AI startup Lindy has become a clear example of how quickly model costs can shape product decisions. The company moved entirely away from Claude and adopted Deepseek, a cheaper alternative that CEO Flo Crivello said changed the economics of the business.
The move matters because it shows a practical pressure point for Anthropic: strong growth can still face resistance if customers decide that frontier model pricing no longer fits their operating budgets.
Why Lindy left Claude
Lindy is a 25-person startup, and its AI costs had become “unsustainable,” according to CEO Flo Crivello. The source article says those costs exceeded personnel costs, making model spending one of the company’s defining business issues.
Crivello said Lindy ditched Claude entirely for Deepseek. The Deepseek setup was hosted by a US company on US soil, which is an important part of how the company described the switch.
The financial result was immediate in business terms. Crivello told CNBC the cost curve “crashed to the ground,” and the change saved millions.
His explanation was not framed as a preference for one model brand over another. It was a survival question. Crivello said he would switch back if Anthropic cut prices: “It’s a matter of survival for the business.”
What this says about AI spending
The Lindy case points to a broader pattern in AI adoption: as companies use AI more deeply, model bills can become harder to ignore. That is especially true when systems burn through tokens at high volume.
OpenAI CEO Sam Altman recently said that AI cost became a “huge issue” for companies with the recent switch to agentic systems burning through tokens. The source connects that concern directly to the kind of spending pressure Lindy faced.
Agentic systems can increase usage because they do more than answer a single prompt. They may run through steps, call tools, perform tasks, and continue working across a workflow. The source does not provide specific usage figures, but the business implication is clear: when token use rises, pricing becomes a central infrastructure decision.
For startups, that pressure can be especially sharp. If AI costs exceed personnel costs, model selection is no longer a minor technical choice. It becomes part of the company’s operating model.
Cheaper models are becoming harder to ignore
The source article also points to a recent analysis by Snowflake's CTO. That analysis showed that affordable Chinese models like GLM-5.2 do not quite match Claude, but can be competitive and can easily win on the price-performance ratio depending on the task.
That distinction is important. The argument is not that every cheaper model is always better. It is that many real business tasks do not require the strongest possible model if another option can deliver acceptable results at a much lower cost.
For companies building AI products, the calculation may include several practical questions:
- Is Claude meaningfully better for this specific task?
- Does the performance gap matter to the user experience?
- Does the price difference change the viability of the business?
- Can a cheaper model handle enough work to reduce overall spending?
Lindy’s decision suggests that, at least for its own use case, the cost difference was large enough to justify a full move away from Claude.
The pressure on Anthropic
The source frames Lindy’s switch as part of mounting cost pressure on Anthropic. Anthropic has seen explosive growth, but that growth could come under pressure as companies tighten AI spending and turn to cheaper Chinese alternatives.
That does not mean customers are abandoning high-end models altogether. Crivello explicitly said he would switch back if Anthropic cut prices. The issue is not only capability; it is whether the cost of using Claude fits the economics of companies building AI-dependent products.
The article also places this pressure in the context of IPO timing. It says Anthropic better get moving on its IPO, while OpenAI apparently already missed its optimal window and postponed. The broader point is that market momentum can shift if customers become more cost-sensitive before companies reach their preferred financial milestones.
For Anthropic, the Lindy example is a warning from the customer side of the market. If AI startups can save millions by using Deepseek or similar models, pricing becomes a competitive feature, not just a finance detail.
What comes next for model choice
Lindy’s move highlights a practical future for AI infrastructure: companies may choose models based on workload, cost, and acceptable performance rather than brand loyalty. A model that is slightly stronger may still lose a deployment if the price-performance ratio does not work for the task.
That dynamic could make AI procurement more selective. Companies may keep evaluating Claude, Deepseek, GLM-5.2, and other models through the lens of business survival, especially as agentic systems increase token consumption.
The central lesson from Lindy is simple. AI capability still matters, but cost now matters enough to change vendor decisions completely. For Anthropic, that makes pricing a strategic issue as much as a product issue.