Remember when everyone freaked out last year about DeepSeek and Chinese AI? Turns out that was the wrong DeepSeek moment to worry about.
Last week, Jensen Huang went on a podcast and said the day DeepSeek releases a model running on Huawei chips instead of Nvidia GPUs would be "a terrible day for this nation." Three days later, DeepSeek released exactly that.
DeepSeek V4 Pro is now live, running entirely on Huawei Ascend 950 GPUs in production. It's the first frontier AI model designed and optimized for non-American hardware. And while it doesn't quite match the absolute peak performance of U.S. models, it gets close enough while costing one-sixth as much to run.
Let me translate what that means: if you're a business in a country that isn't swimming in cash, or you're a developer who doesn't need the last 3-5% of model performance, you now have a viable alternative that costs 20 cents on the dollar compared to ChatGPT or Claude.
Wall Street is completely ignoring this. Everyone got conditioned last year to dismiss Chinese AI competition after the first DeepSeek scare turned out to be overblown. But this time is different, because this time it's about the hardware stack, not just the software.
The Moat Just Got Narrower
The entire AI investment thesis for the past two years has rested on the assumption that Nvidia has an unbreakable moat. U.S. export controls were supposed to keep China from building competitive AI chips. Except now they've done it anyway, and DeepSeek just proved the chips work at scale.
DeepSeek even said their current costs are "high" because Huawei is still scaling production, and they expect prices to drop significantly later this year. If a Chinese tech stack that's "good enough" becomes available at 20% of the cost of the American version, that's going to reshape the entire global AI market outside the Fortune 500.
The biggest companies will still pay premium prices for , , and . But everyone else? Startups, small businesses, governments in developing countries, individual developers - they're going to start asking why they're paying five times more for marginal improvements.
