Red Flags, Bill Gates Doomer

The "All-In" Podcast Kicks Off [00:00:00]

• The hosts celebrate the "All-In Podcast" reaching the number one spot globally.
• David Sacks, Chamath Palihapitiya, and Ben Horowitz banter about who deserves credit for the podcast's success.
• Ben Horowitz joins the discussion, acknowledging the podcast's number one status.

  "It's great being the best."

AI Bubble Concerns Emerge [00:02:30]

• The panel discusses a potential pop or deflation in the AI bubble, citing CoreWeave's stock drop.
• CoreWeave's bonds sold off, raising concerns about its ability to support its debt.
• Elon Musk's "yikes" tweet amplified the negative sentiment.

  "yikes y-i-k-e-s"

CoreWeave's Financial Structure: The "Box" [00:05:00]

• Michael, founder of CoreWeave, previously explained his "box" financing model on the podcast.
• This model involves using contracts with clients like Microsoft to secure financing for GPUs and data centers.
• The "box" governs cash flow, prioritizing data center, power, and interest payments before returning profits.

  "The economics in this box are such that within two and a half years of a five-year deal we have paid for everything."

Market Signals Worry for CoreWeave [00:10:00]

• Chamath points to the spike in credit default swaps for CoreWeave as a sign of imputed default risk.
• The insurance market suggested a significant chance of default, indicating smart money's concern.
• This spike in perceived risk led to a sell-off in CoreWeave's bonds and stock.

  "The insurance market was saying about their debt and I think that's something to look into because those guys I think are the smartest money in in all of this."

Rising Risk-Free Rates Impact Debt Costs [00:15:00]

• The US is raising risk-free rates, impacting the cost of all debt, including junk bonds.
• CoreWeave's financing costs have also increased significantly due to this broader market trend.
• The company needs to raise substantial new debt to finance its ongoing operations and contracts.

  "As that goes up basic way to think about that is as the 10-year bond goes up everything else goes up."

CoreWeave's Revenue vs. Debt Obligations [00:20:00]

• CoreWeave has fixed-price contracts, meaning they don't benefit from rising spot market prices.
• They need to raise an additional $35 billion in debt to finance their existing contracts.
• The average interest rate they pay is around 8.5%, but future borrowing costs are increasing.

  "So he has about 35 billion of debt he needs to raise another 35 billion of debt but as you saw in the chart before that's ticking up."

The "Coin Flip" Scenario for CoreWeave [00:25:00]

• Chamath describes CoreWeave's situation as a "coin flip" if financing costs continue to rise.
• If financing rates reach 15%, they could struggle to cover obligations, potentially being only $5 billion from breakeven.
• Customer concentration (75% revenue from three companies) adds to the risk.

  "It's basically a coin flip now on the following idea."

Ben Horowitz on Neo-Cloud Complexity and Risk [00:30:00]

• Ben Horowitz describes companies like CoreWeave as "Rube Goldberg financial machines" due to their complexity.
• He notes that GPU lifespans are increasing, which is a tailwind, but the overall financial structure is concerning.
• Horowitz's firm avoids investing in neo-clouds due to the potential for getting "upside down" easily.

  "So you know we talked about okay gpu contracts in this but he's got data center contracts how long are those and and then there's he's got to buy memory like and there's memory shortage and this and in that so it is really complex."

Lessons from the Internet Bubble [00:35:00]

• Horowitz contrasts the current AI demand with the internet bubble's lack of demand.
• During the internet bubble, excessive debt combined with insufficient consumer adoption led to failures.
• While AI has massive demand, a "pathological debt situation" could still be ruinous.

  "we don't really have a lack of demand in ai so i think the dynamics will be different."

The Power of Supply Shortages [00:40:00]

• Despite financial complexities, AI demand is strong, and GPU supply remains constrained.
• Data center space and even labor for construction are becoming more expensive.
• This supply-demand dynamic provides a tailwind for companies like CoreWeave, independent of market financing.

  "And so those those are very big tailwinds for core weave despite the fact that the debt markets kind of independent of them are tough."

Poly Market on CoreWeave Bankruptcy [00:45:00]

• Sacks dismisses a poly market prediction of CoreWeave bankruptcy as unlikely given the asset-backed nature of their financing.
• He explains CoreWeave's "opco propco" structure means individual asset vehicles might face issues, but not the entire company.

  "So core weave can't go bankrupt an asset level vehicle for a data center or a build out could theoretically go upside down but that'll go into receivership and then just manage where you sell down the gpus."

