Friday Jul 31
STRATEGYOther

Compute May Get 10x More Expensive

31JUL
COMPUTE10x

Dwarkesh Patel makes a case every AI roadmap should fear. If software engineering is automated by 2028, compute could cost 15x more. Smarter models earn more per chip, so demand sets the price, not supply.

Rent an H100 at what a human engineer costs, and compute looks dirt cheap. That gap, he argues, is the real ceiling on how expensive chips can get.

As models improve at using the same GPU, labs can pay far more for it. Non-frontier labs lose that race first, unable to monetize compute as well. Frontier labs would out-bid everyone else for the same chip supply.

This flips the usual worry: compute may become too expensive to rent, not too scarce. Every roadmap assuming flat GPU costs through 2028 may need a rewrite.

full brief & sources

⚡ Why this matters

  • If Patel is right, every multi-year AI product roadmap built on today's GPU pricing needs a rewrite.
  • It reframes the AI cost debate: the constraint isn't chip supply, it's what labs can afford to pay for the chips that exist.
  • This is an argument being debated by operators right now, not settled fact.

🔍 What happened

  • Dwarkesh Patel published the essay 'Why compute might get 10x more expensive' this week.
  • His model: price an H100 at what a human-equivalent software engineer earns, roughly $250k a year.
  • That's about 15 times today's spot rental price for the same chip.
  • The mechanism: as models get smarter, they extract more economic value per chip, which raises what labs will pay to rent it.
  • Frontier labs, who monetize compute best, would out-bid smaller labs for the same limited supply.

💬 Smart takes

  • Dwarkesh Patel: if software engineering is automated by 2028 and compute costs 15x more, non-frontier labs can't compete for chips against the labs that can pay.
  • Skeptic: this assumes software engineering actually gets automated on that timeline, and every AI timeline bet made so far has run long.

🧭 Where this goes

  1. Likelythis essay gets cited in the next round of AI infrastructure-spending debates.
  2. Possibleat least one mid-tier AI lab cites rising compute costs as a reason it can't keep pace with frontier labs.
  3. PossibleGPU rental spot prices tick up in 2027 as model efficiency improves faster than chip supply.
  4. Wild Cardcompute pricing becomes the actual binding constraint on AI progress before any safety or data limit does.

🥄 The Spoon Take

Everyone's been worried about running out of chips. Patel's argument is scarier: chips stay available, they just get priced like the value they unlock, not like hardware. If he's right, the AI race stops being about who has the most GPUs and starts being about who can afford to rent them.

🤔 Pushback

This is one podcaster's model with a lot of assumptions baked in, not a lab's internal forecast.

·Dwarkesh.com - Compute cost essay