ByteDance, Alibaba, Tencent Order Over $35 Billion in Nvidia Chips

ByteDance, Alibaba and Tencent have placed orders for Nvidia chips worth more than $35 billion, according to a Reuters exclusive. That figure makes the three Chinese groups collectively one of the largest blocs of Nvidia customers by order value, and it lands at a moment when Washington's export rules still govern which Nvidia silicon can legally ship into China.

The buyers are the three biggest spenders on AI infrastructure in the Chinese market. ByteDance runs the Douyin and TikTok platforms; Alibaba operates cloud computing (rented data-center capacity) and the Qwen model family; Tencent owns WeChat and its own cloud unit. All three need GPUs (graphics processing units, the accelerator chips that train and run large AI models) at industrial scale.

Nvidia doesn't break out China revenue by customer, and none of the four companies has published the order figures themselves. Reuters reports the aggregate exceeds $35 billion.

What does the order change for Nvidia?

Nvidia dominates the market for AI accelerators — its data-center GPUs carry the bulk of global AI training workloads, and its fiscal results have repeatedly hinged on a handful of hyperscale buyers. An order book above $35 billion from three Chinese firms alone deepens that concentration.

It also signals continued Chinese demand despite US export controls (the licensing regime that restricts sales of Nvidia's highest-end chips to China). Where top-tier parts are blocked, buyers have turned to compliant variants — chips re-specced to stay under US performance thresholds.

For Nvidia investors, the headline question is delivery: orders convert to revenue only as chips ship, and export-license conditions set the pace.

What does it change for the Chinese buyers?

For ByteDance, Alibaba and Tencent, the orders are a compute bet. Training frontier-scale AI models consumes thousands of GPUs over months; running inference (serving a trained model to end users) consumes more, continuously. The $35 billion-plus commitment is essentially prepaid capacity for both workloads.

It also marks a divergence from Beijing's push for domestic alternatives. Chinese chipmakers have gained ground, but their accelerators and the software stacks around them still don't match Nvidia's CUDA ecosystem (the proprietary programming layer most AI code is written for) in maturity.

The money at stake

More than $35 billion in chip orders. That's the number. Spread across three buyers, it implies per-company commitments in the billions of dollars — with Reuters not specifying the individual split.

The timing matters as much as the total. The orders come amid:

• Persistent US-China tension over advanced-semiconductor exports
• Ongoing debate over whether Nvidia's compliant China variants face further restrictions
• A global AI capex race in which Chinese cloud and platform firms are racing US rivals for accelerator supply

What to watch next

Watch Nvidia's next earnings report for China-segment revenue disclosure — the closest public proxy for how fast these orders convert into shipped silicon.

Watch US Commerce Department moves on export thresholds; any tightening directly affects which of the ordered parts can legally deliver.

And watch whether the three Chinese firms confirm or contest the figures. None has commented publicly so far, and Reuters' report stands as the sole sourcing for the $35 billion number.

One more angle: procurement at this scale usually spans multiple chip generations and delivery windows. An order placed today may not fully ship within a single fiscal year — meaning the $35 billion figure is a commitment, not a single-quarter revenue event.

For now, the signal is plain. Chinese AI spend on Nvidia silicon remains enormous, export rules notwithstanding, and three companies account for a chunk of it that now has a number attached: north of $35 billion.

Grace Kim is a trade-press correspondent covering semiconductors and AI infrastructure.