More or Less · Topics · Episode #160
Nobody Can Explain the Compute Market Right Now
Clip from the full recording, 34:07–40:27 · download mp4 · watch at this point on YouTube
Wall Street is fighting over whether AI compute is about to get much more expensive or much cheaper; Dave says from the ground at OpenClaw, all he sees is bottlenecks everywhere, RAM stocks be damned.
Jess frames the debate raging among the “Gavin Baker types” of tech investing: some think compute is about to get wildly more expensive as long-term contracts come up for renewal at higher rates, better margins for the infrastructure providers; others think the opposite. Jess is skeptical of the “renewal shock” thesis, assuming labs have already planned for it, and asks Dave whether there’s a clear pattern coming on compute pricing.
Dave says he honestly doesn’t understand what the market is pricing — RAM stocks like Micron are down double digits even as, from his vantage at OpenClaw, every signal points the other way: the whole chip and RAM supply chain is running at full capacity and still can’t meet demand, with zero evidence token demand is falling. He makes it concrete: OpenClaw has been so constrained on Mac compute for running multi-agent engineering workflows that the company partnered with Apple on a custom batch of Mac Studios just to unblock its engineers, and the constraint runs across the entire stack — frontier tokens, rented GPUs, and the Mac Studios sitting on engineers’ desks that manage the agents in the first place.
Key points
- Jess: investors are split on whether AI compute is about to get much more expensive, as long-term contracts renew higher, or much cheaper.
- Jess is skeptical labs will get caught flat-footed by contract renewals; assumes they've already planned for it.
- Dave: doesn't understand the market reaction — RAM stocks like Micron are down sharply even as demand signals point the other way.
- Dave: the entire chip/RAM supply chain is running at full capacity and still can't meet demand; no sign token demand is falling.
- Dave: OpenClaw has been so compute-constrained running multi-agent engineering that it partnered with Apple on custom Mac Studio configurations just to unblock engineers.
- Dave: the constraint spans the whole stack — frontier tokens, rented GPUs, and the local Mac hardware needed to manage the agents.
Where they landed
Quotes
“”— Dave
“”— Dave
“”— Jess
Suggested tweets
Micron's down double digits. Dave Morin, from inside OpenClaw: “I don't see any signal that demand for tokens is going down. I only see bottlenecks.”
Tweet this →Investors are fighting over whether AI compute is about to get way more expensive or way cheaper. Dave Morin: “I don't really understand what the market is doing right now.”
Tweet this →OpenClaw got so compute-constrained it partnered with Apple on custom Mac Studios just to unblock its engineers
Tweet this →All topics · Full episode #160 + transcript · Subscribe on YouTube · Spotify