China's $295 Billion AI Bet: The Data-Center Race

RunFreeTools TeamJun 13, 20266 min read

TL;DR — China is drafting a roughly 2-trillion-yuan ($295 billion) plan to stitch its scattered AI data centers into a single national computing grid by 2028, with at least 80% of the underlying technology — including chips — sourced domestically from suppliers like Huawei. The blueprint, first reported by Bloomberg, would lean on state telecoms China Mobile and China Telecom to operate the facilities and effectively shut Nvidia and AMD out of the build-out. It lands the same month DeepSeek is reportedly raising $7.4 billion in its first external funding round, underscoring how Beijing is wiring up both the hardware and the model labs to chase the United States.

What China actually announced

According to a Bloomberg report published on June 9, 2026, Chinese authorities are preparing to spend around 2 trillion yuan — about $295 billion — over the next five years building data centers across the country. Crucially, that figure covers publicly funded construction only. Bloomberg reported the program would be financed mainly through sovereign debt, including ultra-long-term special government bonds, plus state funds earmarked for strategic industries, with bank loans and private capital layered on top. Folding in the power-grid upgrades needed to feed those facilities could push the total projected investment to at least 5 trillion yuan.

The goal is not just more compute. It is consolidation. China's National Development and Reform Commission (NDRC) is drafting a blueprint to connect the country's fragmented patchwork of computing sites into one cohesive national network by 2028. State-owned giants China Mobile and China Telecom would operate the bulk of the facilities and keep them interconnected.

It is worth stressing that, as of mid-June 2026, this remains a draft. Targets, timelines, and the headline figure could shift before anything is finalized — a caveat the reporting itself makes clear.

The 80% rule: locking out Nvidia and AMD

The most consequential detail is a sourcing mandate. The plan reportedly relies on domestic suppliers, including Huawei Technologies, for at least 80% of the underlying technology — most notably AI accelerators. In practice, that target is designed to squeeze Nvidia and AMD out of the new build-out almost entirely.

That is a sharp turn from the status quo, where Chinese labs have long depended on Nvidia silicon. It also raises an obvious question: can domestic supply actually meet the demand? Reporting on the draft flags a real bottleneck. China's limited high-bandwidth memory (HBM) production constrains how many Huawei Ascend-class accelerators can be assembled, and analysts cited by Tom's Hardware estimate domestic suppliers will cover only around 76% of Chinese AI chip demand by 2030, even as that market swells toward $67 billion. The 80% target, in other words, is ambitious by design.

The push builds on existing groundwork. The initiative expands China's "Eastern Data, Western Computing" project, launched in 2022, which routes compute-heavy workloads from the developed eastern seaboard to eight computing hubs and ten data-center clusters in western provinces that offer cheaper land, power, and cooling. The new plan is also described as a key prong of a broader "Six Networks" infrastructure program announced earlier this year, spanning everything from water and electricity to computing.

DeepSeek's first big raise

The hardware push coincides with a watershed moment for China's most prominent model lab. DeepSeek — the Hangzhou company behind the V3 and R1 models that rattled markets in early 2025 — is reportedly raising roughly $7.4 billion in its first-ever external funding round, according to Reuters.

The round targets a post-money valuation of 350 billion to 400 billion yuan, or about $52 billion to $59 billion. Founder Liang Wenfeng is expected to put in roughly 40% of the total himself — around 20 billion yuan — with Tencent reportedly weighing a 10-billion-yuan commitment and battery giant CATL exploring 5 billion yuan. Reuters reported that China's national AI fund, NetEase, JD.com, IDG Capital, and Monolith Management are also among potential participants. The capital is earmarked for computing infrastructure as the company pushes into more demanding AI agents.

For a lab that famously rose to global prominence without venture money, taking outside capital is a notable shift. We covered DeepSeek's broader trajectory in our look at the DeepSeek AI revolution, and the question of whether it can keep pace with US labs in can DeepSeek challenge OpenAI's dominance. The funding round suggests its answer is to buy more compute — exactly the resource the national data-center plan is meant to supply.

US vs China: the spending gap in context

The $295 billion headline is large, but it sits inside a very different financing model than the US one. American hyperscalers are spending private capital at a pace that dwarfs Beijing's public outlay. Bloomberg noted the Chinese figure pales next to the roughly $725 billion that US firms such as Meta and Microsoft are setting aside for AI this year alone.

Dimension	China's plan	US build-out
Headline figure	~$295B over ~5 years (public construction)	~$725B in 2026 alone (private capex)
Primary funding	Sovereign debt, state strategic funds	Corporate balance sheets, hyperscaler capex
Chips	80% domestic (Huawei Ascend et al.)	Nvidia, AMD, custom silicon (TPU, Trainium)
Operators	China Mobile, China Telecom (state)	AWS, Microsoft, Google, Meta, neoclouds
Coordination	Centrally planned national grid by 2028	Market-driven, fragmented across firms

The contrast is as much about strategy as scale. China is pursuing a centrally coordinated, debt-financed grid built on domestic silicon; the US relies on a market scramble funded by some of the most cash-rich companies on earth. Each approach has trade-offs — central planning can align power, networking, and compute, but it inherits the chip and memory constraints described above.

The bigger players behind the build-out

The named facility operators are state telecoms, but the chip and model layers feature a familiar cast. Huawei is the anchor for domestic accelerators through its Ascend line. Alibaba is moving aggressively too: in April 2026, CNBC reported the company switched on a data center running 10,000 of its own chips as part of China's AI push. Tencent shows up on both sides of the ledger — as a potential DeepSeek backer and as a private-sector data-center builder whose spending falls outside the 2-trillion-yuan public estimate.

The model labs are racing in parallel. Beyond DeepSeek, Moonshot AI's Kimi has drawn attention for its long-context capabilities, which we explored in our piece on Kimi AI's large context window. All of them ultimately depend on the same scarce resource: enough domestic compute, and enough HBM to feed it. That dependency is precisely why the chip question — covered in our analysis of the Nvidia Vera chip war — sits at the center of the US-China AI contest.

The bottom line

If the draft survives in something close to its reported form, China will have committed to a state-engineered AI computing grid built on homegrown silicon — a deliberate decoupling from Nvidia and AMD rather than a stopgap. The $295 billion figure is real and credibly sourced, but two caveats matter: it covers public construction only, so the true national spend is higher, and it is still a blueprint subject to change. The binding constraint will be physical, not financial. Whether Huawei and its peers can produce enough advanced chips and high-bandwidth memory to hit an 80% domestic target by 2028 is the question that will decide if Beijing's bet pays off — and DeepSeek's $7.4 billion raise is a sign the country's labs are not waiting around to find out.