China's $295B Plan to Wire Its Data Centers Into One Grid

The United States has spent the last three years treating AI compute as a strategic chokepoint—export controls, chip bans, allied coordination—on the theory that whoever controls the silicon controls the frontier. China just answered with a plan that takes the theory at its word and tries to route around it at national scale. According to reporting first surfaced by Bloomberg on June 9, Beijing is drafting a $295 billion, five-year program to connect thousands of the country's data centers into a single, unified national computing grid—and to build it on a mandate that at least 80% of the underlying technology, including AI accelerator chips, come from domestic suppliers.

If it works, it would be the most aggressive attempt yet to convert a hardware disadvantage into a coordination advantage. If it doesn't, it will run straight into the same wall that prompted it: China still can't make enough advanced chips.

What's actually being built

The blueprint is being drafted by the National Development and Reform Commission (NDRC), China's top economic-planning body, which is a tell in itself. This isn't a company's capex plan or a province's initiative; it's central industrial policy, the same machinery that built China's high-speed rail and solar dominance, now pointed at compute.

The core idea is consolidation. China today has a sprawl of data centers—built in a boom, many underused, scattered across regions with mismatched power and connectivity. The plan would pool those scattered resources into one interconnected network of computing hubs, with state-owned operators like China Mobile and China Telecom running the bulk of the facilities. The stated goal is to knit the country's fragmented compute into a cohesive national grid by 2028.

The framing matters: this is being designed less like a collection of buildings and more like a utility—compute as national infrastructure, allocated and balanced centrally the way a power grid is. That's a genuinely different model from the US approach, where hyperscalers build competing private fleets and the government mostly sets guardrails.

The 80% rule is the whole story

Strip away the gigawatts and the grid metaphor and the load-bearing number is 80% domestic content. Export controls mean China can't build a frontier-scale buildout on cutting-edge Nvidia and AMD parts, so the plan leans hard on homegrown silicon—Huawei above all—to fill the gap.

This is sovereignty by mandate. Rather than wait for domestic chips to win on merit, the policy guarantees them demand: a national-scale buyer committing to fill its grid mostly with local accelerators. For Huawei and China's other chip designers, that's the single most valuable thing a young hardware ecosystem can get—a captive market large enough to fund the next several generations of R&D regardless of how they'd fare in open competition.

It also, by design, locks Nvidia further out. Every percentage point of the grid reserved for domestic silicon is a percentage point Nvidia can't sell into even where export rules might otherwise allow it. Washington's controls were meant to slow China's AI buildout; this plan tries to convert that pressure into a permanent structural preference for Chinese chips at home.

The wall it's heading toward

The skeptical read—voiced bluntly in Tom's Hardware's analysis—is that the 2028 timeline could collide with the limits of local chip production. A mandate that 80% of the silicon be domestic only works if domestic fabs can actually make 80% of the silicon, at the volume and yield a national grid demands. China's advanced-node manufacturing remains constrained—by its own equipment access, by yield, by the same tooling controls that started this whole dynamic. You cannot policy your way past a fab that can't print the wafers.

So the plan contains an internal tension. The more successfully it guarantees demand for domestic chips, the more it exposes whether supply can meet that demand on schedule. A grid that's 80% homemade by mandate but can only source, say, half that in practice ends up either delayed, under-built, or quietly bending the rule. The $295 billion buys coordination and demand certainty; it does not buy lithography.

What it means

For the US, this is the export-control endgame coming into view. Controls successfully denied China the best chips—and in doing so, gave Beijing the strongest possible incentive to build a closed, domestic, state-coordinated compute base that American firms are structurally shut out of. That's the strategic cost of the chokepoint strategy: it works in the short run and breeds a competitor's self-sufficiency in the long run.

For the rest of the world, the more durable signal is the model. China is making an explicit bet that compute is national infrastructure—something a state plans, funds, and operates as a grid, not something it leaves to private fleets. Whether or not the 80% rule survives contact with China's fab reality, that framing is going to spread. The age of treating data centers as a utility, with all the central planning that implies, may have just gotten its defining example.