Skip to content

Topic

AI Compute Costs

The cost of GPU capacity, owned or leased, that AI companies consume to train models and serve inference, and its effect on gross margin and cash burn.

Current stories

investConfirmed8 publishers

OpenAI's two-week training pause is a capex line item, not a thought experiment

The largest frontier reinforcement-learning runs are still on hold, and the new containment rules add roughly 20% to training compute. Loss of control now shows up in schedules and budgets.

Perspective Coverage

8 publishers
Builder
Builder 30%
Operator
Operator 38%
Investor
Investor 32%

Reality

Evidence60
Adoption
Not scored
Hype gap+15
Incentives60
Confidence58
productConfirmed3 publishers

Anthropic spent about $1.60 on compute for each dollar of revenue in 2025, leaked prospectus shows

Anthropic spent $7.33 billion on compute against almost $4.6 billion of 2025 revenue, according to the leaked IPO prospectus Reuters obtained. For teams running Claude, that gap and the company's fixed infrastructure deals say more about future prices than the $42 billion headline loss does.

Perspective Coverage

7 publishers
Builder
Builder 18%
Operator
Operator 23%
Investor
Investor 59%

Reality

Evidence55
Adoption60
Hype gap+20
Incentives65
Confidence55
productOne report1 publisher

An AI chip insider argues inference should get cheap enough that teams stop counting tokens

A SiliconAngle opinion piece written from inside the AI semiconductor business says commodity inference would grow the market. Teams that ration tokens today still have to decide whether their caps are configuration or architecture.

Publishers:siliconangle.com

Reality

Evidence18
Adoption
Not scored
Hype gap+42
Incentives80
Confidence38
investOne report1 publisher

Cognition's $48bn price keeps the same 53x run-rate multiple it carried in May

The Series E doubled the valuation while run-rate revenue rose 83% to nearly $900m, so the price per dollar of revenue barely moved, which leaves the undisclosed gross margin under an $800m compute burn as the open question.

Reality

Evidence38
Adoption60
Hype gap+26
Incentives82
Confidence52