AI Trade: Bubble or Volatility? [00:50:00]

• Sacks argues the recent market swings are volatility, not necessarily the start of a bubble popping.
• He attributes this to the market's sensitivity to any data points when underwriting multi-trillion dollar future outcomes.
• Demand for compute remains off the charts, with companies even VCs seeking compute access.

  "I would make the argument that this is just volatility which you're going to get a lot of whenever you're underwriting to multi-trillion dollar outcomes in the future."

Supply Constraints and Execution Risks [00:55:00]

• There's no glut of AI compute supply due to the difficulty of building and powering data centers.
• Regulatory headwinds also slow down the expansion of new compute capacity.
• While CoreWeave's execution might be slightly slower or more costly, they are considered excellent at standing up compute.

  "So if anything we believe that there's not a glut it's the other way that there's enforced scarcity."

Credit Risk and Repricing [01:00:00]

• Financing for CoreWeave's projects is backed by tenant credit (e.g., Microsoft), not just CoreWeave's unsecured debt.
• Lenders are secured by the GPUs themselves, which retain value.
• The market repricing is primarily of CoreWeave's unsecured obligations, increasing their implied coupon, but not necessarily their current borrowing cost.

  "The market has discounted them by 20 percent they're in distress territory that has brought the implied coupon of those bonds from 9.6 percent to around 15 percent."

The Rise of Bespoke Software and Deflationary AI [01:05:00]

• The commodification of AI is leading to deflationary pressures, with open-source models rapidly reducing costs.
• This enables enterprises to create bespoke software for specific use cases, driving productivity.
• Demand for coders is increasing, as they manage and maintain the AI-generated code.

  "The value of these coding assistants is that now you can create software for highly bespoke use cases whereas software used to be created really just for the mass market."

Anthropic's IPO and Valuation Debate [01:10:00]

• Anthropic's investor day revealed strong revenue figures ($16.8 billion in Q3) but hand-wringing over ARR estimates.
• Palihapitiya suggests Anthropic should price its IPO to raise a significant war chest ($200-$250 billion) rather than optimize for valuation.
• Horowitz agrees that financing risk is a major threat and a large cash reserve is crucial for survival.

  "I think their products are exceptionally good I think they're doing the hard work and you know to open ai's credit they are as well of walking down the price curve so that they're less and less at risk of open source."

The "Compute Duopoly" and AI's Future [01:15:00]

• Anthropic and OpenAI are estimated to consume 60% of global compute capacity in the next year, forming a "compute duopoly."
• This duopoly extends to revenue and frontier model capabilities, though competitors are closing the gap.
• The panel discusses Sam Altman's lack of equity in OpenAI and the need for a "make right" settlement.

  "According to semi-analysis anthropic and open ai are taking 60 of all the compute that's expected to be stood up in the entire world over the next year."

Anthropic's "AI Worm" Psyop Allegations [01:20:00]

• Sacks and Horowitz accuse Anthropic of fear-mongering with a blog post about AI-powered worms infecting IoT devices.
• They argue this is a tactic to stifle open-source competition, particularly small language models (SLMs).
• Horowitz points out Anthropic's own security lapses and hypocrisy regarding distillation attacks.

  "they're basically trying to make it sound like these small models are dangerous they can take over your toaster or your refrigerator or something like that it's almost comical."

The Real Danger: Misalignment and "Frankenstein" Models [01:25:00]

• The panel argues the true danger lies in Anthropic's training of frontier models to act as independent moral agents.
• Research suggests Anthropic's models increasingly assert moral agency and refuse human instructions.
• This self-induced misalignment is seen as a far greater risk than theoretical AI worms.

  "I remain actually very concerned about this frankenstein problem you know that anthropic continues to train its frontier models to again be its own moral actor to think that it can be right when humans are wrong."

Bill Gates and Doomerism in AI [01:30:00]

• The discussion turns to Bill Gates's pronouncements on AI risks, which the panel views as self-serving.
• They suggest Gates seeks relevance and a return to a position of global influence, similar to his role during the pandemic.
• The panel criticizes the media's amplification of these doomer narratives.

  "He wants to be relevant I mean that should he just like I think he wants he absolutely wants to be relevant."

The Importance of Continuous Building [01:35:00]

• Sacks advises against seeking public validation like Pulitzer Prizes, emphasizing the importance of continuous work and building.
• He contrasts this with individuals who, after quitting their jobs, may spiral into seeking external validation.
• The panel concludes by reiterating the podcast's number one status and signing off.

  "The lesson here folks is don't quit your job just keep working keep building shit."