Topic
The compute expense of running AI models at serving time, driven by GPU usage and token consumption, which shapes vendor pricing and margins.
Cathie Wood cites 99.99% annual falls in AI inference costs against Bill Ackman's warning that the Fed's September hike won't slow AI spending. Her figures price AI's output and his worry is demand for its inputs, so both can hold at once.
Reality
Honeywell's CTO put frontier AI at maybe 85% accuracy against the 99.9999% customers demand, and Ecolab said top models can cost more than people. Ecolab has since cut its token costs 70% to 80% by tuning, so cost looks like the easier of the two limits to fix.
Reality
LessWrong post puts the GPU cost of an AI doing an hour of median human work at about 4 cents, against a $25 US median wage. Current API prices narrow that gap sharply, and for the hardest tasks they lift AI cost to the hourly rate of a skilled engineer.
Reality
OpenAI reports 3.1 agent-workdays of agent runtime for every human workday, then spends much of the same report explaining why research did not get 3.1 times faster. The residue lands on review, compute allocation and deciding what to run.
Reality
Seat prices do not change, but the allowance sitting on top of each seat becomes a dollar balance of tokens, and the fallback to a cheaper model when that allowance runs out goes away on the same day.
Reality
Binny Gill argues in a Forbes council column that firms are paying reasoning-model rates for rule-following tasks. The two studies he cites measure consultants and a research router. Neither one measured an enterprise bill.
Reality
A Fortune essay by the authors of the forthcoming When Machines Act argues that the pacing fight misreads how AI reaches the economy. The survey lines it cites put the clock in enterprise data.
Reality
OpenBMB's 2B-parameter model is Apache-2.0, speaks 30 languages, and serves through vLLM's OpenAI-compatible /v1/audio/speech. The real-time factor quoted for it was measured on a different backend than the one the project recommends for production.
Reality
Praveen Neppalli Naga says treating AI cost as an engineering problem lowered Uber's cost per token even as employee use grew. The spending data for the wider enterprise market points the other way.
Reality
Adobe reported record third-quarter revenue of $6.76bn and its billionth monthly active user on September 10. By Shantanu Narayen's own split, freemium expansion explains only about half of the lowered ARR outlook.
Reality
Stanford's AI Index puts a 142-fold parameter cut and a more than 280-fold price cut behind one fixed MMLU threshold. That narrows where building your own still pays, and it lands in a state-law count that doubled in a year.
Reality
Two separate valuers cut Canva the same way after a third came off its growth forecast. The stated cause was compute cost the company could not pass on to customers.
Reality