In May 2026, at the Shanghai International Semiconductor Industry Conference, Huawei unveiled a chip fabrication method that made industry analysts do a double-take: a technique for stacking circuits to improve performance without depending on the most advanced Western lithography equipment. No EUV (Extreme Ultraviolet) machine required. It was a direct response to years of US export controls designed to keep China from accessing cutting-edge chip manufacturing technology.

This single announcement encapsulates the state of the global semiconductor race in 2026: China is not waiting for access to the best tools. It is finding alternative paths. And the question is no longer "Can China compete in semiconductors?" but "How quickly can China narrow a gap that still matters enormously for AI, cloud computing, and wider technological power?"

The semiconductor industry is the foundation of modern technology. Every smartphone, data center, electric vehicle, and AI model depends on advanced chips. The country that controls chip manufacturing controls the pace of technological progress for everything downstream. Here's where the major players actually stand—and why the race is closer than most people realize.

The Current Landscape: Who Leads Where

The global semiconductor industry is not a single competition but multiple overlapping races, each with different leaders:

Chip Design: China Is Closing the Gap Fast

This is where China has made its strongest gains. US export controls imposed in 2022 pushed Chinese firms toward domestic alternatives, and Beijing encouraged that shift even where foreign chips remained legally available. According to multiple industry analyses, domestic designers are now expected to account for the bulk of China's spending on AI chips in 2026, helped by a deep bench of engineers—including many with prior experience at NVIDIA and AMD.

Key players in Chinese chip design:

  • Huawei (HiSilicon): The Ascend 910C AI training chip delivers 1.5 PFLOPS of FP16 performance at 350W. The Atlas 350 inference card delivers 2.87x the performance of NVIDIA's H20 at FP4 precision. Huawei's CloudMatrix system links hundreds of Ascend processors to compensate for weaker individual chips through scale and integration.
  • Alibaba (T-Head): Designing custom AI chips for its cloud infrastructure. Alibaba is reportedly moving to split out its chip design unit as a separate business, signaling investor confidence in its commercial viability.
  • Baidu: Also reportedly spinning out its chip design division. Kunlun AI chips power Baidu's ERNIE model training and inference.
  • Cambricon: A dedicated AI chip company expanding in the protected domestic market.
  • Loongson (Godson): The Loongson 3A6000 uses the self-developed LoongArch instruction set and achieves SPEC CPU 2006 scores above 40—reaching international mid-range desktop CPU levels.

Meanwhile, the US leads in chip design through NVIDIA, AMD, Intel, Qualcomm, and Apple. But the gap in design capability—particularly for AI-specific chips—is narrowing faster than most Western analysts predicted five years ago.

Chip Manufacturing: The Decisive Bottleneck

This is where the battle is being won or lost. China still lacks reliable access to the most advanced lithography tools. ASML, the Dutch company that holds a monopoly on EUV machines, has confirmed that no EUV systems were shipped to China and has denied reports suggesting otherwise.

Without EUV, Chinese foundries rely on older DUV (Deep Ultraviolet) equipment and workarounds:

  • Multi-patterning: Using DUV machines multiple times on the same wafer to achieve finer features. This works but is slower, more expensive, and yields fewer good chips per wafer.
  • Advanced packaging: Huawei's chip stacking technique—bonding multiple less-advanced chips together to achieve performance comparable to a single advanced chip. This is the approach that turned heads in Shanghai.
  • Nanoimprint lithography: Companies like Prinano claim to produce 8-inch wafers for photonic chips without conventional DUV tools. Promising for niche applications but not a solution for mass-producing leading-edge logic chips.

The result: China's most advanced mass-produced chips are at roughly the 7nm node, while TSMC and Samsung are producing at 3nm and moving toward 2nm. That is a meaningful gap in raw transistor density, but Chinese companies are actively working to compensate through architecture, packaging, and system-level optimization.

Equipment imports tell part of the story: China's imports of chipmaking equipment have tripled since 2019, as companies stock up on available tools and domestic equipment makers become more capable. But the decisive EUV technology remains out of reach, and analysts say a domestic EUV equivalent is still years away.

