Post
Post

Republic Research / ARTICLE

The AI hardware trade is cracking hard this morning, and it's worth understanding what actually set it off, because it isn't just profit-taking. The trigger came out of China. A company called Moonshot released a new model, Kimi K3, and it's the largest open-source model ever built. It rivals the top US systems, and it's open, meaning anyone can use it. This is the same fear that hit the market with DeepSeek earlier, and it keeps coming back for a reason. The whole US AI and chip trade is priced on the assumption that we stay far ahead. Every time China closes the gap and drives the cost of these models down, that assumption takes a hit. And it's not just the model side. A Chinese memory maker called CXMT is about to list in Shanghai after an IPO that was more than 200 times oversubscribed, which stands up a fourth major competitor in the memory chips that go into every phone and server. More capacity and more competition in memory is a direct threat to the pricing power that has driven names like Micron and SanDisk higher. Then Meta poured gas on it. They said they're going to build their own compute in-house. That guts the AI cloud middlemen who were counting on Meta as a customer. CoreWeave and Nebius are down on the news. And it flows straight into the chips, because if the hyperscalers build their own, they need fewer of Nvidia's over time. You can see it in Nvidia this morning, falling through its 8 and 20-day. The United States has made an enormous financial bet on AI. This is a build-your-economy-around-one-thing event, closer to a wartime industrial push than a normal tech cycle. Trillions in capital, the biggest companies in the country, and a huge share of the market's gains are all riding on the same trade. So when China keeps closing the gap, the question isn't just whether Nvidia is expensive. It's whether the premium the whole country is paying for AI leadership still makes sense. None of this means the story is over. It means the market is being forced to ask a question it had been ignoring, and that's what a real stress test looks like. Watch how the mega caps hold up here, because that's where this either stabilizes or spreads.

Published Jul 17, 2026Updated Jul 17, 20265 min read
The AI hardware trade is cracking hard this morning, and it's worth understanding what actually set it off, because it isn't just profit-taking. The trigger came out of China. A company called Moonshot released a new model, Kimi K3, and it's the largest open-source model ever built. It rivals the top US systems, and it's open, meaning anyone can use it. This is the same fear that hit the market with DeepSeek earlier, and it keeps coming back for a reason. The whole US AI and chip trade is priced on the assumption that we stay far ahead. Every time China closes the gap and drives the cost of these models down, that assumption takes a hit. And it's not just the model side. A Chinese memory maker called CXMT is about to list in Shanghai after an IPO that was more than 200 times oversubscribed, which stands up a fourth major competitor in the memory chips that go into every phone and server. More capacity and more competition in memory is a direct threat to the pricing power that has driven names like Micron and SanDisk higher. Then Meta poured gas on it. They said they're going to build their own compute in-house. That guts the AI cloud middlemen who were counting on Meta as a customer. CoreWeave and Nebius are down on the news. And it flows straight into the chips, because if the hyperscalers build their own, they need fewer of Nvidia's over time. You can see it in Nvidia this morning, falling through its 8 and 20-day. The United States has made an enormous financial bet on AI. This is a build-your-economy-around-one-thing event, closer to a wartime industrial push than a normal tech cycle. Trillions in capital, the biggest companies in the country, and a huge share of the market's gains are all riding on the same trade. So when China keeps closing the gap, the question isn't just whether Nvidia is expensive. It's whether the premium the whole country is paying for AI leadership still makes sense. None of this means the story is over. It means the market is being forced to ask a question it had been ignoring, and that's what a real stress test looks like. Watch how the mega caps hold up here, because that's where this either stabilizes or spreads.
Article image 1