Lindy Swaps Claude for DeepSeek to Save Millions in AI Costs

As the era of agentic AI systems begins to burn through massive amounts of tokens, startups are facing a brutal reality: the cost of intelligence can quickly outpace revenue. AI startup Lindy has made a decisive move to survive this economic pressure by completely abandoning Anthropic’s Claude in favor of DeepSeek.

The Unsustainable Cost of Frontier Models

For many high-growth AI startups, the expense of running large language models (LLMs) has become a primary threat to solvency. Lindy, a lean team of just 25 employees, found itself in a position where its AI compute expenses were actually exceeding its personnel costs. CEO Flo Crivello recently revealed to CNBC that these costs had become "unsustainable," threatening the very survival of the business.

To combat this, Lindy executed a massive pivot, ditching the Claude ecosystem entirely. By migrating to DeepSeek—which is hosted by a US-based company on US soil to mitigate geopolitical concerns—the startup saw its cost curve "crash to the ground," resulting in millions of dollars in savings.

The Rise of Price-Performance Efficiency

Lindy’s move reflects a broader shift in the industry toward "price-performance" optimization rather than pure model capability. While frontier models like Claude and GPT-4 represent the gold standard for reasoning, they come with a premium price tag that is difficult for scaling startups to maintain.

Industry experts are beginning to validate this transition. A recent analysis by Snowflake's CTO noted that while certain Chinese models, such as GLM-5.2, may not strictly outperform Claude in every benchmark, they are highly competitive and often win on a task-specific price-performance ratio. As companies move toward agentic workflows—where AI agents perform multi-step tasks autonomously—the sheer volume of tokens required makes high-cost models a liability rather than an asset.

Implications for the AI Landscape and Anthropic

This development serves as a warning shot for model providers like Anthropic. As OpenAI's Sam Altman has acknowledged, the transition to agentic systems is making AI costs a "huge issue" for enterprises. If frontier model providers do not address their pricing structures, they risk losing significant market share to more efficient, specialized, or affordable alternatives.

Crivello noted that the decision to use DeepSeek is not necessarily permanent; he stated he would switch back to Anthropic if the company significantly reduced its prices. For now, the priority for Lindy and similar startups is operational survival in a market where the cost of intelligence is becoming a central pillar of business strategy.

Key Takeaways

• Cost-Driven Migration: AI startups like Lindy are shifting from premium models like Claude to more affordable alternatives like DeepSeek to prevent compute costs from exceeding personnel expenses.
• The Agentic Token Crunch: The rise of agentic AI systems, which require massive token consumption, is driving a surge in demand for high price-performance efficiency.
• Market Pressure on Providers: Frontier model companies face growing pressure to lower prices as developers prioritize sustainable unit economics over absolute peak performance.