Reported H200 Demand From Chinese Technology Companies Exceeds Nvidia’s Available Inventory
Редакция ЦОД:Главное
Chinese technology companies have reportedly placed orders for more than 2 million Nvidia H200 accelerators for 2026, versus Nvidia inventory of about 700,000 units, according to S1. The reported supply gap is driving interest in overseas AI infrastructure, including data centers in Southeast Asia, while Chinese import policy remains unresolved.
- S1 reports more than 2 million H200 orders from Chinese technology companies for 2026, compared with about 700,000 units of Nvidia inventory.
- Nvidia has reportedly approached TSMC about increasing H200 production, but the discussions are not independently confirmed.
- Tencent reportedly uses a five-year Oracle arrangement worth about US$7 billion to access high-performance AI chips through Southeast Asian data centers.
- Access to H200 for Chinese customers remains dependent on evolving US and Chinese policy decisions.
- China’s local-chip requirement is limited to new data centers receiving state funding, based on the supplied material.
Chinese technology companies have reportedly ordered more than 2 million Nvidia H200 accelerators for 2026, while Nvidia’s current inventory is estimated at roughly 700,000 units, according to S1. The figures point to a substantial gap between reported demand and immediately available supply, although they do not disclose delivery schedules, shipment destinations, or committed manufacturing volumes.
Against that backdrop, Nvidia has reportedly discussed expanding H200 production with TSMC in response to demand from Chinese customers. The discussions have not been independently confirmed in the material provided.
Chinese platforms are also pursuing access to high-performance AI compute through overseas infrastructure. S1 reports that Tencent has a five-year agreement with Oracle valued at about US$7 billion, providing access to approximately 100,000 high-performance AI chips through data centers in Southeast Asia. This arrangement illustrates how cross-border data-center and cloud infrastructure can provide an alternative route to advanced compute where direct chip access is constrained.
Tencent’s second-quarter capital expenditure was reported at CNY52.8 billion, up 176% year on year, with spending primarily directed toward AI compute and data centers. Separately, ByteDance reportedly plans to raise its 2026 Nvidia-chip budget to about CNY100 billion from roughly CNY85 billion in 2025, conditional on Chinese approval of H200 imports. That plan should not be read as completed procurement.
The regulatory environment remains a central constraint. The United States may permit H200 sales to approved Chinese customers subject to a stated 25% fee, while China is assessing imports. China also requires locally produced AI chips in new data centers receiving any state funding; the requirement applies to that subsidized new-build segment, not necessarily to all Chinese data centers.
If H200 production expands and the reported platform budgets translate into procurement, global demand for AI accelerators, servers and data-center capacity could remain elevated. At the same time, use of Southeast Asian facilities could shift part of Chinese AI workload demand beyond China’s domestic data-center market, while localization requirements support a separate demand segment for domestic equipment in subsidized new facilities.
The reported imbalance between Chinese demand for advanced AI compute and domestic accelerator availability could sustain demand for offshore data-center capacity and cloud-based GPU access. It also highlights a more segmented infrastructure market: export controls and Chinese localization rules may channel subsidized new domestic data centers toward locally produced AI chips while some high-end workloads seek capacity outside China.
- TechNews 科技新報 / КитайскийПервоисточник →
Материал подготовлен редакцией ЦОД: Главное на основе указанных первоисточников.