It’s 2 a.m. in the outskirts of Amsterdam, and the ASML campus is still lit. Not because orders are flooding in—but because engineers are obsessively testing a single, unsettling hypothesis: if you can build chips with 3nm-equivalent performance without an EUV machine, then what exactly are we selling? Machines—or faith in Moore’s Law?
This isn’t science fiction. Last month, Huawei quietly unveiled its “LogicFolding” architecture—a method that vertically stacks 2D logic circuits using advanced packaging and heterogeneous integration to approach 3nm-level performance. No ASML EUV. No TSMC foundry access. Instead, it leveraged SMIC’s N+2 process, in-house EDA tools, and a painstaking, almost artisanal yield optimization loop. It sounds like using an abacus to outcompute a supercomputer. Yet somehow, it works.
I’ve heard countless “indigenous substitution” pledges over two decades in this industry. This time feels different. Huawei isn’t shouting slogans; it’s rewriting the foundational logic of chip design. While the world obsesses over GAA transistors and High-NA EUV delays, Huawei sidesteps the dead end of transistor scaling altogether—shifting the battlefield from the silicon plane into three-dimensional space. This isn’t retreat. It’s strategic elevation. Like how iPhone didn’t improve physical keypads—it erased them entirely.
But here’s the real question: if performance can be approximated through stacking, how irreplaceable is ASML’s EUV really? Consider this: a single High-NA EUV costs over €350 million, takes three years to deliver, and requires maintenance rivaling a small nuclear plant. Huawei’s approach may suffer from lower yields and higher power—for now—but it’s sovereign, rapidly iterative, and crucially, license-free.
That’s Washington’s nightmare scenario. Sanctions were meant to strangle. Instead, they forced a tracheotomy. And the irony? If this new paradigm matures, it could undermine the West’s monopoly pricing on leading-edge nodes. TSMC’s recent quiet pullback on 3nm capacity expansion hints at more than just AI demand volatility.
Then there’s NVIDIA, standing right in the storm’s eye. On one side, Blackwell chips roar ahead. On the other, its China-tailored H20 stumbles in adoption. Jensen Huang bets that AI compute will always be scarce. But what if Huawei’s LogicFolding paired with Ascend 910B delivers 80% of A100 performance at 30% of the cost for inference workloads? Will enterprises still pay full price for unshackled silicon?
Don’t underestimate that “80/30 rule.” In business, performance isn’t absolute—it’s the optimal point on a cost-efficiency curve. Forced by sanctions to accept subpar solutions, Chinese customers are discovering something unsettling: you don’t actually need top-tier FLOPS to run viable large models. Like how the pandemic revealed remote work could be just as productive—once habits shift, they rarely revert.
ASML CEO Peter Wennink once said, “Technology has no borders.” Reality disagrees. Machines face export controls. Talent hits visa walls. Even EDA licenses bow to White House whims. Huawei’s move proves true technological sovereignty isn’t about owning the most advanced tool—it’s about building system resilience that doesn’t hinge on any single node.
This contest has long transcended semiconductors. It’s about who defines “advanced,” who controls the narrative of progress. The U.S. hoped to freeze China’s tech timeline through blockade. History suggests otherwise: containment breeds more resilient innovation paths. The USSR, denied computers, pioneered analog computing systems. Japan, restricted from DRAM exports, seized control of materials and equipment upstream.
Now it’s China’s turn. And Huawei? It’s merely the first to crack the wall.
So back to that haunting question in the ASML lab: if you can match 3nm without EUV, what are you really selling? A machine—or the last moat defending Western technological hegemony? And when that moat dries up, who becomes the new ferryman?