‘I told you so’: China’s chip sector is making breakthroughs—and U.S. export controls are the reasons why

An academic critic argues U.S. export controls have not stopped China’s chip and AI progress and may have accelerated domestic alternatives, though China’s chip self-sufficiency has not yet translated into commercial competitiveness. Nvidia’s data-center chip share in China fell to 34% in the first half of 2026, and its AI accelerator shipments dropped 44% year over year, according to IDC data cited by Reuters. Chinese firms are using workarounds including domestic chips, advanced packaging and 3D stacking, while the article notes unresolved questions about scalability and yields; a slowdown in chip scaling or AI demand could narrow the West’s lead.

Analysis

The key investable distinction is strategic substitution versus commercial substitution. China can reduce exposure to restricted chips without matching the cost, reliability, or scale of U.S.-based compute; domestic model performance and chip shipments alone do not establish competitive inference economics. That leaves near-term demand for frontier compute intact, but raises the probability that a growing share of Chinese demand becomes inaccessible to U.S. vendors and that export rules produce diminishing strategic returns.

For NVDA, this is a medium-term ceiling on China opportunity and a source of policy volatility—not, on this evidence, a reason to extrapolate a broad collapse in AI demand. AMD may face the same policy uncertainty, while TSM can still participate in global accelerator demand across vendors; its Taiwan concentration remains a distinct tail risk. ASML’s EUV bottleneck retains strategic value, though a longer-lived Chinese workaround effort could gradually shift investment toward packaging and system-level innovation rather than frontier lithography. The CXMT architecture claim is not independently validated: qualification, yield, volume, and customer adoption matter more to MU and SKHY than a launch announcement.

Over days, expect headline-driven dispersion around controls and China sales. Over 1–3 months, watch company disclosures for China revenue, export-license changes, and evidence of domestic-chip deployment at scale. Over 6–18 months, the central risk is not simply Chinese technical progress but a split market that limits U.S. vendors’ addressable demand while preserving their advantage elsewhere. The article’s strongest contrarian point is that controls may accelerate self-sufficiency; its weakest inference is treating technical progress as proof of commercially competitive serving costs.

Key Decisions for Investors

• Consider a modest, catalyst-driven relative-value position: long TSM versus short NVDA if export-control headlines drive NVDA-specific underperformance while broad AI infrastructure demand remains firm. Enter only after relative weakness is confirmed; keep the position small given TSM’s Taiwan tail risk and its exposure to NVDA’s supply chain. Reassess if TSM order commentary weakens or NVDA demonstrates durable China access under revised rules.
• Do not short MU or SKHY on CXMT’s architecture announcement. Treat it as a watch item; require independently verifiable yield, production volume, and customer qualification before underwriting share loss. A credible ramp plus pricing pressure would strengthen the bearish case; failed qualification or limited output would falsify it.
• Track NVDA and AMD China-related disclosures and U.S./allied licensing decisions over the next 1–3 months. A further tightening that reduces addressable sales supports relative underperformance; stable rules, meaningful licensed demand, or evidence that Chinese deployments remain commercially uneconomic weakens that thesis.
• Avoid treating claims about Chinese models or chip capabilities as proof of a near-term global AI-demand reversal. The thesis would change if hyperscaler spending or accelerator orders broadly roll over, not merely if Chinese buyers substitute domestic hardware.