CharlieKey takeaways from your favorite podcasts
← 28 October 2026TBPN

Gavin Baker on Anthropic's S-1, GPU Prices, and SpaceX

Gavin Baker argues Anthropic's S-1 was an upside surprise with diversified customers, while compute demand is so tight that even forward prices are rising, rewarding competent GPU operators and flexible pricing.

Watch on YouTube ↗

Charlie's summary

  1. Anthropic's S-1 was an upside surprise, with a more diversified customer base than critics claimed

    Gavin says the S-1 showed only $10 billion burned, plus encouraging cash generation and unit economics. He was struck that about half the Fortune 500 use Anthropic yet it is only “0.1% of their expenses”, which suggests enterprise diffusion is very early. Against claims of concentrated spend, he points to 250 accounts spending over $10 million and 11 or 12,000 over $100,000.

    ▶ 0:28
  2. Open-weight inference clouds are growing faster than OpenAI or Anthropic

    Gavin says Fireworks, Baseten, Together, Modal and Cognition are at scale, open-weight first, and growing faster than either lab. His firm led Fireworks' last round and found it hard to price because revenue grew so fast that the multiple was “effectively going down” every day. He thinks this is far less known on the East Coast.

    ▶ 5:03
  3. The compute market is in contango: GPUs cost more to reserve for the future than today

    Gavin says October rental prices for essentially every GPU rose daily. Silicon Data's founder said the market is now in contango, so a cluster a year out costs a lot more. Portfolio companies that got 4,000 Blackwells in mid-2026 are shocked by prices for September 2027. Even with new supply coming, finding compute well into 2027 is hard. He thinks public markets are missing this.

    ▶ 12:45
  4. Short-term spot-priced contracts beat long fixed offtake deals, and running GPU clusters is like running a Formula 1 team

    Gavin says CoreWeave runs GPUs well but locked compute into long contracts now far below market. Elon's reasoning, he says, was that prices rise, so selling spot allows paybacks “well under a year” with big prepayments. Standing up a cluster is hard operational work, so he hopes capacity concentrates in proven operators like CoreWeave, Crusoe, Nebius, SpaceX and hyperscalers. Nvidia's pre-assembled DSX pod could cut energizing time from months and free working capital.

    ▶ 15:18

Give your email to unlock the rest of the key points — for every episode.

Do more with Charlie

Free — sign in with your email, no password.