Coinbase CEO outlines 5 ways to cut AI costs without capping tokens

Coinbase is betting that smarter infrastructure, not usage caps, can keep its AI costs from spiraling as token consumption hits record highs.

Coinbase Chief Executive Officer Brian Armstrong outlined five strategies the crypto exchange is using to manage rising artificial intelligence costs without restricting how many tokens its engineers can use, according to a post on X on Friday.

"Ultimately, humans shouldn't be choosing models — AI can automate this task," Armstrong said, describing the company's approach to routing prompts to the most appropriate large language models based on task difficulty.

The first strategy involves selecting cheaper default LLMs. Coinbase is experimenting with Chinese open-weight models such as GLM 5.2, developed by Z.ai, and Kimi 2.7, from Moonshot AI, as defaults through its LLM gateway, Armstrong wrote. These models cost significantly less than those from frontier American AI labs like Anthropic and OpenAI.

The second approach, which Armstrong discussed earlier in June, routes prompts to models matched to task complexity — using frontier models for planning but cheaper ones for execution where they would be overkill. The remaining strategies include better caching to reduce inference costs, keeping context lean by starting new sessions when switching between tasks, and improving company-wide visibility into AI spending.

A graph attached to Armstrong's post showed token usage at Coinbase recently reached one of the highest levels in the company's history, while AI spending fell to nearly half its peak level. "The goal isn't to suppress usage. It's to build the infrastructure that makes exponential growth sustainable," he wrote.

The push to contain AI costs comes less than two months after Coinbase laid off 14% of its staff, partly because AI had changed how people work. In a post in May, Armstrong said engineers at the company were using AI to "ship in days what used to take a team weeks."

Coinbase's approach contrasts with the broader industry trend of imposing usage caps on employees to curb rampant token consumption. Companies including Uber and Amazon have reassessed their AI investments after budgets were depleted faster than expected, according to reports. A Flexera study published in early 2026 found that roughly 3 in 5 organizations acknowledged AI overspend had increased year over year.

Armstrong said all Coinbase engineers can use as many tokens as they want, but they can see their usage. The company expects "more impact" from employees who spend more on AI, he wrote.

This article is for informational purposes only and does not constitute investment advice.