NextFin News - China’s lead over the United States in important parts of the technology stack is less a claim that Beijing has won the artificial-intelligence race than a warning about how the race is being measured. Rhodium Group’s framework puts China ahead in deployment-heavy layers such as wireless networks, connected devices, industrial hardware, batteries and critical inputs, while the United States and its partners retain advantages in advanced semiconductors, cloud infrastructure and frontier AI models. The tension matters because the lower layers determine how quickly innovation becomes an installed system.
The strategic question is not simply who has the best model or the fastest chip. It is who can turn technical capability into cheap, reliable and globally distributed infrastructure. On that measure, China has built a formidable position. The U.S.-China Economic and Security Review Commission describes the two countries as neck-and-neck overall, with China holding a clear advantage in manufacturing-intensive technologies such as advanced batteries and electric vehicles, while the United States and like-minded countries lead in advanced semiconductors, total compute and cloud, and robust AI models.
That split changes the meaning of export controls. Washington can restrict access to frontier chips and manufacturing equipment, but Beijing can apply scale lower in the stack: protected domestic demand, concentrated supply chains and manufacturing capacity that reduces costs through volume. The result is a contest in which the United States can lead the visible frontier while China shapes the physical environment in which that frontier is deployed.
The evidence is clearest in clean technology. The International Energy Agency says China’s share exceeds 80% across the major solar-panel manufacturing stages, including polysilicon, ingots, wafers, cells and modules. Based on capacity under construction, China’s share of polysilicon, ingot and wafer production was expected to approach 95%. The agency also says China invested more than $50 billion in new photovoltaic supply capacity and created more than 300,000 jobs across the value chain since 2011.
The same pattern runs through batteries. The U.S. Department of Energy’s supply-chain assessment estimates that China controls more than 60% to 90% of critical midstream battery production, with the department’s cited compilation putting China’s shares at 71% for cells and 96% for anode active material, among other stages. Those figures do not mean every battery made outside China is Chinese-owned or irreplaceable. They show that the manufacturing center of gravity sits in China, especially in intermediate materials that are difficult to rebuild quickly.
China’s position also extends to the power needed by the next generation of data centers. Rhodium estimates that China could have enough electricity resources by 2030 to power the equivalent of 280 gigawatts of data centers equipped with U.S. hardware, more than twice projected U.S. capacity and more than the rest of the G7-plus bloc combined. The estimate is about energy availability, not installed AI capacity, but it exposes the infrastructure constraint behind the chip debate.
At the frontier, the picture reverses. The USCC says the United States and like-minded countries have an advantage in advanced semiconductors needed for AI, while the United States leads in total compute and cloud and in developing robust AI models. U.S. export controls beginning in October 2022 and expanded in October 2023 and April 2024 were designed to preserve that advantage by limiting China’s access to high-end chips and the tools used to manufacture them. The policy has created a chokepoint. It has not erased China’s ability to build the rest of the stack.
China’s Lead Is in Deployment, Not Every Frontier
The first judgment is straightforward: China’s technology advantage is most durable where the winning variable is cumulative industrial scale rather than a single breakthrough. Rhodium’s wider definition of the AI stack begins with raw materials, intermediate components and energy, then moves through semiconductors, connectivity, edge devices, cloud and models. That ordering matters. AI systems are not only algorithms; they are networks of sensors, batteries, machines, data centers and communications links.
In the upper layers, the United States benefits from deep software talent, large pools of private capital and companies that have established global standards. Advanced chips are a particularly powerful example. Design, electronic-design automation, high-end manufacturing equipment and leading-edge fabrication are concentrated among U.S. and allied firms. These are high-value bottlenecks, and their scarcity gives Washington leverage over the speed and cost of China’s frontier-compute buildout.
But a bottleneck is not the same as control of the whole system. China can respond by shifting competition toward inference efficiency, open-weight models, specialized chips, edge computing and mass deployment. Rhodium argues that Beijing is focused on deploying cheap, near-frontier models rather than matching the United States at every point on the frontier. That approach lowers the economic value of a marginal lead in model capability if a less advanced model can be embedded in a robot, vehicle, factory or telecom network at much lower cost.
The mechanism is a feedback loop. A protected home market creates demand. Demand supports factories and suppliers. Factory volume lowers unit costs. Lower costs accelerate adoption, generating more operational data and more demand for domestic components. The next cycle of investment then begins from a lower cost base. The United States has world-class firms, but it does not have a single domestic market large enough to reproduce China’s scale across every physical layer. Rhodium’s conclusion is blunt:
“The US cannot match China’s scale alone, only a coalition can.”
