Does China Still Want Nvidia Chips?

The Unwanted Chip

The Trump administration just engineered a controversial deal to let Nvidia resume selling its less powerful AI chips to China, complete with a novel 15% tax for the US Treasury. For a brief moment, it looked like a transactional win-win: Nvidia would regain access to a massive market, Chinese tech giants would get sought-after US hardware, and the US government would get a cut.

The only problem? Beijing is now telling Chinese companies not to buy them.

This is a bizarre turn of events, but it’s not just a matter of political pride. The uncomfortable truth for Washington is that the downgraded, US-approved H20 chip is simply not that competitive anymore. China is rejecting the H20 not just because it wants to build its own tech, but because it already has a comparable domestic alternative. Huawei’s new Ascend P910C accelerator, by most metrics, outperforms the H20, boasting more than twice the floating-point performance.

This technological reality is buttressed by a powerful strategic imperative. Even if the chips were equal, Beijing has deep-seated fears about dependency and national security. The push to "buy domestic" is being driven by concerns that US hardware could come with built-in backdoors. As Bloomberg reports:

Chinese officials are worried that Nvidia chips could have location-tracking and remote-shutdown capabilities — a suggestion that Nvidia has vehemently denied. Trump officials are actively exploring whether location tracking could be used to help curtail suspected smuggling of restricted components into China, and lawmakers have introduced a bill that would require location verification for advanced AI chips.

When the domestic chip is both more secure and more powerful, the choice for Beijing becomes simple. This reveals that the real game isn't about the H20 at all. It's about whether Huawei's massive, rack-scale systems can compete with Nvidia's top-tier Blackwell products that remain banned from China.

Huawei's approach is a brute-force marvel of engineering, born of necessity. Since US sanctions prevent it from building a single chip as powerful and efficient as Nvidia's best, it has adopted a different strategy: compensate with sheer numbers. Its CloudMatrix system connects more than five times as many accelerators as Nvidia's top-tier rack and takes up 16 times the floor space. The result is a power-hungry and expensive behemoth that can, through sheer scale, achieve comparable performance on paper.

The episode highlights the fundamental limits of US export controls. The Trump administration thought it was setting a clever trap by allowing a nerfed, older-generation chip into China to keep them dependent. But China is sidestepping the trap entirely. Its domestic alternative forces a difficult calculus on its own tech giants: while Huawei's system offers superior raw performance to what the US is now selling, it comes at a significantly higher cost in both price and power. Chinese tech firms are now caught between their government's strategic imperative to "buy domestic" and their own commercial imperative to build the most efficient and cost-effective AI infrastructure to win their own hyper-competitive local market. It is a costly, generational bet on self-reliance, and one that is set to reshape the global technology landscape.

The Platform Tenant

The modern tech landscape has a peculiar tenancy rule: even its kings are sometimes just renters in someone else's castle. This week, Elon Musk, proprietor of the digital fiefdom known as X, began publicly railing against the terms of his lease in Apple's walled garden. He accused the company of unfairly rigging its App Store to favor OpenAI's ChatGPT and threatened legal action over what he called an "antitrust violation." The public spat is more than just another billionaire tweetstorm; it is the opening shot in a new and critical conflict over who will be the gatekeepers of the AI era.

The fight is not merely about which app gets to be number one. It is about who controls the primary distribution channels for AI on the more than two billion devices that run our lives. The old power brokers of the mobile age are now in a position to dictate the winners of the next technological revolution. Musk's core accusation is that Apple is using its platform power to give its chosen partner, OpenAI, an unfair advantage. As he complained on X:

Apple is behaving in a manner that makes it impossible for any AI company besides OpenAI to reach #1 in the App Store, which is an unequivocal antitrust violation.

xAI will take immediate legal action..

One way to think about this is a classic platform power play. Apple controls the App Store, the sole gateway for software distribution on the iPhone. For over a decade, the company has wielded this power to its advantage. Now, it is extending that gatekeeper role into the AI era. Apple's deep partnership with OpenAI effectively anoints it as the default, trusted AI provider within the world's most valuable tech ecosystem. For a user downloading their first AI app, the one blessed by Apple is the obvious choice. This is not just a recommendation; it is a powerful exercise of platform control.

Musk's complaint signals that the battle for AI dominance is quickly evolving beyond who has the smartest model into a classic platform war, reminiscent of the PC and mobile eras. The new chokepoints of the AI industry are becoming clear: operating system integration, hardware manufacturing, and, as this episode makes clear, the app stores that serve as the final marketplace.

There is, of course, a final layer of irony here. Musk, as the owner of X, is a platform operator who has himself been accused of manipulating his own "town square" to his own benefit. But his complaint underscores a fundamental reality of the modern tech landscape: even the most powerful tech lords are subject to the rules of someone else's walled garden. The outcome of this fight will not just determine the fortunes of Musk and Altman, but will set the rules for the entire competitive landscape of AI for the next decade.

The Scoreboard

• Semiconductor: SK Hynix expects AI memory market to grow 30% a year to 2030 (Reuters)
• Semiconductor: Trump flip-flops on Intel CEO, calls him ‘success’ days after demanding resignation (CNBC)
• Software: GitHub CEO to Leave Microsoft Amid Steep AI Competition (Bloomberg)
• E-commerce: StubHub IPO is back on for September after ticketing company delayed plans on tariff concerns (CNBC)