Why China’s EUV breakthrough is such a big deal

The breaking of the West’s AI containment strategy

In many ways, 2025 was a year of China playing catch-up in the AI race. This was expected. What wasn’t expected was the pace.

When I first covered DeepSeek’s release in The Spectator back in February, I described it as China’s Sputnik moment. A touch of editorial hyperbole — but the point was straightforward. DeepSeek demonstrated that there was another route to developing capable AI systems and getting them into a SOTA state that didn’t rely solely on compute, flipping the logic of the AI race on its head (and incidentally wiping out over $1 trillion in tech stocks in the process).

At the same time (and potentially more importantly), the decision to make it open source signalled something else entirely: that distribution, not just raw capability, had become a blue-ocean opportunity for Chinese expansion.

Now, with its breakthrough on extreme ultraviolet (EUV) lithography, China has breached the third element of the AI race that was widely assumed to be fully blockaded. Advanced chips — long the lynchpin of American AI strategy, first denied through export controls and later held up as potential leverage — are no longer quite as unreachable in the East as Washington believed.

This new development signals significant tailwinds for China in building its own rival stack.

Strategically, this is a disaster. Washington spent years arguing that controlling access to advanced chips would preserve a decisive advantage in artificial intelligence. That logic later evolved into a more transactional bet: that allowing limited sales of American hardware might keep China dependent on the US stack. On December 8th, President Trump authorised the sale of Nvidia’s H200 chips to “approved customers” in China, reportedly in exchange for a 25 per cent cut for the US Treasury. The assumption was clear — dependency could be managed, even monetised.

China’s EUV progress suggests that assumption was misplaced.

So what happened?

According to a Reuters investigation published yesterday, China has built a working prototype of an EUV lithography machine inside a secure facility in Shenzhen — the single most critical technology underpinning advanced semiconductor manufacturing. Crucially, the machine is not yet producing commercial chips. But it exists. It generates EUV light. And it fills nearly an entire factory floor.

EUV lithography is the one capability the West believed it had successfully ring-fenced. Only one company — ASML in the Netherlands — has ever mastered it. No EUV system has ever been sold to China. For years, US export controls were designed around a simple assumption: deny EUV, and you deny China the ability to produce cutting-edge chips at scale.

The Reuters reporting suggests that assumption no longer holds.

The prototype, completed in early 2025 and now undergoing testing, was reportedly built by a team of former ASML engineers who reverse-engineered key elements of the system. Many were recruited from abroad, some working under aliases, their involvement classified under national security. China supplemented this expertise by scavenging components from older lithography machines, sourcing parts through secondary markets, and leaning heavily on domestic research institutes to replicate what could no longer be imported.

The result is crude compared to ASML’s production-grade machines. Precision optics — particularly mirrors of the quality supplied by Germany’s Carl Zeiss — remain the largest technical obstacle. The machine has yet to produce working chips, and even those close to the project suggest 2030 is a more realistic target than the government’s official 2028 goal.

But that is beside the point. The strategic failure has already occurred. EUV was the chokepoint the West believed it had successfully locked behind export controls. Instead, it has become a delayed one.

What this episode reveals is not Chinese technical superiority, but Western overconfidence in denial as a strategy. Once a technology exists in the world — once it has been proven, deployed, and understood — preventing a determined, well-resourced country like China from recreating it becomes a question of time, not possibility. Export controls slowed China down. They did not stop it. In some respects, they likely accelerated Beijing’s push toward full-stack technological sovereignty.

What this means for AI policy in 2026 is a total rethink. Jensen Huang and others have repeatedly argued that the United States should continue selling advanced hardware into China, on the logic that dependence on American chips would slow the emergence of a rival stack. For a short while, this was plausible. But since Beijing banned domestic firms from relying on foreign hardware in September — and now with EUV in play — that strategy looks increasingly untenable.

In hindsight then, what seemed like a high-risk bet from the outside looks a lot less of a gamble at all. If Beijing can produce advanced chips domestically within the next five years, access to American hardware ceases to be leverage.

This reframes the next phase of competition. If chips are no longer the decisive bottleneck, energy infrastructure becomes the constraint that matters most. AI capability scales not just with silicon, but with power — grid capacity, generation, and the ability to sustain energy-intensive computation at national scale.

On that front, the balance already looks uncomfortable. China treats AI as a national industrial project and is building the power system to match it. Beijing already generates roughly twice as much electricity as the United States and continues to add capacity at a pace the West struggles to match. By contrast, much of America’s grid is ageing, congested, and stuck in permitting limbo — an increasingly serious liability in an era where compute demand is exploding.

On top of this, if open source has become the primary arena where adoption, standards, and developer loyalty are decided, then the United States cannot afford to treat it as a side effect of private-sector benevolence. It needs its own coherent open-source strategy — one that treats open weights, permissive licensing, and developer ecosystems as strategic assets rather than ideological risks.

That does not mean abandoning proprietary frontier research — in this regard the West still is best — but it means recognising that leadership now requires a parallel public arena where U.S. and allied models can circulate freely, be improved in the open, and become defaults across global developer workflows. Without that, America risks winning the frontier while losing the ecosystem — a position that will hinder any hopes for lasting power.

At this point, China is no longer just competing on models. It is positioning itself to win on energy infrastructure, technological distribution and diffusion (there’s a reason your Christmas toys may be loyal to the CCP), and now even core AI capabilities.

As David Sacks, Trump’s AI tsar, has repeatedly argued, winning the AI race means having the world built on America’s stack. Yesterday’s news will not have landed lightly on the White House lawn.