China’s AI Capex Boom Masks Energy Vulnerability and Demand Cliff Risks
Record trade volumes obscure structural fragility as $98 billion AI infrastructure bet collides with Strait of Hormuz disruption and monetization lag.
China’s 2024 trade volumes hit a record 43.85 trillion yuan ($5.98 trillion), up 5% year-over-year, but the headline strength conceals a precarious bet on AI infrastructure capex just as energy costs spike and post-investment revenue realization falters.
The growth story is narrower than it appears. While General Administration of Customs data showed China’s exports climbing 13% by volume in late 2024, imports advanced only 2%—a divergence signaling weak domestic absorption and reliance on external demand. Beneath the surface, a concentrated surge in AI-related imports drove much of the apparent resilience: AI product imports jumped 65% in the first half of 2025 to $272 billion, according to Federal Reserve analysis, before collapsing in the second quarter as geopolitical tensions and export controls tightened.
Chinese AI spending is now projected at 600–700 billion yuan for 2025, per Bank of America Securities, with government funding accounting for roughly 400 billion yuan. Matty Zhao, co-head of China equity research at the bank, framed the surge as a direct response to U.S. competition. Goldman Sachs forecast a 65% increase in cloud provider capex, with leading internet firms committing over $70 billion to AI and data centers in 2025 alone.
Energy Shock Exposes Import Dependency
The AI buildout coincides with a sharp deterioration in Energy Security. On March 4, 2026, Iranian forces effectively closed the Strait of Hormuz following escalating conflict with the United States and Israel. Brent crude surged from $71.32 per barrel on February 27 to a peak of $126 by March 8, disrupting roughly one-fifth of globally traded oil. Daily tanker transits through the strait collapsed from 138 to just six by March 24, with 21 confirmed vessel attacks, according to United Against Nuclear Iran.
China imports 54% of its crude from the Middle East, with half transiting the Hormuz chokepoint, per Columbia University’s Center on Global Energy Policy. Iranian shadow tankers alone supplied 12% of China’s oil in 2025—roughly 1.38 million barrels per day. Strategic reserves of 1.2 billion barrels provide a three-to-four-month cushion at current consumption rates, but a prolonged closure beyond mid-2026 would force difficult choices between prioritizing AI infrastructure power demands and broader industrial needs.
“Closure of the Strait of Hormuz would disrupt roughly a fifth of globally traded oil overnight—prices wouldn’t just spike, they would gap violently upward on fear alone.”
— Ali Vaez, Director of Iran Project, International Crisis Group
The International Energy Agency described the disruption as “the largest supply disruption in the history of the global oil market.” For China’s AI sector, which relies on energy-intensive data centers, sustained crude above $100 per barrel would compress margins on already capital-intensive projects while raising questions about the viability of rapid expansion plans.
Capex-Revenue Misalignment Deepens
Even before the energy shock, cracks were forming in AI investment economics. AI services revenue now covers only 15% of AI-specific capex, a figure that declined for the first time in 2025, according to Capital Blueprint analysis. Enterprise adoption lags severely: only 17% of organizations have deployed AI at production scale, leaving an 83% gap between infrastructure buildout and monetization.
- AI services covering just 15% of capex in 2025, down from prior year
- 83% of enterprises not yet at production-scale AI deployment
- Chinese model efficiency gains (DeepSeek at 25% share) eroding hardware demand
- U.S. hyperscaler capex hitting $660–690 billion in 2026, risking overcapacity
Chinese domestic models—led by DeepSeek and similar offerings—have captured roughly 25% market share by delivering comparable performance at lower computational cost. This efficiency undermines the hardware revenue thesis: fewer chips needed per inference task means weaker demand growth even as usage scales. The phenomenon mirrors concerns in Western markets, where Futurum Group projects U.S. hyperscalers will spend $660–690 billion on AI infrastructure in 2026—roughly 75% of total capex—while Microsoft’s cloud gross margin contracted 400–500 basis points due to investment intensity.
Supply Chain Bifurcation Accelerates
China’s push for semiconductor self-sufficiency adds another layer of risk. In November 2025, Beijing banned all state-funded data centers from purchasing foreign AI chips, signaling confidence in domestic production. Domestic chips are expected to power 30–40% of China’s AI compute by 2026, up from under 10% in 2024, according to the Foundation for Defense of Democracies.
Yet yield and performance gaps persist. Export controls have bifurcated semiconductor markets, forcing Chinese firms to rely on less advanced nodes and domestically fabricated alternatives that lag frontier models. The Federal Reserve noted that Chinese imports fell sharply in the second quarter of 2025 amid tightening restrictions, reversing earlier momentum. This creates a paradox: China is investing heavily in AI infrastructure while simultaneously losing access to the cutting-edge components that maximize return on that infrastructure.
China’s 15-year AI roadmap targets a $100 billion domestic industry by 2030, backed by the National Venture Capital Guidance Fund (RMB 1 trillion) and Big Fund III semiconductor subsidies ($47–70 billion). The government’s share of 2025 AI capex—roughly $56 billion—reflects state prioritization of strategic autonomy over near-term profitability.
What to Watch
Three pressure points will determine whether China’s AI investment wave becomes a foundation for long-term growth or a stranded asset story. First, crude prices and Strait of Hormuz transit normalization: if disruptions persist beyond Q2 2026, energy costs will force trade-offs between AI expansion and industrial stability. Second, enterprise AI adoption velocity: the 83% production-scale gap must close rapidly or capex will outpace revenue indefinitely, triggering margin compression and valuation reassessment. Third, domestic semiconductor roadmap execution: China’s goal of 30–40% self-sufficiency by 2026 faces yield challenges, and any further tightening of export controls could widen the performance gap with Western infrastructure.
For now, China’s trade figures offer a narrative of resilience. The underlying composition—export volume growth without price power, import dependence on volatile regions, and concentrated capex in unproven revenue models—suggests fragility. The confluence of energy shocks and demand-side uncertainty has not yet forced a reckoning, but the window for adjustment is narrowing as both geopolitical and market dynamics accelerate.