AI Chips: Where the Race Gets Interesting

The AI chip market is where semiconductor self-sufficiency matters most in 2026, because AI training and inference are the primary drivers of chip demand growth. Here's how the key players compare:

  • NVIDIA (US): Still the global leader. H100 and B200 GPUs dominate AI training worldwide. The CUDA software ecosystem creates enormous switching costs.
  • Huawei Ascend (China): The 910C is being deployed in 10,000+ card training clusters at Baidu, Tencent, and Alibaba. Annual production capacity has reached 500,000 units. The Atlas 350 inference card is the only mass-produced chip supporting FP4 low-precision inference, delivering 2.87x NVIDIA H20's performance at that precision level.
  • AMD (US): MI300X competitive in training, gaining market share.
  • Google TPU (US): Custom silicon for internal workloads, not widely available externally.

The critical insight: Chinese developers are optimizing AI models specifically for Huawei silicon, using lower-precision formats like FP8 and FP4 to extract more performance from available hardware. As one industry analyst noted, "the best Chinese chips still lag behind NVIDIA's older flagship products, and most domestic hardware is being used for inference rather than the far more demanding task of training frontier AI systems." But the gap is closing—and for inference workloads, which represent the majority of deployed AI, Chinese chips are increasingly competitive.

The Self-Sufficiency Scorecard

Let's assess China's progress toward semiconductor self-sufficiency across key categories:

Fully Mature (Ready for Mass Deployment):

  • Embedded MCU chips (1:1 replacement for STM32, 30-50% cheaper)
  • Industrial interface and analog chips (70%+ domestic substitution rate)
  • Desktop and server operating systems (large-scale deployment in government and finance)
  • Distributed databases and middleware (commercial use in financial core systems)
  • Ascend AI training and inference chips (deployed in 10,000+ card clusters)

Rapidly Advancing (Approaching International Parity):

  • 7nm manufacturing processes (14nm automotive-grade in mass production, 7nm in active development)
  • High-end server CPUs (Kunpeng 930 with 64 cores at 3.0GHz)
  • High-end EDA tools (domestic tools progressively covering digital and physical design)
  • Memory chips (ChangXin Memory Technologies preparing for IPO)

Still Requiring Breakthroughs:

  • Sub-7nm lithography equipment and core components (domestic alternatives still in development)
  • High-end industrial software (CAD/CAE/EDA for advanced design flows)
  • High-precision sensors (high-accuracy, high-stability chips)

How China's Approach Differs from the West

The semiconductor race is not just about technology—it is about fundamentally different approaches to solving the same problem:

US Approach: Control the Bottleneck

The United States has focused on controlling the critical chokepoint—EUV lithography and advanced chip manufacturing through alliances with the Netherlands (ASML), Japan (Tokyo Electron), and South Korea (Samsung, SK Hynix). The strategy is to maintain an unbridgeable gap in manufacturing technology, ensuring that any country subject to US export controls cannot access the most advanced chips.

This approach has been effective in the short term but carries risks. It incentivizes China to develop alternatives, and every year of restrictions increases the economic and strategic motivation to succeed. The US has also restricted AI model access—limiting GPT-5.6 and other frontier models to approved partners—which further accelerates China's drive for independence.

China Approach: Build a Parallel Stack

China's strategy is to build an entire domestic technology stack from the ground up—not just chips, but the software, tools, and systems that go with them. The result in 2026 is a surprisingly complete ecosystem:

  • Chip architecture: LoongArch (Loongson), RISC-V variants (multiple companies), ARM-based Kunpeng (Huawei)
  • AI chips: Ascend (Huawei), Kunlun (Baidu), custom silicon at Alibaba and Tencent
  • Operating systems: openEuler (Huawei), Kylin, UOS for servers; HarmonyOS for mobile/IoT
  • AI frameworks: MindSpore (Huawei), PaddlePaddle (Baidu), competing with TensorFlow and PyTorch
  • Database: openGauss (Huawei), OceanBase (Alibaba), TiDB (PingCAP)
  • EDA tools: Empyrean, Primarius, X-EPIC progressively covering design flows

This parallel stack is not yet as mature as the Western ecosystem, but it exists—and it is being deployed at scale in government, finance, telecommunications, and increasingly in commercial applications. The bet is that a complete domestic stack, even if individually weaker in some components, is more resilient than a stack dependent on foreign technology.

