The Real AI Disruption Isn't Intelligence. It's the Price

The real disruption of AI isn't intelligence: it is the price. American startups are turning to Chinese models like DeepSeek and Kimi because they cost a fraction of Western rivals, and when the gap is this wide, geopolitics becomes background noise.

Take DeepSeek: about $0.28 per million tokens. Kimi goes even lower, at $0.15. Now look at Claude Opus, which can charge $1,500 for the same work. That’s not a small discount; it’s an abyss. Imagine choosing between a thousand-euro Michelin dinner and a three-euro perfect sandwich. When you’re feeding an army of algorithms, the choice isn’t really a choice , it’s an obligation. And it works. Pinterest tested Chinese models and found they cost 90% less while being 30% more accurate. When a company sees numbers like that, the conversation shifts.

The Real Disruption Is Arithmetic, Not Intelligence

An estimated 80% of American startups using open models are already “speaking Chinese” under the hood. The market has already decided who won this hand, even as we wonder whether AI is a bubble. In Hong Kong, the gold rush is palpable. Companies like Minimax see investors lining up to lend billions in a single afternoon , $14 billion raised in just the first months of 2026. The wallet has spoken, and it speaks Mandarin.

Disruption at the Margin: How Scarcity Forged a New Engineering

What fascinates me most is how they got there. The United States tried to hobble China by restricting access to powerful Nvidia chips. Instead of complaining, Chinese engineers rewrote the rules. They built architectures using hundreds of small experts working together, cutting memory consumption by 90%. They learned to run giant models on scrap chips. It’s the engineering of scarcity turned into a sharp weapon , a lesson in doing more with less that Silicon Valley had forgotten.

This is not just a cost advantage; it is a fundamentally different approach to AI. While the US and Europe raced to build ever-larger monolithic models, Chinese teams focused on efficiency and accessibility. The result is a stack that makes advanced AI as cheap and ubiquitous as a utility.

What Europe Can Learn , Or Lose

Us Europeans? Here’s where it gets bitter, and I say this with a heavy heart. We are great at studying, reflecting, producing talent , who then take a plane to work elsewhere. We have culture, ethics, but we’re slow. Stuck between endless regulations and a chronic inability to team up. While we write the perfect instruction manual, others are already building the house. The lesson is always the same: it’s not the one who invents the light bulb who wins, but the one who lights everyone’s streets at a bargain price. It happened with the internet, with phones, and it’s happening again. AI today is like water: cheap, accessible, ready to flow everywhere.

The real question, the one I ask myself at night when I close my laptop, isn’t what’s happening in Beijing or San Francisco. It’s whether we, watching from the sidelines, will start building something or remain spectators of a future written by others.

Frequently Asked Questions

For context: the price war is the economic side of the story I told in data centers spinning out of control and in the invisible AI trends of 2026: when compute gets cheaper, the disruption stops being about intelligence and starts being about access.