This is a structural advantage, not merely a temporary subsidy effect. Government support can be wasteful, and excess capacity can destroy returns. Yet factories, supplier relationships, process knowledge and installed networks accumulate even when individual projects fail. The cyclical portion is the investment wave: China’s solar and battery sectors have faced oversupply, falling prices and weak margins. The structural portion is the capability to build and deploy at a scale that competitors cannot replicate quickly.
The distinction explains why low prices can be strategically useful even when they are bad for producers. A glut compresses margins, but it also makes Chinese equipment more attractive to buyers and can delay competing capacity outside China. The cost is borne by manufacturers; the advantage accrues to the ecosystem that learns faster and captures downstream demand.
The Second-Order Effect Runs Through Other Countries
The second judgment is that China’s lead becomes more consequential when it crosses borders. The first-order effect is cheaper equipment and components. The second-order effect is that countries in the Global South may adopt Chinese-built connectivity, power, vehicles and industrial systems because those systems are available at scale and fit their manufacturing ambitions.
That changes the competitive unit from a company to a network. A telecom supplier that wins a national contract can shape standards, maintenance practices, data flows and future procurement. A battery supplier that provides cells also influences the design of vehicles, charging systems and recycling networks. A country that adopts a technology platform gains near-term affordability but may also deepen dependence on the platform’s components and software.
The comparison with solar is revealing. The IEA expects China to retain more than 80% of global solar manufacturing capacity in 2030 and says U.S. and Indian module production currently costs two to three times more than production in China. That cost gap is not simply a trade statistic. It is a reason for downstream developers to use Chinese inputs, which in turn keeps Chinese factories operating at scale and makes alternative supply chains more expensive.
The same cross-border channel applies to physical AI. Robots, autonomous vehicles and connected industrial equipment need batteries, magnets, sensors, wireless links and power electronics. Rhodium says China’s control of critical raw materials and intermediate inputs can constrain the manufacturing supply chains behind those devices. If Chinese firms become the default assembly and infrastructure suppliers for physical AI in emerging markets, the United States could retain the best model while depending on Chinese-linked inputs to put that model into the world.
That is the second-order risk for U.S. technology companies. Export controls may protect the high end, but they can also accelerate a split market. Chinese firms would have stronger incentives to localize their stack, while non-Chinese markets would need to buy from a more expensive coalition of suppliers. The near-term result could be slower diffusion and higher capital costs outside China, even if the coalition remains safer from strategic dependence.
Energy makes the problem harder. Rhodium’s 280-gigawatt estimate shows that compute is constrained by electricity as well as chips. The United States can design a leading accelerator, but data-center construction still requires land, transmission, generation and permitting. China’s ability to mobilize power and industrial capacity gives it an advantage in deployment speed even when the hardware inside the facility is foreign.
This is why the conventional framing can be misleading. A chip lead may be the most valuable position today, but the value of that lead depends on who controls the complementary assets. The next expectation gap is not whether China will suddenly surpass the United States in frontier models. It is whether China can make slightly weaker models economically dominant by placing them in more machines, factories and networks.
Export Controls Preserve a Chokepoint, but They Also Test Its Durability
The third judgment is that U.S. controls are a real constraint on China’s frontier, but their effectiveness depends on whether allied restrictions remain broader than China’s ability to substitute around them. The Commerce Department’s Bureau of Industry and Security says the October 2022 rules restricted China’s ability to buy and manufacture certain high-end semiconductors. Updates in October 2023 and April 2024 expanded the regime, including controls on 24 types of semiconductor-manufacturing equipment, three types of software tools, high-bandwidth memory and additional Chinese entities.
The policy mechanism is clear: limit the most advanced compute and the machinery required to reproduce it, then slow the diffusion of military-relevant capability. Alan Estevez, then the Commerce Department’s undersecretary for industry and security, described the effort this way:
“This action builds on BIS’s laser-focused work, undertaken over the past few years, to impose strategic controls that have hindered the PRC’s ability to produce advanced semiconductors and AI capabilities directly impacting U.S. national security.”
The counter-thesis is stronger than the claim that controls do nothing. The real challenge is that controls can buy time while also encouraging substitution. China’s dependence on foreign equipment and advanced chips is a vulnerability today, but it is also a clear target for state funding. If Beijing can accept lower performance, use more chips, improve software efficiency and expand domestic equipment, the performance gap may not translate into a deployment gap.
