China Chipmakers Claim 41% of AI Chip Market; US Data Centers Stalled by Import Gap

China's domestic AI chip vendors crossed a milestone market-share threshold as American data center builders confront a supply chain gap closer to home.

🗄️ Chinese Chipmakers Hit 41% of China's AI Accelerator Market — Nvidia's Share Falls to 55%

Decoded: Chinese GPU and AI chip vendors collectively captured 41% of China's AI accelerator card market in 2025, according to IDC data reviewed by Reuters on April 1. Total shipments of AI accelerator cards across all vendors reached approximately 4 million units in China last year. Nvidia retained the overall lead, shipping around 2.2 million cards for a 55% market share — a significant decline from its prior dominant position. AMD shipped roughly 160,000 cards for a 4% share. Chinese vendors collectively shipped approximately 1.65 million cards. Huawei Technologies led all domestic vendors, shipping around 812,000 AI chips — roughly half of all Chinese-branded shipments. Alibaba's chip unit T-Head placed second with approximately 265,000 cards. Baidu's Kunlunxin and Cambricon each shipped around 116,000 cards. GPU startups MetaX, Hygon, and Iluvatar CoreX accounted for the remainder of Chinese vendor share. China's central government launched a new wave of AI infrastructure spending in 2025, with local governments accelerating intelligent computing centers and implicit directives to buy domestic. (Reuters/IDC, April 1, 2026)

Why it matters: Nvidia's (NVDA) China AI chip revenue has faced sustained pressure since 2022 from successive U.S. export control rounds — but this IDC dataset quantifies what that pressure looks like in shipped units: 41% of an approximately 4-million-unit market now going to domestic alternatives in a single year. Huawei's 812,000-chip shipment figure is particularly notable because it arrives alongside Reuters' March 27 report that ByteDance and Alibaba are planning additional orders after testing — meaning 2026 Huawei volumes could exceed 2025 further. AMD's 4% share indicates that even U.S. competitors with lower export control constraints are not benefiting meaningfully from Nvidia's China retreat. The structural driver — government-mandated buy-Chinese procurement — is a policy variable, not a market one, making Nvidia's China recovery path dependent on geopolitical thaw rather than product competition.

🏛️ US AI Data Centers Can't Be Built Without Chinese Electrical Imports

Decoded: The U.S. AI data center construction boom is structurally dependent on Chinese-manufactured electrical equipment, Bloomberg reported on April 1. The domestic manufacturing base for transformers, switchgear, and battery storage systems — core electrical infrastructure required to power and connect large-scale data centers — cannot meet demand at the pace of the AI build cycle. The supply shortfall has forced U.S. data center developers to import this equipment from China, creating construction delays and raising exposure to tariff disruption. U.S. transformer production capacity is constrained by decades of offshored manufacturing; lead times for large power transformers from domestic suppliers now run 2–4 years in some categories. Eaton (ETN), a primary domestic supplier of switchgear and power distribution equipment for data centers, has flagged extended order lead times as demand from AI buildouts has outpaced its manufacturing capacity. (Bloomberg, April 1, 2026)

Why it matters: The AI data center capex narrative has centered on GPU supply, power generation, and land — but the Bloomberg report identifies a less visible bottleneck: the electrical equipment layer between the power source and the compute hardware. Every data center requires transformers to step down grid voltage, switchgear to distribute and protect circuits, and battery storage for continuity. If this equipment can only be sourced at scale from China, the construction timelines for hyperscale AI campuses — including those being built by Microsoft, Meta, Google, and Amazon — carry embedded exposure to U.S.-China trade policy. Tariff escalation on Chinese electrical imports would directly extend data center completion timelines and inflate per-facility capital costs at a moment when all major hyperscalers have publicly committed to multi-hundred-billion-dollar AI infrastructure spending through 2028.

Stay decoded. See you tomorrow.

— The Get AI Decoded Team