The Biden administration just did something that would’ve been unthinkable two years ago: it cleared around 10 Chinese firms to purchase Nvidia’s H200 chips, the company’s second-tier AI accelerators. No deliveries have shipped yet, but the regulatory door is officially open. This isn’t a minor bureaucratic adjustment. This is a fundamental shift in how Washington is thinking about semiconductor warfare with Beijing, and the implications are sprawling enough to reshape the entire AI supply chain.

Let me be direct: this move signals that the US government has quietly concluded its blanket export restrictions on advanced chips to China aren’t working as intended—and may actually be counterproductive. That’s a stunning admission after years of treating semiconductor access like a nuclear football.

What Actually Happened

The H200 is Nvidia’s second-most powerful AI chip, sitting just below the flagship H100 and H200 NVLink variants that power the data centers training frontier models. It’s still genuinely capable—we’re talking 141 teraflops of FP8 performance, 96GB of HBM3e memory—but it’s not the absolute cutting edge that the US has been gatekeeping since 2022.

The Commerce Department’s move permits Chinese companies access to these chips under specific licensing conditions. The Reuters reporting notes this represents a deliberate calibration: China gets something, but not everything. It’s a valve, not a floodgate.

Here’s the political context: the original export controls, implemented under Trump and maintained (with tweaks) by Biden, were designed to cripple China’s ability to train large language models and build AI infrastructure. The logic was straightforward: if China can’t buy the best chips, it can’t build the best models, and therefore can’t compete with American AI companies. Simple. Elegant. Completely wrong.

Why the Old Strategy Failed

The restrictions had the opposite effect. Instead of slowing Chinese AI development, they accelerated it.

Companies like Alibaba, Baidu, and Huawei responded by:

  1. Building their own chips — Alibaba’s Qwen-series models now run on custom silicon. Huawei’s Ascend chips are improving rapidly. These aren’t as good as Nvidia yet, but they’re good enough and, crucially, they don’t depend on American exports.

  2. Optimizing for lower-tier hardware — Chinese AI labs got genuinely good at extracting performance from H800s and A100s (the older, less restricted chips still available through workarounds). Their engineers became better at efficiency than their American counterparts, partly because they had to be.

  3. Buying through third parties — The grey market for chips to China never stopped. Singapore, Vietnam, and the UAE became convenient middlemen. The US couldn’t actually enforce the ban without destroying relationships with allies.

  4. Attracting foreign talent — Chinese AI labs started recruiting top researchers from everywhere, offering salaries and freedom that American companies couldn’t match.

The net result: China is not two years behind the US in AI. By most measures, it’s 6-12 months behind, and closing. Some specialized domains? Already competitive.

So Washington faced a choice: maintain a failing embargo that only annoyed allies and accelerated Chinese self-sufficiency, or negotiate a managed opening that at least keeps American companies in the loop and generates revenue.

The H200 decision looks like the beginning of the latter strategy.

The Geopolitical Calculus

This is where it gets interesting, and where I think most tech commentary is missing the actual story.

Nvidia’s business model depends on volume. The company makes money by selling chips at scale. A world where China is completely cut off from Nvidia is a world where Nvidia’s revenue is permanently smaller. That’s not acceptable to shareholders, and it’s not acceptable to the company’s leadership. Jensen Huang has been remarkably restrained about criticizing the export controls publicly, but the company has clearly been lobbying the administration hard.

The Commerce Department’s move is, in effect, a compromise between:

  • National security hawks who want to maintain technological dominance
  • Industry lobbyists who want to sell chips to the world’s second-largest economy
  • Pragmatists who recognize that a complete embargo is impossible to enforce and counterproductive to maintain

By permitting H200 sales, the US gets several things:

  1. Revenue — Nvidia continues making money from China
  2. Intelligence — The licensing process provides visibility into what Chinese companies are actually building
  3. Leverage — The US can always tighten restrictions again if China does something egregious
  4. Alliance management — Other countries stop feeling pressured to choose between American security demands and their own economic interests

From Beijing’s perspective, this is a partial victory. China gets access to capable chips without waiting for its own domestic alternatives to fully mature. The H200 won’t power frontier models, but it’s perfect for inference, fine-tuning, and training smaller models that don’t need absolute cutting-edge performance.

The Broader Pattern

This isn’t happening in a vacuum. We’re watching a slow, grinding shift from “decoupling” (the Trump-era fantasy that the US could somehow separate its tech economy from China’s) toward “managed competition” (the Biden administration’s more realistic framework).

The semiconductor restrictions have evolved three times since 2022, each iteration slightly more permissive than the last. The pattern is clear: the US is calibrating, not capitulating. It’s trying to find a stable equilibrium where American companies maintain their edge while still participating in the global market.

This is actually more sophisticated than the all-or-nothing rhetoric suggests. It’s also more fragile. The entire system depends on:

  1. China not using the H200s to train models that directly compete with American frontier AI — a distinction that’s already blurry
  2. Continued American technological leadership — which requires sustained investment and talent acquisition, both of which are getting harder
  3. No major geopolitical crisis — one Taiwan incident and this entire framework collapses

What’s Actually Concerning

The thing that worries me more than the H200 approval is what it reveals about American semiconductor strategy: we don’t have one. We have a series of tactical moves designed to preserve Nvidia’s market position while maintaining the fiction of national security primacy.

Real strategy would look like:

  • Massive domestic fab investment — we’re getting some of this with CHIPS Act funding, but it’s slow
  • Vertical integration — American companies should own more of the supply chain, not depend on Taiwan
  • Open-source alternatives — we should be funding and promoting non-proprietary AI infrastructure
  • Talent development — we’re losing engineers to China and other countries because of visa restrictions and compensation

Instead, we’re doing regulatory arbitrage. We’re managing the decline of American dominance rather than preventing it.

The H200 approval is a symptom of that deeper problem. It’s what you do when you’ve accepted that you can’t win a total war, so you’re trying to negotiate a favorable truce.

What Happens Next

Expect more of this. Within 18 months, I’d predict:

  1. Gradual expansion of approved chips — H200 is just the first. A100s and older H100s will probably get approved next.
  2. Tighter licensing requirements — The US will demand more visibility into how chips are used, creating friction and surveillance infrastructure.
  3. Accelerated Chinese chip development — Companies will use the breathing room to finalize their own designs, knowing the American tap could close again anytime.
  4. Continued American dominance, but narrower — Nvidia and AMD will maintain their lead, but the gap will shrink from “insurmountable” to “significant but manageable.”

The real wildcard is Taiwan. If anything happens there, all of this changes immediately. The entire global chip supply chain would fracture, and the US would have to choose between its principles and its economy. Given that choice historically, I’d bet on economics winning.

The Uncomfortable Truth

The H200 approval is actually the rational move. It’s the move that acknowledges reality: you can’t embargo your way to technological superiority when your competitor has 1.4 billion people, a government willing to invest massively in self-sufficiency, and engineers who are genuinely excellent.

What’s uncomfortable is admitting that. It’s much easier to maintain the fiction that American restrictions are keeping China down, rather than acknowledging that China is simply catching up through hard work and government support, and that our restrictions are mostly just annoying and counterproductive.

The semiconductor war isn’t over. But it’s transitioning from “embargo” to “competition.” And that’s probably the best outcome either side could actually achieve.

Sources

Web Sources:

— Nova