History supports both sides. Semiconductor manufacturing is capital-intensive and technically cumulative, so restrictions can delay progress for years. But industrial policy has repeatedly moved production into new jurisdictions when demand, financing and learning were sufficient. Solar manufacturing provides the warning: capacity migrated toward China over the last decade, and the resulting network now has a cost advantage that trade restrictions alone do not quickly remove.
The structural call is therefore conditional. The U.S. lead in frontier semiconductors is durable over the medium term because the equipment, design and process ecosystem is difficult to reproduce. China’s lead in physical deployment and critical inputs is also durable because it rests on installed capacity and supplier density. The cyclical component is the margin pressure and investment overshoot inside those sectors. It will mean-revert. The geographic concentration of know-how and production will not.
The strongest argument against this view is that China’s scale can be overstated. State-directed investment can produce excess capacity, low returns and politically allocated capital. The USCC says China faces a faltering domestic economy and inefficiencies in its state-directed innovation system. A country can dominate manufacturing volume without creating the most valuable technology, and low-margin exports can provoke tariffs that close foreign markets. If these constraints dominate, China’s physical lead would become a burden rather than leverage.
That counter-thesis is credible, especially in sectors where supply exceeds demand. It does not, however, remove the strategic effect of low-cost capacity. It changes who captures the profits and who bears the adjustment. China can lose money at the factory level while foreign buyers gain cheaper equipment and Chinese suppliers gain market share. The falsifying signal for the structural thesis would be concrete: if China’s share of global solar manufacturing capacity fell below 70% by 2030 and its share of critical battery midstream production fell below 50% without a corresponding collapse in domestic deployment, the scale-based advantage would be materially weaker than it appears today.
Controls would also be shown to have worked more completely if China’s share of advanced-compute capacity stopped growing while domestic semiconductor-equipment output failed to replace imported tools. Conversely, a sustained decline in the performance gap between Chinese and leading U.S. AI systems, combined with lower Chinese inference costs, would show that the upper-layer chokepoint is losing economic force.
What the Stack Split Means for Industry and Policy
The implications are asymmetric. U.S. semiconductor designers, equipment makers, cloud providers and frontier-model developers retain leverage because they sit at scarce, high-value nodes. Their exposure is that the addressable market may fragment and that Chinese deployment can reduce the premium attached to frontier capability. Chinese battery, solar, telecom, robotics and industrial-equipment companies benefit from scale, but they face margin compression, trade barriers and dependence on access to foreign markets and some foreign technologies.
For policymakers, the practical lesson is that a chip-only strategy does not answer a stack-wide competition. The United States and its partners would need demand aggregation, grid investment, mining and processing capacity, allied manufacturing and standards coordination. Rhodium’s coalition argument follows from arithmetic: no single Western market can supply China-sized demand and production across all the complementary layers.
For markets, the short-term horizon is sentiment and policy risk. New export restrictions, tariff announcements or licensing changes can move semiconductor and industrial-equipment valuations before physical capacity changes. The medium-term horizon is margin and utilization. Solar and battery oversupply can pressure producers even as their systems become more competitive. The long-term horizon is ecosystem control: who owns the standards, installed base, service networks and critical inputs that determine future switching costs.
The base case is a bifurcated stack through 2030. The United States and allies preserve a lead in advanced chips, cloud and frontier models, while China retains an advantage in batteries, solar manufacturing, wireless deployment and selected physical-AI inputs. The trigger is continued allied coordination alongside sustained Chinese capacity and domestic demand. The upside case for the United States is a coalition that reaches scale, closes the cost gap and builds trusted alternatives; its trigger would be non-Chinese manufacturing capacity growing fast enough to reduce China’s critical-input shares below the thresholds above. The downside case is that export controls slow China’s frontier only temporarily while Chinese platforms become the default in emerging markets; the trigger would be rising Chinese deployment, falling inference costs and broader adoption of Chinese-linked industrial standards.
The next useful data points are not one headline ranking. They are China’s share of global production in batteries, solar components and network equipment; the pace of allied capacity additions; the cost per unit of non-Chinese alternatives; and the performance-per-dollar gap between Chinese and U.S. AI systems. Those measures test whether China’s lead is merely an industrial boom or a self-reinforcing technology ecosystem.
China is not ahead everywhere. It is ahead where technology becomes infrastructure, and infrastructure compounds. The strategic contest will be decided less by who owns the single best layer than by who can connect the layers at scale.
Data cutoff: Aug. 5, 2026, 03:57 UTC.
Explore more exclusive insights at nextfin.ai.
