China AI compute demand surges 417% as supply gaps widen

A growing supply-demand gap

Data from the state-backed China Academy of Information and Communications Technology reveals a massive imbalance in the nation's AI infrastructure. During the first quarter of 2026, the demand for AI computing power surged by 417% compared to the previous year. In contrast, the supply of that same computing power only grew by 128%.

This discrepancy means that demand growth was roughly 3.3 times faster than the growth of available supply. Because demand is moving so much faster than the ability to produce hardware, the market has shifted into a sellers' market. In this environment, companies that can secure production capacity essentially guarantee a steady flow of customers.

The reports highlight a structural tension where the physical limits of hardware production are dictating the pace of AI development. As reported by Caixin, this imbalance is creating a situation where the ability to manufacture chips is becoming a primary lever of power. The data suggests that the rapid expansion of AI applications is currently hitting a wall of physical manufacturing constraints.

Domestic chipmakers see revenue spikes

The high demand for local hardware has led to significant financial gains for Chinese chip designers. Moore Threads, a company that designs graphics processors to compete with Nvidia, reported a massive revenue jump of 1,773.92% in its third-quarter filings. This represents about 19 times an increase in revenue compared to the same period a year earlier.

Other domestic players are also seeing growth, though some face different hurdles. Cambricon reported a 104.47% increase in third-quarter revenue, even though it fell short of analyst expectations. Bloomberg attributed that specific shortfall to ongoing manufacturing constraints and a lack of available memory chips.

The financial health of these companies varies based on their position in the supply chain. While Moore Threads saw a massive percentage jump from a small base, Cambricon's results reflect the difficulties of scaling production. These figures illustrate a market where the winners are those who can navigate the complexities of manufacturing and component shortages.

Shortages impact AI model developers

The lack of available computing power is not just a problem for hardware makers; it is actively affecting the software side of the industry. Because supply cannot keep up with the 417% demand spike, leading AI model developers have had to make difficult choices. Companies such as Z.ai, Moonshot AI, and Alibaba were forced to pause or limit new subscriptions for individual users.

These limitations were observed specifically between April and July as the shortage tightened. When developers cannot secure enough compute power to run new models, they must throttle the growth of their user bases. This creates a bottleneck where the ambition of software developers is restricted by the physical reality of chip availability.

For the end user, this means that the rapid rollout of new AI services may be slowed by infrastructure limits. Instead of a seamless expansion of services, users may find themselves waiting for capacity to open up. This highlights how the 'memory wall' and manufacturing limits act as a ceiling for the entire AI ecosystem.

The role of memory and manufacturing

A critical component in this supply chain is DRAM, a type of memory chip essential for high-performance computing. The Chinese DRAM maker CXMT reported a 613.18% increase in third-quarter revenue, with a gross margin of about 89.53%. The company attributed this success to a global shortage of DRAM caused by the massive demand for computing power.

The shortage of memory chips is a recurring theme in the reports from various manufacturers. Bloomberg noted that Cambricon's inability to meet higher revenue estimates was tied directly to these memory shortages and manufacturing constraints. This suggests that the bottleneck is not just in the design of the chips, but in the raw materials and the factories required to build them.

As global demand for computing continues to rise, the price and volume of components like DRAM are expected to remain under pressure. The high gross margins reported by CXMT indicate that the current shortage is allowing manufacturers to command higher prices. This creates a cycle where the high cost of components may further impact the profitability of AI model development.

Capacity as a geopolitical lever

The situation in China illustrates a broader concept where physical constraints create geopolitical leverage. In this framework, the power dynamics of AI are determined by who has the physical capacity to produce hardware, rather than who has the best software models. Because production capacity is so scarce, it becomes the primary factor in deciding which companies can successfully scale their operations.

This shift moves the focus away from pure innovation and toward industrial logistics. When a market becomes a sellers' market, the ability to secure a spot in a factory line becomes a strategic advantage. For China, this means that the domestic ability to manufacture chips and memory is now a central pillar of its AI strategy.

Looking forward, the main question is whether the gap between demand and supply will narrow in the coming months. While the current data shows a significant imbalance, the industry is watching to see if manufacturing improvements can keep pace with the 417% demand growth. Until then, the physical constraints of the chip stack will continue to define the limits of the AI landscape.