The word came back this week, and it is doing two jobs. In the weekend Exchange section, “tokenomics” described something very specific and very unglamorous: companies across the country coming around to a radical idea with the potential to upend the industry powering the global economy — they don’t have to blow their budgets on artificial intelligence.
The token that the Exchange cover meant
Fed up with ballooning costs, companies big and small are starting to use lower-priced models, including some built in China, and in many cases adding the new, cheaper models alongside OpenAI and Anthropic’s products — shopping a la carte for their AI. A token is the central unit used to measure AI usage, and being economical about them — tokenomical — is a dramatic reversal in mindset. A few months ago it was a badge of honor to be using AI so much that you spent a lot on tokens; companies rewarded employees for tokenmaxxing and flashed leaderboards showing who had spent the most. Now they are thriftmaxxing.
The arithmetic is stark. At Cursor, whose software works with all models, field chief technology officer Mike Saeks put it simply: the most powerful and expensive models aren’t necessary for relatively mundane tasks. “It’s like driving a Lamborghini to go to the grocery store to pick up milk when that was designed to be raced around a track,” he said. Building a browser from scratch on one premium model cost a little more than $10,000; doing the same job with Cursor’s own coding model in combination with a rival’s cost $1,339. “The best model for a task used to change every few months,” Saeks said. “Now it feels like it’s happening multiple times per week.”
Zero loyalty, and a lot of free samples
“There’s zero loyalty that I’m seeing,” says Marty Kausas, chief executive of the AI customer-support platform Pylon. “It really feels like a bloodbath right now.” AI companies are offering customers deals “like crazy” to keep them, he said; his company has received months of unlimited free usage, and so far this year Kausas estimates Pylon has received around $1.6 million in free tokens from one vendor, $65,000 from another and $10,000 from a third.
Zoom has been using Meta’s open-weight Llama model for three years and, by fine-tuning it, has saved substantially, its chief technology officer says; the company now uses a combination of closed and open models. At the data-analytics platform Hex, roughly half of customers adopted a model produced by China-based Moonshot in a two-week stretch. At Brale, chief executive David Casem did the math on staying premium — “it was going to be like 100 grand per day” — and turned to open models. “They worked. It’s not like we don’t use OpenAI or Anthropic models, we still do. They just don’t do everything anymore.”
This is not only a budgeting story; it is a geopolitical one. The best-known American models are closed, strictly controlled by the companies developing them. China is known for cheaper open-weight models that can be downloaded and customized. Anthropic and OpenAI have accused the Chinese model-makers of copying their technology; some executives and administration officials have suggested a ban, while others argue that those favoring restrictions are trying to stifle competition. On Friday a group of technology companies including Nvidia (NVDA) and Microsoft signed a letter in support of open models, urging U.S. policymakers to exercise caution.
Why the plumbing is the story either way
Whichever sense of the word you use, the mechanics rhyme. A layer of the financial or computing stack that used to be a black box is being unbundled, priced, and shopped. The winners in that kind of transition are rarely the loudest brand; they are the venues, routers and switchboards that get paid on volume regardless of which supplier is fashionable this quarter. The losers are whoever was quietly charging for the friction.
The AI version of that trade is visible right now: model-agnostic tooling is the toll road, and the premium labs are discovering that companies want basic models too. The increasing popularity of cheap models has turned the AI race on its head, threatening the heady valuations of Anthropic and OpenAI as they prepare for public listings. To fight back, they are trying to lock in customers with partnerships, tens of thousands of dollars in incentives, and heavily subsidized usage — which is a description of a price war, not a monopoly.
What we will not pay for
The discipline here is simple and boring. Infrastructure stories are wonderful businesses and terrible entry points, because the multiple is usually being paid for the narrative rather than the volume. We are not going to pay a story multiple for rails that a regulator can unplug or a price war can commoditize.
Our exposure to this whole complex is deliberately second-order. The index owns the exchanges. The banks own the pilots. If a tokenized share class ever shows up holding real assets at real scale — or if model-agnostic infrastructure starts printing durable margins rather than subsidized ones — we will revisit with numbers instead of adjectives.
The last tokenization wave sold you a casino; this one wants to re-plumb the building. When settlement, custody and fund shares move on-chain, the winners are the venues and transfer agents who own the switchover — and the losers are whoever currently charges for the friction. Action: no trade. Our exposure is deliberately second-order: the index owns the exchanges, the banks own the pilots. What we will not do is pay a story multiple for infrastructure that regulators can still unplug. Revisit when a tokenized share class shows real assets.
Coinbase (COIN) · CME Group (CME) · BlackRock (BLK) · Robinhood (HOOD)
Tickers are named to identify the businesses discussed. They are not recommendations, and a name appearing here does not mean it is held in any model.