China's AI Chip Market Splits From the West: RMB 800 Billion at Stake

China's cloud capital expenditure could reach RMB 800 billion by 2027, implying RMB 400–480 billion in AI chip procurement if semiconductors account for 50–60% of that spending. That projection, laid out by China Asset Management Company (ChinaAMC), frames a market that is splitting into two distinct stories: a domestic Chinese supply chain riding accelerating import substitution, and an overseas market where the same AI boom now faces harder questions about pricing, capital efficiency and delivered earnings.

The divergence matters beyond portfolio positioning. AI infrastructure has been the single largest driver of semiconductor demand globally, and the source analysis argues the two markets now reward different bets — volume growth in China, financial discipline abroad.

Who is making the call?

ChinaAMC is one of China's largest asset managers. Founded in April 1998 and headquartered in Beijing, the firm reported RMB 3.15 trillion (US$465.3 billion) in assets under management as of June 30, 2026. Through its Hong Kong subsidiaries and a global partnership network, ChinaAMC provides international investors access to China's capital markets. Its house view carries weight precisely because it steers allocation across the Chinese chip supply chain that Western firms increasingly cannot hold.

What underpins China's near-term AI chip opportunity?

The domestic thesis rests on two compounding forces: expanding cloud capex and accelerating substitution of foreign chips with domestic silicon. The arithmetic is straightforward in the source's framing.

• Cloud capex could hit RMB 800 billion in 2027, up from the current buildout trajectory.
• Chip procurement would then run RMB 400–480 billion, on the assumption that semiconductors take 50–60% of spending — an assumption ChinaAMC itself treats as a range, not a certainty.
• Foundry capacity is expected to double by 2027, relaxing a constraint that has historically throttled domestic AI accelerator output.
• Domestic supply-share validation and overseas supply policies remain the critical variables — meaning the thesis still depends on chips proving they work at scale, and on export controls not shifting again.

The caveat is embedded in the analysis: capacity doubling does not guarantee yield or adoption. Supply-share validation is the gate, and it has not yet fully passed.

Which catalysts could validate China's supply expansion?

ChinaAMC names three near-term checkpoints, all dated and observable rather than speculative. Investors watching whether the domestic thesis is real should track:

• Q4 shipments from Huawei and Cambricon, the two flagship domestic accelerator vendors whose volumes signal genuine demand rather than announcements.
• Progress in domestic HBM qualification (HBM, high-bandwidth memory, is the stacked memory critical for AI training performance and a longstanding Chinese weak point).
• Cloud vendors' December capex guidance, which converts rhetoric into budget commitments for the following year.

One concrete datapoint stands out: CXMT's HBM is expected to enter small-scale production in Q4. CXMT (ChangXin Memory Technologies) is China's leading DRAM maker, and even small-scale HBM output would mark a milestone in reducing dependence on South Korean and US-allied memory suppliers. "Small-scale" is the operative phrase — qualification, not mass volume, is what the source actually promises.

In plain words: the Chinese thesis is testable within two quarters, on shipments and budgets rather than promises.

Where could overseas AI suppliers find pricing power?

The overseas story is not one of slowing demand. ChinaAMC says robust AI spending continues to support demand. What has changed is the scrutiny: pricing power, capital efficiency and earnings delivery are becoming increasingly important — analyst language for "the market will no longer fund capex on faith."

The clearest pricing lever sits in printed circuit boards. PCB supply constraints have extended lead times in some segments to approximately six months, supporting price increases across both materials and finished boards. For PCB and substrate suppliers, that backlog is genuine pricing power — the kind that shows up in margins rather than in press releases.

The risk side is equally specific. Hyperscaler free-cash-flow pressure remains a live concern: disappointing AI returns could hit supply-chain valuations. The big US cloud providers are spending tens of billions annually on AI infrastructure; if revenue from those investments lags, the pressure transmits directly down the component chain. Six-month PCB lead times cut both ways — they support prices today, but they also reflect a buildout that must eventually justify itself.

In plain words: overseas suppliers still have demand and some pricing power, but the market now demands proof of returns.

What is the medium-term outlook?

Despite near-term market volatility, ChinaAMC sees both domestic and overseas AI compute markets retaining a positive medium-term outlook. The asymmetry is in the short term: domestic AI compute enjoys a better near-term outlook, in the firm's assessment, because Chinese capex and substitution are accelerating simultaneously, while overseas players must first demonstrate earnings delivery against already-heavy spending.

The unsaid part deserves attention. ChinaAMC does not quantify what "medium term" means, and its 50–60% chip-share assumption is a band wide enough to move procurement estimates by RMB 80 billion. The 2027 figures are scenario arithmetic, not guidance.

What should readers watch next?

The checkpoints are on the calendar. Huawei and Cambricon Q4 shipment disclosures, CXMT's HBM small-scale production ramp, cloud vendors' December capex guidance, and hyperscaler free-cash-flow results will each either confirm or undercut one leg of the two-market thesis. If domestic supply-share validates while overseas AI returns disappoint, the divergence ChinaAMC describes widens; if both deliver, both markets win — but on different evidence.