American companies are getting stingy about artificial intelligence — and that shift could reach all the way to Wall Street.
According to a Wall Street Journal report highlighted by Techmeme, US firms large and small have flipped from what the paper calls "tokenmaxxing" — spending freely on AI usage — to "thrift-maxxing," hunting for the cheapest way to get a job done. In practice, that means mixing cheaper Chinese models alongside offerings from OpenAI and Anthropic rather than committing to a single expensive provider.
A quick translation for non-engineers: AI systems charge by the "token," roughly a chunk of text going in or coming out. Every customer support reply, code suggestion, or document summary burns tokens, and the bill scales with usage. When a company runs millions of those requests, the difference between a premium model and a budget one stops being a rounding error and becomes a line item somebody has to defend.
The Journal frames this as more than a procurement tweak. Companies mixing and matching models is "changing the economics and power players of the industry," the report says — and it poses a threat to the IPO valuations of the leading American labs. Those valuations rest on an assumption that customers will keep paying premium rates at growing volume. If businesses treat models as interchangeable parts, picking whichever is cheapest for each task, that assumption weakens.
The report does not name specific Chinese models or the companies making the switch, and the broader dollar figures aren't detailed in the summary available.
Why it matters: if the most advanced AI becomes something buyers shop for on price rather than brand loyalty, the enormous valuations attached to America's top AI labs — and the investment case for their eventual public offerings — rest on shakier ground than the hype suggests.