Flo Crivello, CEO of AI startup Lindy, made a decision that would have been unthinkable two years ago. He ditched Anthropic's Claude entirely. Switched to DeepSeek, hosted on US soil. And watched his AI costs — which had exceeded his entire personnel budget — crash to the ground.
"It's a matter of survival for the business," he told CNBC. He'd switch back to Claude if Anthropic cut prices. But they haven't. So he won't.
This isn't a story about nationalism or patriotism or which country builds better AI. This is a story about math. And the math says American AI labs are charging prices that no longer make sense for the companies actually using their products.
The Cost Crisis Nobody Wants to Talk About
Lindy is a 25-person startup. Their AI costs had become "unsustainable" — exceeding what they paid their entire team. This is the dirty secret of the AI boom: the infrastructure costs are eating companies alive, and the labs selling the shovels are getting rich while their customers go broke.
Sam Altman recently admitted that AI costs became a "huge issue" for companies as they switched to agentic systems that burn through tokens at unprecedented rates. OpenAI's own CEO is acknowledging what every startup founder already knows: the current pricing model doesn't work for anyone except the labs themselves.
Enter DeepSeek. Hosted by a US company on US soil, so no geopolitical complications. Competitive performance on most tasks. And a price point that makes Claude look like a luxury good.
Snowflake's CTO recently ran the numbers on GLM-5.2, another Chinese model. It doesn't quite match Claude Opus 4.7 on every benchmark, but it's competitive — and the price-performance ratio isn't even close. For most real-world tasks, the difference is negligible. The cost difference is massive.
The Arrogance of American AI Pricing
Here's what American AI labs don't understand: startups don't care about your benchmark scores. They care about whether they can afford to stay in business.
Anthropic is reportedly seeking a $900 billion valuation. OpenAI is aiming for a $1 trillion IPO. These are numbers that make sense only if you assume customers will keep paying premium prices forever. But Lindy just proved they won't.
The switch to DeepSeek isn't ideological. It's economic. Crivello isn't a China hawk or an America basher. He's a CEO trying to keep his company alive. When your AI bill exceeds your payroll, you don't have the luxury of brand loyalty.
And here's the kicker: he'd switch back to Claude if the price was right. This isn't a permanent defection. It's a negotiation. Lindy is voting with its wallet, and the message is clear: cut prices or lose customers.
What This Means for the AI Industry
The Lindy story is a canary in the coal mine. If a well-funded, AI-native startup can't afford American model prices, what does that mean for the rest of the industry?
It means the pricing power of American AI labs is built on sand. They've been charging monopoly prices in a market that's about to get very competitive, very fast. Chinese models are improving rapidly. Open-source alternatives are getting better. And customers are discovering that "good enough" at 1/10th the price is actually better than "best in class" at a price that bankrupts you.
Anthropic's response to this pressure has been to build bigger models, not cheaper ones. They're chasing the frontier while their customers are fleeing to the mid-tier. It's a classic innovator's dilemma — except the disruptors aren't startups in a garage, they're Chinese labs with government backing and a cost structure that American companies can't match.
🔥 Hot Takes
1. Anthropic's $900B valuation assumes customers are captive. They're not. Lindy just proved that switching costs in AI are near zero. If DeepSeek is hosted in the US and accessible via the same API, the only difference is the model name in your code. Anthropic and OpenAI have spent years building moats around their technology, but they forgot to build a moat around their pricing. When a 25-person startup can save millions by changing one line of code, your pricing power is an illusion.
2. The "US vs China AI race" narrative is a distraction from the real competition: price. American politicians want to frame this as a national security issue. But for startups like Lindy, it's a spreadsheet issue. DeepSeek isn't winning because it's Chinese. It's winning because it's affordable. If Anthropic cut prices 80% tomorrow, Lindy would switch back. The loyalty isn't to China — it's to survival.
3. AI labs are building a two-tier market, and they're pricing themselves into the luxury tier. The top tier — enterprises with infinite budgets, government contracts, and strategic AI investments — will keep paying premium prices for Claude and GPT. Everyone else — startups, small businesses, developers — will migrate to Chinese models, open-source alternatives, or smaller American labs. The result: American AI becomes a luxury good for the Fortune 500, while the rest of the world builds on Chinese infrastructure. That's not a win for American AI dominance. That's a market segmentation that favors China.
The Bottom Line
Lindy's switch to DeepSeek isn't an isolated incident. It's the beginning of a trend. As AI costs continue to rise and Chinese models continue to improve, more startups will make the same calculation: is the marginal performance gain worth the massive price premium?
For most companies, the answer is no. And as more companies vote with their wallets, American AI labs will face a choice: cut prices and accept lower margins, or watch their market share erode to competitors who understand that in a commoditizing market, price is the only moat that matters.
Flo Crivello chose survival over loyalty. Millions of dollars in savings say he made the right call. The question is whether Anthropic and OpenAI are listening.