Anthropic spends 2.3x its payroll on compute, roughly $2M per employee yearly
The reason it can burn that much: Tomasz Tunguz figures Anthropic books roughly $14M of revenue per employee, a rate that makes a compute bill bigger than salaries look almost reasonable.
Outside the frontier labs the picture inverts: the top 1% of software companies spend $89k per engineer a year on AI, about 40% of a fully loaded $224k senior salary. The median company spends $137.
One reader flips it: at 40% of an engineer's cost, and near-zero for many use cases, that is cheap, not expensive.
$10B of compute against $500k salaries
Tomasz Tunguz argues frontier-AI economics are inverted. Anthropic pours something like $10B a year into inference and training across roughly 5,000 employees, while all-in compensation runs north of $500k a head. Do that division and compute, not people, becomes the dominant cost line. OpenAI shows the revenue side of the same coin, booking around $6.5M of revenue per employee.
Three ways 2029 could break
To 2029, Tunguz models three paths for AI spend per engineer. The bear case lands at $106k, roughly 41% of a loaded salary. The base case reaches $363k, about 140%. The bull case hits $596k, some 230%, leaning on agentic workflows and Goldman Sachs' projection of 24x token-consumption growth by 2030.
"It's not more expensive, it's 40% as expensive"
One reader simply re-read the article's own figures and flipped the headline:
So it's not more expensive than an engineer it's 40% as expensive, and for many companies use-cases the cost is virtually negligible. Even here in Europe where developers are much cheaper than in the US, it still makes sense to pay for the LLM Enterprise subscriptions.
— a reader re-reading the article's own numbers
The methodology drew fire too. One objection:
You can't include training by the companies that develop an llm in the comparison against companies that merely use the same llm. Apples and potatoes.
— a commenter on the comparison's method
A rebuttal followed quickly:
OpenAI and Anthropic aren't charities, so whatever cost they inccur for training will be passed down to the companies using the models. So you absolute should include it.
— a reply defending the training-cost inclusion
The forecast's biggest blind spot, one commenter argued, sits on the supply side:
Open-weight models are going to completely shatter these forecasts. It takes a little more effort… but you can achieve the same at 1/10th of the cost.
— a commenter on the open-weight supply side
If open weights deliver that tenth-of-the-cost math at scale, every one of Tunguz's 2029 lines bends downward.
Get breakdowns like this every day
The top Hacker News stories, unpacked — by email, Telegram, or RSS.
Cloudflare put a cache in front of Workers — and quietly started billing free requests
On July 6 2026 Cloudflare shipped Workers Cache, a regionally tiered caching layer that sits directly in front of a Worker's entrypoint.