AI companies must buy compute far in advance.
Mostly true True in substance, wrong in a detail.
Factual: Statements about the present or the past, checked against the record: did they hold up?
The second judge would not resolve this claim; the verdict shown is the first judge's.
What would count as true or false
Judged as of 2026-05-09. True if credible industry evidence shows that frontier AI companies generally secure compute well ahead of use: multi-year cloud or data-centre contracts, chip orders or capacity reservations made months to years before deployment. Mostly true if advance purchasing is the dominant practice but some companies also rely meaningfully on spot or short-term capacity. Mixed if advance and short-term buying are roughly equally common among AI companies. Mostly false if advance purchasing happens but is the exception, and the 'must' overstates the practice. False if evidence shows AI companies typically buy compute on demand without long lead times.
Set before the research began.
Evidence (4)
-
E1 Anthropic’s Daniela Amodei Charts Leaner A.I. Strategy as IPO Race With OpenAI Heats Observer · 5 June 2026
“The structure of these deals is you have to commit to a certain amount of compute reasonably far in advance”
-
E2 Higher usage limits and a SpaceX compute deal Anthropic · 6 May 2026 Own record
“A 5 GW agreement with Google and Broadcom, which will begin coming online in 2027;”
-
E3 GPU Contracts Assume Steady Demand; AI Workloads Do Not Orply · 22 October 2026
“The biggest downside of the GPU scarcity is not that it's hard to get GPUs (you just have to pay money), but that almost all capacity is sold as 3 year rectangles starting 3-6 months in the future.”
-
E4 Nvidia GPUs become core AI resource as firms mobilise all channels to secure supply Digital Today · 12 October 2026
“Some providers offer GPUs in relatively small units without prepayment or long-term contracts. But the more demand concentrates on a provider, the more it may require prepayment or take months before GPUs are actually up and running.”
Reasoning
Rated mostly true: buying ahead is the main practice, but short-term buying also exists. An independent account published by Orply says almost all GPU capacity is sold as 3-year blocks that start 3-6 months in the future (E3). Anthropic's own announcement lists a 5 GW agreement with Google and Broadcom that begins coming online in 2027, which is capacity committed ahead of use (E2). Another independent report says some providers sell GPUs in small units without prepayment or long-term contracts, though prepayment and longer lead times come with concentrated demand (E4). That short-term option is why the claim's 'must' is slightly too strong.
Cite this claim
Daniela Amodei, “AI companies must buy compute far in advance.” (said 9 May 2026). Verdict: mostly true. Reliability API, claim c_6d8ade638480, method 0.1.0+28da144b, dataset release v1.1.2. https://epistacklabs.com/claims/c_6d8ade638480
This claim is in dataset release v1.1.2, with the same verdict.
Reply to this claim
More factual claims by Daniela Amodei
-
Mostly true
“you’re essentially making a bet on the future.”
AI companies must buy compute far in advance, essentially making a bet on future demand.
-
Resolved true
“It takes a lot of compute, and that compute is in scarce supply.”
Training AI models takes a lot of compute, and compute is in scarce supply.
-
Mostly true
“If that scenario ever changed drastically, it would cause problems because both Anthropic and OpenAI have already purchased a lot of compute for the future, which is very costly.”
Anthropic and OpenAI have already purchased a lot of compute for the future, which is very costly.
-
Resolved true
“Training these models is extremely expensive. It takes a massive amount of compute power, which is limited in supply.”
-
Mostly true
“I think that's largely because we're seeing this trend of training large generative models to just be almost prohibitively expensive, right.”
Training large generative models is becoming almost prohibitively expensive.
Verdicts are made automatically; how verdicts are made.