During the recent US-China summit (May 13-15) in Beijing, many analysts wondered who actually held the Trump cards, pardon the pun. An industry-by-industry review revealed a sobering reality. China has closed all critical gaps across key sectors.
For nearly a decade, the prevailing strategy in Washington D.C. rested on a deceptively simple premise. Restrict China’s access to advanced semiconductors, and thereby choke its technological ascent. Export controls, sweeping sanctions, and entity blacklists were deployed with precision, and they were all designed to cement a permanent technological ceiling. The underlying assumption was that without American technology, Chinese innovation would plateau, locking it generations behind the West.
As of spring 2026, that assumption is not merely outdated. It is dangerously obsolete. China no longer needs US advanced chips in the way it did just two years ago.
Consider the resigned admission from Nvidia’s own chief executive. In May 2026, Jensen Huang sat for an interview with the Special Competitive Studies Project, a US government-backed advisory group. His words sent a chill through Wall Street and Silicon Valley alike. “Today, in China,” Huang said, “we have now dropped to zero.” He was referring to Nvidia’s share of China’s AI accelerator market. Two years ago, that share stood above 80 percent. At its peak, Nvidia commanded more than 90 percent of the Chinese market.
The US tech containment strategy simply backfired. The Chinese approach of forced catch-up has given way to strategic bypass. Across the vast majority of the global semiconductor market, China has already achieved functional independence. What remains is not a question of whether Beijing will reach self-sufficiency, but how much of the Western semiconductor industry will survive once it does. The containment experiment has failed. The only variable left is the pace of the fallout.
To understand how this happened, we must look past the obsession with cutting-edge node counts and examine the industry as it actually operates, in layers. The first layer, mature nodes at 28 nanometers and above, accounts for roughly 80 percent of global chip demand. These are the unsung workhorses powering electric vehicles, industrial automation, power grids, and consumer electronics. China has already achieved near-total self-sufficiency here. By 2025, domestic fabs were running mature-node production at scale.
With Beijing’s mandate that new facilities source 70 percent of their equipment domestically by 2027, this independence will soon be structurally locked in. Western firms still selling products into this segment do so only because China tolerates the competition, not because it requires them.
The second layer, advanced memory, has seen a similar decoupling. High-volume memory chips like 3D NAND and DDR5 are the lifeblood of the digital economy, and Chinese champions YMTC and CXMT have systematically dismantled Western dominance. Wafer capacities are expanding, layer counts are pushing past 232, and domestic manufacturing equipment now comprises over half of the tooling in select memory lines. The gap with South Korea and the United States is no longer measured in decades, but in years.
Even the third layer, the leading-edge logic sub-7 nanometer frontier once deemed untouchable, has yielded to Chinese engineering. SMIC, China’s flagship foundry, has achieved mass production of 5-nanometer-class chips using older deep ultraviolet lithography. Yields are lower and costs are higher than those of global giant TSMC, but for the domestic security infrastructure and AI deployment, good enough is entirely sufficient. The system works. More importantly, it is scaling.
The most profound shift of the past two years, however, did not happen in fabrication plants. It occurred in AI deployment. Training foundational models undeniably requires the most advanced silicon, and here, Chinese labs still trail Nvidia’s flagship offerings. But inference, the actual large-scale application of AI models, represents a vastly larger market. And on that front, China has already won. The turning point arrived with DeepSeek V4, a model explicitly optimised to run on Huawei’s Ascend 950 accelerators. Almost overnight, the performance gap for inference workloads evaporated. Major Chinese tech firms pivoted to domestic hardware not out of patriotic necessity, but because the economics and strategic autonomy made it the superior choice. Nvidia has effectively lost the Chinese AI inference market, and there is no viable path back.
This brings us to a critical asymmetry that Washington has largely ignored, the balance of vulnerability. Does China need US chips more than the US needs China’s rare earths? The answer reveals a lopsided reality. For China, restricted chip access creates a technological ceiling, a manageable delay that slows AI training but leaves the broader economy intact. For the United States, a cutoff in heavy rare earths triggers immediate industrial paralysis. Electric vehicle production stalls, aerospace supply chains fracture, and defense manufacturing grinds to a halt within months. This vulnerability explains the frantic diplomacy behind closed doors.
