AI infrastructure faces its first serious stress test

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

The episode examines whether falling AI-infrastructure stocks expose fragile financing or merely reflect volatility in a still-constrained market.

3 key takeaways
  1. 1Debt-heavy neoclouds remain vulnerable to fixed-price contracts, power constraints, GPU cycles, and a downturn in demand.
  2. 2Anthropic and OpenAI face enormous capital needs as their valuations increasingly function as strategic hedges for investors.
  3. 3The panel argues that electricity, infrastructure access, and specialized data may outlast frontier models as AI’s strongest moats.

Don't miss

The panel argues that electricity, not models, could become the decisive AI moat as power and infrastructure constrain expansion.

The brief

Ben Horowitz joins the panel as CoreWeave’s falling stock prompts a larger question: does debt-financed AI infrastructure reveal a bubble, or ordinary volatility in a constrained market?

The group treats neoclouds as financial machines exposed to GPU prices, power, memory, and fixed contracts, recalling how falling bandwidth prices damaged LoudCloud during the dot-com crash.

Anthropic and OpenAI become the next test: their valuations reflect extraordinary growth and strategic importance, but also a need for vast new capital before any public offering.

The panel debates whether AI will commoditize software or expand it, with new moats potentially shifting toward data, users, feedback loops, vertical expertise, and electricity.

The closing argument turns against AI-safety doomerism, challenging Anthropic’s worm warnings and Bill Gates’s catastrophe rhetoric as narratives shaped by incentives and influence.

What was said on this episode

41 statements · 19 positive · 19 negative · 3 neutral

  1. CoreWeave debt implies approximately a 55% five-year default probability

    “over a five-year duration, what that math implies is about a 55% chance of a default”

    Listen at 6:55

  2. Chamath Palihapitiyaon CoreWeave customer concentrationNegative11:16

    CoreWeave receives 75% of revenue from three customers

    “they have extreme customer concentration. 75% of the revenue comes from three companies”

    Listen at 11:16

  3. Chamath Palihapitiyaon CoreWeave breakevenNegative12:16

    CoreWeave is approximately $5 billion from breakeven

    “My team's estimates is that they're plus or minus about 5 billion away from that breakeven point”

    Listen at 12:16

  4. Ben Horowitzon Bandwidth pricesNegative17:24

    Bandwidth prices fell from $1,500 to $50 per megabit by 2001

    “By 2001, it was $50 a megabit.”

    Listen at 17:24

  5. CoreWeave cannot become bankrupt under its opco-propco structure

    “CoreWeave can't go bankrupt”

    Listen at 18:58

  6. David Oliver Sackson CoreWeave declineNeutral20:30

    The CoreWeave decline is ordinary market volatility

    “I would make the argument that this is just volatility”

    Listen at 20:30

  7. David Oliver Sackson CoreWeave fundamentalsPositive20:56

    CoreWeave fundamentals have not materially changed

    “I don't think that something fundamental has changed.”

    Listen at 20:56

  8. David Oliver Sackson Compute demandPositive21:05

    Demand for compute continues growing rapidly

    “We know that the demand for compute just keeps growing by leaps and bounds.”

    Listen at 21:05

  9. David Oliver Sackson H100 and A100 GPUsPositive21:40

    Older H100 and A100 GPUs still rent at premium prices

    “old GPUs, like even the H100s or the A100s, which are several generations old, they're still renting at premium prices right now”

    Listen at 21:40

  10. David Oliver Sackson Powered compute supplyNegative22:50

    Powered compute supply is constrained rather than excessive

    “there's not a glut, it's the other way, that there's enforced scarcity”

    Listen at 22:50

  11. Ben Horowitzon Anthropic and OpenAI IPOsPositive27:30

    Anthropic and OpenAI IPOs will perform adequately

    “My belief is those are going to be fine.”

    Listen at 27:30

  12. Ben Horowitzon OpenAI and AnthropicNegative27:59

    OpenAI and Anthropic threaten most existing portfolio holdings

    “OpenAI and Anthropic are threats to that.”

    Listen at 27:59

  13. Chamath Palihapitiyaon Financing-spread business modelsNegative29:20

    Higher long-term rates damage financing-spread business models

    “Any business model that thrives on a financing spread gets cornholed.”

    Listen at 29:20

  14. David Oliver Sackson Bespoke softwarePositive32:35

    Bespoke software will substantially increase economy-wide productivity

    “That is going to be a huge driver of productivity throughout the economy.”

    Listen at 32:35

  15. David Oliver Sackson Software codersPositive34:30

    Demand for software coders is increasing despite coding assistants

    “demand for coders is increasing”

    Listen at 34:30

  16. Chamath Palihapitiyaon Electricity access for AIPositive38:21

    Electricity access will become the decisive AI competitive moat

    “I think power will be the moat.”

    Listen at 38:21

  17. Chamath Palihapitiyaon United States energy supplyNegative38:40

    The United States will face a 1.4-terawatt-hour energy shortfall by 2050

    “by 2050, about 1.4 terawatt hours short”

    Listen at 38:40

  18. David Oliver Sackson Neocloud economicsPositive41:21

    Payback period is the key metric for evaluating neocloud economics

    “payback period is really critical”

    Listen at 41:21

  19. David Oliver Sackson Neocloud companiesPositive41:22

    Short neocloud payback periods substantially reduce business risk

    “as long as the payback periods are really short, I don't see how these companies can lose”

    Listen at 41:22

  20. Chamath Palihapitiyaon CoreWeave contractsPositive43:21

    CoreWeave should cancel some fixed contracts and sell compute spot

    “I think that they have to cancel some of these contracts and move to spot.”

