Lindy dumps Anthropic's Claude for Deepseek, saves millions

Lindy, a 25-person AI startup, ditched Anthropic's Claude entirely on Jun 26, 2026, replacing it with Deepseek, a model hosted by a US company on US soil. CEO Flo Crivello told CNBC the cost curve "crashed to the ground," and the move saved the company millions after AI spending had exceeded personnel costs.

Why did Lindy switch from Claude to Deepseek?

Lindy replaced Claude with Deepseek to cut a bill that had become larger than payroll; the company says AI costs were "unsustainable" and exceeded personnel costs for the 25-person startup. Crivello told CNBC that the swap to Deepseek reduced expenses dramatically and that he would revert to Claude only if Anthropic lowered its prices, calling the decision "a matter of survival for the business."

The switch also aimed to keep data and hosting in the United States: Deepseek is hosted by a US company on US soil, a detail Lindy emphasized when announcing the change. The company framed the move as strictly financial and operational rather than technical — the priority was lowering token and inference spend that had been burning through the startup's budget.

What cheaper alternatives are companies using, and how do they compare?

An analysis cited by the story, from Snowflake's CTO, found that Chinese models such as GLM-5.2 do not quite match Claude on performance but can be competitive and often win on price-performance depending on the task. That math encourages startups and buyers to choose lower-cost models when budgets tighten.

The coverage also notes OpenAI CEO Sam Altman calling AI cost a "huge issue" after a shift to agentic systems that consume many more tokens. Taken together, those signals pushed Lindy and others to prioritize cost-per-result over small performance gaps.

Why it matters

Cost pressure at the application level can reshape vendor economics and go-to-market strategies. Lindy's switch shows that even a fast-growing startup will replace a higher-cost provider when AI expenses outpace salaries. The company framing this as survival ties vendor pricing directly to startup viability, a dynamic that can force established model providers to reconsider pricing or risk customer churn.

The piece also links broader market effects: companies tightening AI spending, the competitiveness of lower-cost alternatives like GLM-5.2 on price-performance, and public-company timing (the story suggests Anthropic faces pressure around an IPO) all amplify why this single customer decision matters beyond one startup.

What to watch

Watch whether Anthropic cuts prices or announces new commercial tiers: Crivello said Lindy would switch back if Anthropic reduced costs. Also track further independent comparisons of Claude, GLM-5.2 and other lower-cost models that measure price-performance by task, since the Snowflake CTO analysis named task-dependent wins for cheaper models.