EU Approach: Regulate and Subsidize

The European Union has taken a different path: heavy regulation (the EU Chips Act) combined with subsidies for domestic manufacturing. TSMC is building a fab in Germany; Intel has expanded in Ireland and Germany. The goal is to reduce dependency on Asian manufacturing. But Europe has virtually no domestic chip design companies competing at the frontier, and its strategy relies heavily on attracting foreign investment rather than building indigenous capability.

South Korea and Japan: The Critical Suppliers

South Korea (Samsung, SK Hynix) and Japan (Tokyo Electron, Disco, Screen) occupy a unique position. They possess world-class manufacturing and equipment capabilities but are politically aligned with the US. Their technology is critical to both sides of the semiconductor race, making them key swing players. Samsung's 3nm GAA (Gate-All-Around) manufacturing process is technically ahead of TSMC's current node, but yield challenges have limited its commercial impact.

The AI Training Bottleneck: Where the Gap Still Matters Most

The most consequential semiconductor gap between China and the US is in AI training hardware. Training frontier AI models like GPT-5, Claude, or Qwen3.7 requires massive GPU clusters—often 10,000+ cards running for weeks or months. NVIDIA's CUDA ecosystem creates enormous network effects: most AI researchers train on NVIDIA hardware, optimize for CUDA, and default to NVIDIA for production deployments.

China's alternative—Huawei Ascend clusters with the MindSpore framework—is functional but less mature. Companies like Baidu, Alibaba, and Tencent have deployed Ascend-based training clusters, but the ecosystem of tools, libraries, and developer expertise is still catching up. Most domestic hardware is being used for inference (running trained models) rather than training (building new models from scratch).

This matters because the countries that control AI training infrastructure shape the direction of AI development. If China cannot train frontier models as efficiently as the US, it risks falling behind in the next generation of AI capabilities. However, the efficiency-first approach—using FP8 precision, system-level optimization, and architectural innovation—could prove surprisingly effective over time, especially as models become more optimized and less brute-force.

What the Market Thinks

Investor sentiment tells an important story. Hong Kong-listed semiconductor stocks have rallied sharply over the past year. Alibaba and Baidu are moving to spin out their chip design units as independent businesses. ChangXin Memory Technologies is preparing an IPO. These moves signal that investors increasingly see China's domestic chip industry as a serious commercial bet, not just a government-subsidized vanity project.

Meanwhile, ASML's stock has been volatile as investors weigh the impact of China restrictions on the company's revenue. The tension between geopolitical strategy and commercial reality creates ongoing uncertainty for the entire global semiconductor supply chain.

The Bottom Line: A Race with No Clear Finish Line

The semiconductor self-sufficiency race between China and the rest of the world does not have a simple leader. It is a multi-dimensional competition:

  • Chip design: US leads, but China is closing fast, especially for AI-specific chips
  • Manufacturing equipment: US and allies dominate (ASML, Applied Materials, Lam Research), but China is building alternatives
  • Advanced manufacturing: TSMC and Samsung lead at 3nm/2nm; China is at 7nm and improving
  • AI chips: NVIDIA dominates training; Huawei is increasingly competitive for inference
  • Legacy chips: China is rapidly gaining market share in mature-node chips used in cars, appliances, and industrial equipment
  • Full-stack ecosystem: China is the only country building a complete parallel stack from chips to AI frameworks; this takes time but reduces vulnerability

China is edging closer to self-sufficiency but remains short of the capability needed to match the West at the frontier. The gap in manufacturing technology is real and significant—perhaps 5-7 years by most estimates. But the trajectory is what matters, and China's trajectory is upward while the West's policy of restriction is creating the very competitor it seeks to prevent.

For the global technology industry, the implication is clear: the era of a single dominant semiconductor supply chain is ending. What emerges will be two partially overlapping ecosystems, each with different strengths and different vulnerabilities. The companies and countries that can operate effectively in both will have the greatest advantage. Those that bet everything on one side may find themselves exposed when the other catches up.