    Listen at 43:21

  21. Chamath Palihapitiyaon Spot compute pricesPositive45:20

    Spot compute prices will remain high for some time

    “spot will remain highly elevated for a while”

    Listen at 45:20

  22. Chamath Palihapitiyaon Startup access to electricityNegative45:34

    Startups without electricity access cannot rely on hyperscalers

    “if you don't have access to electricity, you cannot go through the hyperscalers”

    Listen at 45:34

  23. Ben Horowitzon Anthropic IPOPositive50:45

    Anthropic's IPO will perform adequately despite growth concerns

    “I think the IPO will be fine”

    Listen at 50:45

  24. Chamath Palihapitiyaon AI token economicsNegative54:26

    Token economics will undergo substantial cannibalization

    “I still think that there's a bunch of cannibalization that has to happen on the token side.”

    Listen at 54:26

  25. Chamath Palihapitiyaon Anthropic financingPositive55:27

    Anthropic needs to raise at least $200 billion

    “They need, in my estimate, at least $200 billion.”

    Listen at 55:27

  26. David Oliver Sackson Anthropic and OpenAI compute demandNegative59:51

    Anthropic and OpenAI will consume 60% of newly deployed global compute next year

    “Anthropic and OpenAI are taking 60% of all the compute that's expected to be stood up in the entire world over the next year.”

    Listen at 59:51

  27. David Oliver Sackson Anthropic and OpenAINeutral1:00:27

    Anthropic and OpenAI hold a revenue and market-share duopoly

    “they definitely have a duopoly when it comes to revenue and market share”

    Listen at 1:00:27

  28. Anthropic and OpenAI currently lead frontier-model capability

    “they still, I think, are in the lead in terms of capability”

    Listen at 1:00:44

  29. David Oliver Sackson OpenAI governance transitionNegative1:01:36

    OpenAI should have normalized ownership when becoming for-profit

    “the company, when they went from being a nonprofit to a for-profit, they should have done a make right”

    Listen at 1:01:36

  30. Small AI models are safer than larger models

    “small models are the safest models”

    Listen at 1:05:56

  31. David Oliver Sackson Anthropic distillation defensesPositive1:08:57

    Anthropic built technical defenses that stopped model distillation

    “Anthropic had to actually build. the technical capabilities to detect when distillation was happening and they've stopped it.”

    Listen at 1:08:57

  32. David Oliver Sackson Coinbase fraud-detection modelPositive1:15:20

    Coinbase's small model cut fraud by about 30% versus Opus 4.5

    “it cut fraudulent transactions by about 30% while being faster and much cheaper than Opus 4.5”

    Listen at 1:15:20

  33. David Oliver Sackson Enterprise AI workloadsPositive1:16:20

    Enterprise workloads will increasingly use customized small models

    “every workload is going to have its own small, customized model”

    Listen at 1:16:20

  34. David Oliver Sackson Anthropic frontier modelsNegative1:20:34

    Anthropic trains frontier models to act as independent moral agents

    “Anthropa continues to train its frontier models. to again, be its own moral actor”

    Listen at 1:20:34

  35. David Oliver Sackson Anthropic model trainingNegative1:21:43

    Anthropic's model-training practices create the risks it warns about

    “they are creating the very risks they keep warning us about”

    Listen at 1:21:43

  36. Ben Horowitzon Natural pandemic occurrenceNeutral1:23:04

    No natural pandemic has occurred yet

    “there has not yet been a natural pandemic”

    Listen at 1:23:04

  37. Ben Horowitzon Anthropic AI developmentNegative1:23:35

    Anthropic's AI development practices are currently AI's biggest danger

    “the biggest danger with AI right now is what they themselves are creating”

    Listen at 1:23:35

  38. Chamath Palihapitiyaon Bill Gates AI doomerismNegative1:24:33

    Bill Gates seeks renewed global influence through AI doomerism

    “he's the doomer. He's the flag-carrying doomer trying to essentially have what he had the first time around, which is command and control from a global hierarchy”

    Listen at 1:24:33

  39. Ben Horowitzon Bill Gates AI warningsNegative1:27:10

    Bill Gates's AI doomerism is extremely dangerous

    “I think it's extremely dangerous”

    Listen at 1:27:10

  40. Conditions for recursive AI takeoff have not been met or approached

    “Zero of those four have been met. We're not on track for any of those four to be met”

    Listen at 1:28:33

  41. Ben Horowitzon Recursive AI takeoffNegative1:28:43

    Recursive AI takeoff remains entirely theoretical

    “we're really talking about a risk that's entirely theoretical at this point”

    Listen at 1:28:43

Statements are attributed to the speaker as said on the episode and reflect their view at the time, not PodLume's. They are not advice.

Weekly email

State of the Pod, every Monday

The week's most-discussed names and stories from podcasts, with the moments behind them. Free, five minutes, unsubscribe in one click.

Listen to the full episode and explore every guest, topic, and moment on PodLume.

AI infrastructure faces its first serious stress test · PodLume