Morgan Stanley expects Chinese companies to supply 86% of their country’s AI chip market by 2030, a market the bank estimates will be worth $67 billion. If the projection holds, it would mark one of the fastest semiconductor supply shifts in modern history, driven largely by a U.S. export-control campaign that was designed to slow Beijing’s AI ambitions but may be accelerating its push for self-reliance.
The forecast, circulated in a Morgan Stanley research note and reported by Reuters, lands at a moment when the gap between American restrictions and Chinese workarounds is becoming the central tension in global chip policy. Today, Chinese-designed processors account for a relatively small slice of the country’s AI accelerator consumption, with Nvidia’s A100 and H100 chips (and their various export-compliant derivatives) still powering much of China’s cloud and AI training infrastructure. Reaching 86% domestic supply in roughly five years would require Chinese firms to scale production, close a meaningful performance gap, and convince domestic buyers to switch, all while navigating a tightening regulatory net from Washington.
The export-control ratchet
The regulatory pressure behind this forecast is real and well documented. The U.S. Department of Commerce, through its Bureau of Industry and Security, has imposed three major rounds of semiconductor export restrictions targeting China since October 2022. Each round lowered the performance thresholds that trigger a license requirement and closed loopholes that had allowed chips to reach Chinese buyers through third countries.
The October 2022 rules set initial limits on chips above certain processing-power and interconnect-bandwidth benchmarks. A year later, the Commerce Department tightened those thresholds after Nvidia redesigned chips specifically to skirt the original cutoffs. A further expansion in late 2024 added new controls on high-bandwidth memory, additional chipmaking equipment, and a wider set of Chinese entities. The cumulative effect: virtually every cutting-edge AI chip and much of the equipment needed to manufacture one now requires a U.S. government license before it can be shipped to China.
These are not paper rules. Exporters must submit license applications through the Commerce Department’s SNAP-R portal, and violations carry criminal penalties, civil fines, and denial of future export privileges. The system is operational, continuously processing requests, and actively enforced, which means the pressure on Chinese buyers to find alternatives is binding, not theoretical.
Where Chinese chipmakers actually stand
The most visible evidence of Chinese progress is Huawei. Its HiSilicon subsidiary designed the Ascend 910B AI accelerator, which Chinese cloud providers including Baidu, Alibaba, and Tencent have begun deploying as a substitute for restricted Nvidia hardware. An upgraded version, the Ascend 910C, has been sampled to major customers, though independent benchmarks of its real-world training performance remain scarce.
On the manufacturing side, Semiconductor Manufacturing International Corporation (SMIC) demonstrated in 2023 that it could produce chips at a 7-nanometer-class process node, a feat that surprised Western analysts who had assumed China was years away from that capability without access to the latest extreme ultraviolet (EUV) lithography machines from the Netherlands-based ASML. SMIC achieved this using older deep ultraviolet (DUV) equipment, a workaround that is slower and more expensive but functional.
Still, most independent assessments place Chinese AI chips roughly two generations behind Nvidia’s current lineup. Nvidia’s H100 and its successor, the B200, are built on TSMC’s most advanced nodes with cutting-edge packaging technology that Chinese fabs have not yet replicated at scale. The gap matters because AI training workloads are extraordinarily sensitive to chip performance: a two-generation lag can translate into significantly higher energy costs, longer training times, and limitations on the size of models that can be built.
Beijing is spending heavily to close that gap. China’s national and provincial governments have channeled tens of billions of dollars into semiconductor subsidies through vehicles like the National Integrated Circuit Industry Investment Fund, commonly known as the “Big Fund.” A third phase of the Big Fund, announced in 2024 with a reported capitalization exceeding $47 billion, is the largest yet. But money alone has not historically been sufficient to leapfrog entrenched technological leads, and China’s chip industry still depends on imported equipment for many critical manufacturing steps.
Why the 86% number deserves scrutiny
Morgan Stanley’s projection is an analytical forecast, not a measurement. Investment banks build these estimates on models that incorporate assumptions about technology adoption curves, government policy trajectories, and competitive dynamics. Such forecasts are useful as directional signals, but they are not predictions in the scientific sense.
Several variables could push the actual outcome well above or below 86%. On the upside, a broader U.S. crackdown, say, a full embargo on all semiconductor equipment sales to China, could accelerate domestic substitution even faster than Morgan Stanley expects. On the downside, Chinese chipmakers could hit technical walls that slow their progress, or Beijing could shift subsidies toward other priorities. The $67 billion market-size estimate itself depends on assumptions about how quickly Chinese companies adopt AI across industries ranging from autonomous driving to financial services.
Enforcement gaps add another layer of uncertainty. Restricted chips have continued to reach China through gray-market channels and intermediary countries, a pattern documented in U.S. Department of Justice indictments and Commerce Department entity-list additions targeting diversion networks. If significant volumes of advanced foreign chips keep flowing into China despite the controls, domestic suppliers would face stiffer competition than the Morgan Stanley model may assume. Conversely, if Washington and allied governments in Japan, the Netherlands, and South Korea tighten enforcement further, the incentive to buy domestic would strengthen.
What this means for Nvidia, cloud giants, and everyone else
The financial stakes are immediate. Nvidia has disclosed that China-related revenue restrictions have already cost it billions in potential sales. Before the October 2022 controls, China accounted for roughly 25% of Nvidia’s data-center revenue; that share has since dropped, though the company has partially offset losses by selling different chip configurations to other markets. If Chinese buyers shift overwhelmingly to domestic alternatives by 2030, the lost revenue would be permanent, not cyclical.
For global cloud providers and hardware manufacturers, a bifurcated supply chain is the most likely practical consequence. Companies selling into both Western and Chinese markets would need to maintain separate product lines: one built on U.S.-allied chips and compliant with export rules, another designed around Chinese processors and standards. That duplication raises engineering costs, complicates supply logistics, and ultimately gets passed on to business customers and consumers.
Third-country markets are the next battleground. A more self-sufficient Chinese chip sector would not stay contained within China’s borders. Chinese AI chip companies, backed by state subsidies and battle-tested against domestic competition, could emerge as aggressive competitors in Southeast Asia, the Middle East, Africa, and Latin America, regions where governments are less aligned with Washington’s export-control framework and where price often matters more than geopolitical allegiance.
The policy paradox Washington has not resolved
For U.S. policymakers, the Morgan Stanley forecast crystallizes an uncomfortable question: are export controls slowing China’s AI capabilities in the near term while accelerating its long-term independence? The evidence so far suggests both things may be true simultaneously. Chinese AI labs have reported difficulties obtaining enough high-end chips for frontier model training, a sign the controls are biting. But the same restrictions have turbocharged government and private investment in domestic alternatives, creating a pipeline of Chinese chip companies that did not exist, or existed only on paper, five years ago.
Until more transparent data emerges from both regulators and industry, the balance of probabilities will remain contested. What is not contested is the direction of travel: China is building its own AI chip ecosystem at a pace that few analysts predicted when the first export controls landed in October 2022. Whether that ecosystem reaches 86% market share by 2030, or falls short, the shift is already reshaping how chipmakers, cloud companies, and governments plan for the next decade.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
