Nvidia’s $70B China AI Order May Deliver Just 1% Near Term

Chinese AI groups have ordered about 6 million Nvidia RTX Pro 5500 chips worth roughly $70 billion, but market sources expect only about 1% to arrive within four months.

Alibaba (09988.HK), Tencent Holdings (00700.HK) and ByteDance are the named buyers. More than $11,000 per chip, or roughly NT$360,000, puts the reported order value near NT$2.3 trillion.

How far apart are the orders and deliveries?

People familiar with the matter said buyers expect tens of thousands of chips over the next four months, about 1% of the total. The gap does not signal cooling Chinese demand; Nvidia’s production capacity reportedly cannot keep pace with the global AI data-center buildout.

• First batch: Several thousand chips are expected before the end of this year.
• Second batch: Tens of thousands of chips are expected around February next year.

These are the only two confirmed shipment arrangements cited. Even the longer-range figure remains an estimate: internal estimates from major buyers point to roughly 700,000 RTX Pro 5500 chips by the end of 2027. Nvidia declined to confirm volumes, which could change as it adjusts capacity.

Why is capacity moving to newer products?

Nvidia has warned customers that capacity will remain tight for some time because of product-scheduling conflicts. The company relies on Taiwan Semiconductor Manufacturing Co. (2330.TW) to make its AI chips, and TSMC’s shift of manufacturing resources toward newer products such as Rubin is expected to constrain RTX Pro 5500 supply.

Nvidia is allocating scarce capacity first to advanced, higher-margin products, mainly to serve customers such as OpenAI and Anthropic. The RTX Pro 5500 ranks behind those high-end customers in the product lineup.

Market observers characterized the near-term opportunity as potential demand that is “visible but out of reach.” For Nvidia, actual revenue recognition depends on TSMC’s capacity allocation and Rubin’s mass-production schedule, not on Chinese customers’ willingness to order. The $70 billion therefore represents reported order value, not recognized revenue.

Why can China still buy the RTX Pro 5500?

The RTX Pro 5500 does not cross the threshold of U.S. export controls. Washington bars Nvidia from exporting its most powerful AI accelerators to China, citing national-security and military concerns that the chips could accelerate Chinese AI development and strengthen military capabilities.

The Trump administration recently eased restrictions on some older-generation chips, but only small quantities are currently reaching China. The RTX Pro 5500 has therefore become one of the few high-end AI chips still available in the Chinese market.

A Nvidia spokesperson said U.S. companies face constraints from outdated export-control rules covering gaming products launched nearly five years ago and from China’s own restrictions on imports of American products. Nvidia declined to comment on specific shipment volumes. Alibaba, Tencent and ByteDance did not respond to requests for comment.

Can domestic suppliers close the gap?

Market sources said Huawei and Alibaba’s T-Head still lag U.S. competitors in technology and manufacturing capability, despite continued progress by Chinese hardware companies. That leaves China’s AI-computing shortage difficult to fill completely in the near term.

Large Chinese cloud providers are already purchasing AI inference chips from Huawei, Cambricon (688256.SS) and T-Head. U.S. restrictions on advanced chip-manufacturing equipment and technology services, combined with a prolonged shortage of high-bandwidth memory, constrain Chinese production and push up prices.

Cambricon’s latest earnings report reflected capacity constraints. Huawei, China’s largest AI-chip supplier, raised prices for its top-tier AI accelerators by 60% around September. People familiar with the matter described the resulting squeeze as “demand spilling over, supply constrained.”

What does this mean for revenue and competition?

For China’s AI industry, a delivery rate near 1% means computing-power expansion will lag leading U.S. companies, while large-scale model training and inference remain constrained.

TSMC’s scheduling has had limited near-term impact on the overall demand structure for advanced process and packaging capacity. Still, the crowding-out of products such as the RTX Pro 5500 provides a clue about Nvidia’s product mix and gross margin.

Through the end of 2027, the decisive variable will be TSMC’s allocation as Rubin ramps: more capacity for the RTX Pro 5500 would expand Nvidia deliveries, while continued prioritization should keep domestic accelerators scarce, sustain price pressure and sharpen competition among Huawei, Cambricon and T-Head.