A firm blew its 2026 AI budget by April, then capped staff at $1,500/month
The company set its AI budget in 2025, before coding agents got hungry. By April the money was gone, and engineers got a hard monthly token ceiling.
One ride-hailing firm tried to plan a year of AI spend in advance and ran out in four months. The fix was a per-person cap of $1,500 a month on individual AI tools.
The catch nobody priced in: none of the model vendors are profitable, so today's bills are already subsidized. The number companies are straining against is the cheap one.
And the cited stat about Chinese models 'overtaking' US ones in usage rests on a single aggregator's traffic, not the APIs most customers actually hit.
The budget set in 2025 that didn't survive to summer
The Financial Times reported that companies are pulling back on AI as the bills outrun what anyone forecast. The detail that traveled furthest in the HN thread was a single ride-hailing firm: it set its 2026 AI budget, blew through the whole thing by April, and slapped usage caps on staff, limiting each employee to $1,500 in monthly token spend on individual AI tools.
The reason that number looks absurd in hindsight is timing. The budget was drawn up before the latest generation of coding agents landed.
"Right, because they set their 2026 budget in 2025. And in 2025 nobody could predict how good (and token-hungry) coding agents would get after November 2025… I'd be surprised if any company that set an AI budget for 2026 hasn't blown through it by now." — Simon Willison, creator of Datasette
The number they're fighting is the subsidized one
The sharper economic read came early in the thread: the costs straining budgets aren't even the real costs.
"None of the AI companies are profitable, i.e. they're all still charging substantially less than what it costs to actually deliver the service. With things in that intermediate state, it's hard to know what a future stable state will be like." — a commenter on the economics of unprofitable vendors
If the metered price sits below cost today, the line companies are wincing at is the discounted one. Several voices expected the pressure to push workloads on-prem, where spend can be fixed rather than metered.
The stat that doesn't say what it sounds like
The FT cited OpenRouter data showing Chinese models overtaking US ones in token consumption. Willison flagged the asterisk:
"That's a bit of a dodgy statistic. OpenRouter only tracks their own users — the vast majority of API customers for OpenAI and Anthropic presumably go straight to their APIs." — Simon Willison, on what the aggregator can and can't see
The incentive that built the overspend
A few people in the thread had watched the spending culture form from the inside. One described employers literally rewarding token burn:
"I knew this was going to be the end result after seeing so many companies reward employees based on how many tokens they were using. My company even pulled stats and gave physical awards out at an in-person retreat." — an engineer whose company handed out awards for token usage
A blunter take framed the whole cycle as a management failure waiting for its PIP:
"CEOs laid people off to replace them with AI, but turns out AI is more expensive and does a worse job. If I made a blunder of that scale, would I or would I not be put on a PIP?" — a commenter scoring the layoff-then-spend cycle
Not everyone took the bear side. An engineer who uses LLMs daily for coding pushed against both camps, arguing the tools have obvious value when you already know what you want and read every line they produce. The honest middle: faster output, real spend, and a culture that confused token volume with productivity.
Get breakdowns like this every day
The top Hacker News stories, unpacked — by email, Telegram, or RSS.
Readers beat a 170k-word vocab quiz by always picking the longest answer
The web app guesses how many of English's 170,000 words you know from a 100-question quiz. One reader wrote a six-line console script that just clicks the longest option every time — and scored in the high 70s to mid 80s.