The South Shore Press
← Back to Business
Business

China's Open-Source AI Model Just Wiped Out Weeks of Chip Stock Gains

A Chinese startup's new Kimi K3 model rattled the AI trade this week, but the real story is how thin the market's confidence has become, not how good the model is

By Howard Roark
China's Open-Source AI Model Just Wiped Out Weeks of Chip Stock Gains
Credit: South China Morning Post

On Thursday afternoon, a Chinese AI lab called Moonshot released an open-weight model called Kimi K3, claiming performance that rivals the best commercial systems out of Silicon Valley. By Friday morning, chip stocks across three continents were in freefall. Japan's SoftBank fell nearly 10 percent. Korea's Kioxia halted trading after hitting its daily limit down 16 percent. Taiwan's semiconductor index slid into a technical correction. Nasdaq futures dropped more than 1.5 percent before the opening bell.

This is the third time in roughly eighteen months that a Chinese open-source model has triggered this exact reaction, and the pattern is worth understanding because it tells you more about Wall Street's positioning than about the technology itself. The model is genuinely capable, ranking third on independent evaluations and topping certain coding benchmarks. But nothing about its release should have surprised anyone paying attention. Chinese labs have been closing the gap with American frontier models for two years, each time at a fraction of the reported training cost. What moved markets wasn't the surprise of Chinese competence. It was that the trade built around American AI supremacy had grown so crowded, and so dependent on a specific narrative holding, that any crack in that narrative triggers an outsized reaction.

Consider the timing. The selloff landed in the same week that TSMC and ASML, the two companies that actually manufacture the chips underpinning the entire AI buildout, both beat earnings expectations and raised their forward guidance. TSMC lifted its 2026 revenue growth outlook to more than 40 percent and raised its capital spending plan to as much as $64 billion, up from a prior ceiling near $56 billion. That is about as unambiguous a signal of real, physical demand as a market gets. And investors sold the stock anyway, preferring to price in a future where a cheaper Chinese model erodes the pricing power of the entire compute supply chain, rather than a present where demand for that supply chain is accelerating.

There's a legitimate economic question buried in the panic: if capable AI models can be built and distributed for a fraction of the cost that American labs are spending, what does that mean for the trillion-dollar-plus capital expenditure plans hyperscalers have committed to over the next several years? It's a fair question. But it's also one that gets asked and re-litigated every time a new open-source model drops, and each time the answer so far has been that demand for compute keeps growing regardless, because capability gains get consumed by new applications almost as fast as they arrive. The bottleneck, per the same research houses now marking down chip stocks, isn't a shortage of AI talent or algorithms. It's power, land, and manufacturing capacity, exactly the physical constraints that a cheaper software model does nothing to relax.

For Long Island readers, the direct exposure to any single chip stock is likely limited. But the indirect exposure is not. Pension funds, 401(k) target-date portfolios, and the broader retirement savings of Suffolk County residents carry meaningful weight in mega-cap technology names that have driven a disproportionate share of market returns over the past two years. When that concentration unwinds even briefly, as it has this week, it moves account balances for people who never chose to make a bet on Chinese AI competition. The lesson isn't that the AI buildout is fake. The physical evidence, from chip manufacturers' order books to data center construction backlogs, says otherwise. The lesson is that a market this concentrated in a single narrative is fragile to any development that complicates that narrative, even when the underlying business fundamentals haven't changed at all.

Watch what happens in the next two weeks as more hyperscalers report earnings and update their 2027 capital spending plans. If those numbers keep climbing, as several already have, it will be hard to square continued panic about Chinese competition with continued acceleration in the very spending that competition is supposed to threaten. That contradiction, more than any single Chinese model release, is the thing actually worth watching.

You Might Also Be Interested In