Nvidia H200 Chips Reach China as Beijing Eases Curbs: ByteDance and Tencent Receive 10,000 Each

Small batches of Nvidia's H200 AI chips have been allowed to enter mainland China, according to a Financial Times report published Wednesday, marking a significant shift in Beijing's posture toward US-made accelerators as the global AI race escalates.

The report, which was picked up by Reuters and multiple financial outlets, states that ByteDance and Tencent — two of China's largest technology companies — have each received around 10,000 Nvidia H200 chips as Beijing permits limited imports while continuing to push domestic alternatives. The FT described the move as an effort by Chinese authorities to aid the country's leading AI companies even as they promote homegrown silicon.

The development is the latest chapter in one of the most closely watched stories in AI policy, and it arrives amid a steady stream of developments covered in breaking AI news from the world's two AI superpowers.

A Long Road to Delivery

The chips' arrival caps an unusual back-and-forth between Washington and Beijing that has stretched across most of 2026.

In January, Reuters reported that China had given the green light to ByteDance, Alibaba and Tencent to purchase Nvidia's H200 chips, with sources describing import approvals covering more than 400,000 GPUs.

The United States followed in the spring, clearing export licenses for H200 sales to Chinese buyers — approvals reportedly worth an estimated $10 billion in total, with licenses for roughly ten Chinese firms and caps of up to 75,000 units each.

But through mid-July, almost nothing had actually shipped. Under Secretary of Commerce Jeffrey Kessler told Congress that only a small number of H200 chips had reached China or Hong Kong, calling the volumes trivial. Nvidia had repeatedly warned investors that China revenue would remain limited despite the licenses.

That logjam now appears to be breaking. The Financial Times reports that chips are moving into the country in small shipments, with ByteDance and Tencent each having taken delivery of approximately 10,000 H200s — modest quantities relative to the approved volumes, but a meaningful signal that both governments are permitting the trade to flow. Moneycontrol, citing the FT report, characterized the shipments as small batches entering the country under the eased curbs.

Why Beijing Changed Course

China's leadership has spent two years restricting access to foreign AI hardware in an effort to force its tech sector onto domestic chips from Huawei, Cambricon and other local suppliers. But those supply chains have struggled to meet the voracious compute demands of Chinese AI labs competing with American frontier models.

According to the FT's account, Beijing is now calibrating that policy: allowing limited H200 imports to relieve pressure on its champion companies, while keeping the overall push toward self-sufficiency intact. The H200 occupies a strategic middle ground — powerful enough to matter for training and inference workloads, yet an older-generation part that Washington has deemed exportable under current rules, unlike Nvidia's flagship Blackwell-class chips.

The strategy echoes reporting from earlier this month that Moonshot AI trained its Kimi models on a 20,000-chip H200 cluster leased through Alibaba — evidence that Chinese labs are finding creative paths to Western silicon even under constraint.

What It Means for Nvidia

For Nvidia, the trickle of deliveries is unlikely to move quarterly results on its own. Analysts noted in July that the initial volumes were too small to change guidance, given the company's overall scale.

But the signal value is considerable. China was once one of Nvidia's largest markets, and the H200 licensing regime represents Washington's chosen framework for keeping that market partially open — tightly licensed, capped in volume, and restricted to previous-generation hardware. If Beijing continues easing import limits and shipments scale from thousands toward the hundreds of thousands of approved units, the revenue implications grow quickly.

The broader context also matters: Nvidia has moved to insulate its AI business through massive compute deals with US customers, including the landmark arrangement with OpenAI in Ohio backed by a multi-billion-dollar guarantee. Every incremental chip delivered into China adds to an already record-setting demand picture for AI accelerators.

The Balancing Act Ahead

Plenty of uncertainty remains. Chinese authorities could tighten the spigot again if domestic chip yields improve or if political tensions flare. US officials have shown willingness to revisit the licensing framework, and lawmakers have already criticized the H200 approvals as a loophole that hands Beijing advanced computing power.

For now, though, roughly 20,000 H200s are sitting inside two of China's AI giants, and more shipments are reportedly in motion. In an AI race increasingly defined by compute access, even small batches carry outsized weight — a fact not lost on either Washington or Beijing as the year heads into its final stretch.

Stay Ahead of AI

Export controls, chip diplomacy and compute geopolitics are reshaping the AI landscape in real time. Follow AI industry coverage for the latest on every policy shift.

Read more AI news →