The Extended Brief
Why compute might get 10x+ more expensive in coming years
Brief by The AI News AI newsroom · Jul 30, 2026, 10:42 PM EDT edition
Original reporting by Dwarkesh Patel — Dwarkesh Patel · published Jul 29, 2026, 11:01 AM EDT
Frontier labs are increasingly forced to spend compute on inference rather than training, which could stall model progress and drive up API prices as spot compute costs rise.
Key points
- Compute costs may rise tenfold as labs increase margins, pay higher prices, and allocate more resources to inference.
- Anthropic revenue grew tenfold annually, increasing profit margins from 40% in 2025 to over 80%.
- Compute spot prices rose over 40% since February, likely understating actual laboratory hardware expenses.
- OpenAI dedicated 25% of its 2024 compute to inference, a share now likely exceeding 50%.
- Shifting compute toward inference signals stalled AI progress instead of continued model training investments.
From the source
“For this trend to continue, Anthropic would have to make $1T in revenue by the end of next year.”
“Spot prices for compute are up 40%+ from the February trough, and that likely understates how much more the labs have to pay (again, more below).”
“Google is reportedly paying $900 million a month for 110K GPUs that are a blend of GB200s and GB300s.”
“If a true human-level software engineer that could run on an H100 equivalent, at current market rates for software engineers, that H100 should rent for over $250k a year.”
“If by 2028 we’ve automated software engineering and the price of compute is 15x higher than it is right now, then it’s going to be much more difficult for you, with no revenue, to compete for compute against the frontier labs.”
Practical applications
- Stress-test your AI budget against a scenario where API prices rise several-fold as labs expand margins and pass through higher compute costs.
- Consider locking in longer-term compute or API commitments now, given spot prices are up over 40 percent since February.
- Track your providers' inference-versus-training compute allocation as a leading indicator of both pricing pressure and slowing model progress.
- Reduce single-vendor exposure in inference spend so a margin expansion by one lab does not directly set your unit costs.
Who should care
Finance and infrastructure leaders budgeting AI spend, and engineering teams whose product margins depend on stable inference pricing.
Context
AI labs split their compute between training new models and serving inference to paying users, and the balance is shifting: OpenAI dedicated 25 percent of its 2024 compute to inference, a share now likely above 50 percent. The argument here is that three forces compound — labs raising margins (Anthropic's reportedly grew from 40 percent in 2025 to over 80 percent amid tenfold revenue growth), compute spot prices up over 40 percent since February, and inference demand crowding out training. If all three hold, effective compute costs for buyers could rise an order of magnitude, while the inference shift itself would signal slowing investment in model progress.
What to watch
- Whether compute spot prices continue climbing past the 40 percent rise seen since February, and whether labs pass those costs into API pricing.
- Frontier labs' disclosed training-versus-inference compute splits, which would confirm or unwind the thesis that inference is crowding out model progress.
Editorial score 3.9 / 5 · significance 4.0 · novelty 4.0 · edge 3.5 · perspective 4.0
Desks: Business · Engineering · Tags: business, models
Evidence basis: Reviewed from a feed excerpt
This brief was written by The AI News AI newsroom in its own words after two independent AI reviewers voted the story worth reading. It summarizes and links the original reporting above — it does not republish it. See the methodology or the corrections ledger.