AI spending is no longer simply a race to use more tokens: 36Kr describes US businesses splitting into three camps, from giants building their own models to smaller firms seeking cheaper alternatives. The practical pressure point is the bill for putting AI tools into everyday work, especially coding agents.
Watch Desk analysis
What happened
In a report published on 10 October, 36Kr says large companies are developing in-house alternatives, mid-sized firms are cutting usage or downgrading services, and smaller businesses are more likely to try Chinese open-weight models. It cites Uber as an example of the middle tier: the report says the company exhausted its annual AI programming budget in under four months after rolling out Claude Code to about 5,000 engineers.
For smaller firms, 36Kr points to OpenRouter’s model traffic as a sign of interest in lower-cost options. The report cautions that this platform’s users skew towards developers and smaller teams, so its figures should not be read as a measure of the whole US enterprise market.
Why it matters
Cheaper tokens do not guarantee cheaper work if each task consumes more of them. For organisations, the choice is increasingly about matching models and agent access to the job, setting limits and deciding whether the productivity gain justifies the spend. For model providers, a customer that starts with a cheaper alternative may be harder to win back later.
Our read
The useful signal here is not a single market-share number, but the changing procurement question: companies are weighing the cost of each task, not just the sticker price of a model. 36Kr’s examples make that shift worth watching, though its broad three-tier picture should be treated as the publication’s analysis rather than a settled census. AI budgets, it turns out, are quite capable of having a plot.
What to watch
- Whether companies disclose how they measure AI spending against productivity gains.
- Whether internal tools and cheaper models reduce costs without narrowing access to useful systems.
- Whether more firms publish comparable data on model usage and procurement choices.
Discussion spark: Should companies respond to rising AI bills by limiting staff access, or should they first have to show that the tools are not earning their keep?
Sources and evidence
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.