So what did President Trump actually secure on rare earths during the May 2026 Beijing summit? The answer depends on which government you believe in. The White House announced that China had agreed to address US concerns over rare earth shortages, specifically naming yttrium, scandium, neodymium, and indium. Yet China’s own official readout of the summit made no mention of rare earths at all.
The US is discovering that rebuilding a rare earth supply chain is an exercise in climbing a mountain with a broken leg. Rare earth refining is not merely mechanical. It is a deeply complex industrial chemistry. China spent four decades mastering the intricate knowledge required to separate seventeen elements into 99.9999 percent pure compounds. It is the kind of arcane expertise that exists in the hands of engineers, not in patent filings. And then, there is the cost factor. A Chinese-made neodymium magnet still costs nearly half as much as its Western counterpart.
Without permanent, massive subsidies, no American company can compete. Projections now suggest China has a two-thirds chance of breaking the extreme ultraviolet lithography barrier by 2030, while the US has less than a thirty percent chance of achieving rare earth independence in the same window. China is leveraging its pain into capability. The US is merely hoping its pain disappears.
Paradoxically, the two- to three-year lag in frontier chip access has become one of China’s greatest strategic assets. Latecomer advantage theory is no longer an academic abstraction. Chinese researchers and engineers can observe which American AI architectures scale, which collapse under their own weight, and which represent dead ends. Such a wait-and-see approach has spawned genuine breakthroughs. Teams have developed mixture-of-experts models that slash compute requirements. They have secured industry leadership in four-bit quantization. They have also engineered novel attention mechanisms that outperform standard transformers. The list goes on and on.
This advantage extends beyond software. China’s manufacturing ecosystem is rapidly integrating physical AI, robotics, autonomous systems, embodied intelligence, where deployment speed and supply chain stability matter far more than raw benchmark scores. A deliberate, measured pace allows Beijing to optimize for inference efficiency. Engineers can identify failure modes before commercial rollout. They can also build complementary infrastructure tailored to proven architectures. The Nvidia restrictions inadvertently gifted Beijing a clean slate. Had Chinese tech giants remained tethered to American hardware, they would have been building their technological future on a foreign foundation.
Western analysts have consistently underestimated Chinese timelines. Predictions claimed domestic 28-nanometer lithography would take a decade. SMIC operationalized it by 2025. Experts declared 5-nanometer production without EUV impossible. It is running today. Forecasts placed a Chinese EUV prototype in the 2030s. Shenzhen’s test systems are already generating coherent light, with functional chips anticipated by 2028. These are not aspirational slogans from policy documents. They are funded, tracked milestones embedded in the 15th Five-Year Plan. By 2027, China targets 70 percent equipment localisation for mature fabs. By 2030, memory and leading-edge logic will approach comparable self-sufficiency. The machine is moving, and it is accelerating.
The United States designed its containment strategy around the expectation of perpetual Chinese dependency. That expectation has collapsed. The long game framing was always a projection of Western industrial rhythms onto a system that operates on a different clock entirely. China has already captured the domestic market, forced Western competitors into strategic retreat, constructed a parallel supply chain, and converted enforced lag into a structural advantage. The rare earths comparison makes the stakes unmistakable. The US cannot afford a cutoff, while China not only can endure one, but has already neutralised its strategic impact.
The question for Washington is no longer how to maintain Chinese dependency. It is how much of the American semiconductor ecosystem will remain competitive when China shifts from import substitution to full-scale global export. The answer, unfolding in real time, is far less than policymakers anticipated.
The end of dependency is not on the horizon. It has already arrived.
Purchase a copy of my novel — The Electric Reckoning.
(It encapsulates how this world works in a fictionalised form)
Share, restack or quote my posts.
No posts
Comments
Nothing yet. Say the first thing.
Sign in to join the conversation.