The United States is repairing the "Stargate," while China is refining the "Token Factory."

When North America's leading AI cloud vendors experience quarterly revenue growth surge beyond three digits and their order backlogs surpass $100 billion, the market tends to interpret this as "computing power demand has yet to peak." However, demand is merely a faade; the true determinant of seating arrangements at the table lies in the systemic barriers on the supply sideonly the U.S. and China possess the capability to simultaneously manage large-scale clusters at the ten-thousand-card level, gigawatt-level power, a self-sufficient chip ecosystem, and model iteration capabilities. Reports from Oxford University and Stanford University corroborate this: these two countries concentrate the vast majority of scaled computing power factories globally, with a mere 2.7% performance gap in cutting-edge models, entering a "parallel running" phase. Other countries are either limited to regulation or capital, lacking a complete supply chain, rendering them at different tables. The outcome of this game is defined by Jensen Huang's "Token Factory" theory presented during the GTC conference: intelligent computing centers are no longer mere storage warehouses but factories that produce tokens; commercial value = token throughput per watt usable power. Power goes in, tokens come outwhoever can turn this conversion chain into a measurable, priceable, and tradeable industrial assembly line will obtain pricing power for the next generation of "general commodities." I. How the U.S. Plays: Outsourcing National Strategy to Market Alliances The symbol of America's response is "Stargate." This super-engineering project, supported by OpenAI, SoftBank, Oracle, and sovereign funds, plans to invest $500 billion over four years with a capacity nearing 7GW, and the Texas Abilene flagship park's single-phase capacity reaching 1.2GW, was dubbed "a computing power factory at the level of a nuclear reactor." Its significance goes beyond scaleit is the first to transform "national-level computing infrastructure" from government bidding to a market behavior among giants, replacing administrative approval rhythms with the speed of capital. Its strategy can be summarized into three points: alliances among giants, high-stakes capital; binding to a single technology ecosystem for vertical integration, pursuing extreme computing density; and within a heavy capital closed loop, the logic of engineering continuously grapples with the logic of capital. Leading cloud vendors have signed multi-year long-term contracts, which essentially extend the same logic: locking in future power and cabinet usage for the next three to five years with guaranteed orders, turning CAPEX into cash flows that can be monetized. The actions of Nvidia best illustrate the American logic. Over the past 16 months, it has completed more than 145 mergers and investments, totaling approximately $90 billion spent, from inference chip teams to model open-source platforms to code generation technology licenses, almost securing every choke point from energy to basic models to applications. In the U.S., full-stack capabilities represent a will of capital. The "Token Factory" concept serves as the ultimate measurement for this vertical integration: it is no longer about who has more GPUs, but rather how much tokens can be "refined" per watt of power. Computing power transforms from a cost center into a profit center, while data centers evolve from "electronic warehouses" into "production workshops"this narrative resonates not only on Wall Street but also forms the underlying operating system of the American approach. II. How China Plays: Official Direction and Private Sector Competition The government first manages to create certainty in the most challenging external conditions: the "East Data, West Computing" policy sets a structural framework, "computing-power-electricity synergy" is incorporated into top-level design, and the consensus that "the end of computing power is power" is firmly established. The state lays down a public foundation of "available water, adequate electricity, and usable land," allowing market entities to develop independently. At the same time, "tokens" are evolving from technical jargon into industry language. In March 2026, the National Bureau of Statistics officially confirmed the Chinese translation of Token as "," clarifying its properties as "measurable, priceable, and tradeable"; on another note, China's average daily token usage has surpassed 500 trillion, nearly increasing by 5,000 times over the past two years. When the "token economy" is included in local industrial plans, China's "Token Factory" transcends mere corporate behavior, evolving into an industry chain shaped from the top down, from infrastructure to settlement standards. Range Intelligent Computing Technology Group (300442.SZ) exemplifies the "foundation first" model. In the first half of 2026, its AIDC business revenue grew by 126% year-on-year, surpassing traditional IDC for the first time. It replicates the earlier logic of its North American counterparts: first addressing physical bottlenecks in computing power placement, power supply and cooling, and high-density deployment, before moving up to provide high-value-added services. More critically, it has secured multiple contracts for whole rack deliveries at the ten-thousand-card level, starting to touch the edges of computing power schedulingtransitioning from "turnkey" to "token delivery," only a layer of software away. Sharetronic Data Technology (300857.SZ) and Jiangsu Lettall Electronic (603629.SH) represent collective actions of "position takers": one extends from server manufacturing to intelligent computing centers and optical modules, attempting to reverse "bottlenecks" into self-manufacturing at the hardware level; the other shifts from computing power leasing to self-built capacity for controllable bottom-layer computing power. These actions confirm the same consensuswhile full-stack may be challenging, no one wants to miss out; and "positioning" itself is about tightening every bolt in the token industry chain. GBA AI COMP (01396) has further refined the concept of "full-stack." This Hong Kong-listed company saw revenue grow over tenfold in the first half of the year, with intentions orders exceeding 37 billion RMB, and its FP16 dense computing power scale reaching 50,000 PFLOPS. Its "full-stack" is more than a simple cumulative process; centered around its self-developed cloud platform "Quantum Pack", it possesses both autonomous dispatch capabilities for large clusters and can connect energy foundations, computing clusters, token delivery, and application incubation into a closed loop: internally adjusting power and computing capabilities according to training inference loads, while externally providing token services via standard interfaces, and optimizing upstream resource allocation in reverse. This shift from "selling computing power" to "delivering tokens" neatly aligns with the logic of the "Token Factory." While most peers are still addressing the question of "whether there is," it has quietly sketched out the delivery outline of China's Token Factory. III. Different Paths, Possible Coincide The best perspective is never to seek a Chinese replica of an overseas giant. What truly merits attention is that these two routes are pointing towards the same destination: turning astronomical CAPEX into sustainable cash flows and token output. The end of electricity is computing power, and the end of computing power is pricing power. Whoever first refines "electricity" into measurable, tradeable, and exportable "tokens" holds the ticket for entry into the next round. In China's batch of submissions, the first to reveal this card has positioned itself in a more advanced place.