Orders and approval: The Information reportedly says that “several” Chinese customers have ordered the chip, but no reliable reporting identifies them. There is also no public confirmation that Beijing has approved sales of this particular LPU product; an order is not equivalent to an import or regulatory clearance.

Market role: If allowed to ship, it could modestly ease China’s shortage of high-end AI compute for serving models, especially inference. But it would face increasingly capable domestic alternatives, including Huawei’s Ascend 950DT, slated for cloud deployment in 2026 and aimed at training and decoding/inference workloads. Evidence is insufficient to make a credible performance or price comparison with Baidu’s hardware or to say the Nvidia chip would decisively outperform domestic options.

H200 is the more immediate route back: Beijing has begun allowing limited H200 imports, with ByteDance and Tencent reportedly receiving roughly 10,000 processors each. That is a limited exception intended to assist leading Chinese AI firms, not a broad reopening of the market.

Broader strategy and risk: The prospective LPU chip appears to be a hedge: Nvidia can target inference with a separately adapted product while it pursues constrained H200 sales and develops China-compliant, lower-power products. The strategy remains exposed to both U.S. export-control changes and Beijing’s discretionary approvals; earlier reports said Nvidia required full upfront payment for H200 orders because of that approval uncertainty.

Orders and approval: The Information reportedly says that “several” Chinese customers have ordered the chip, but no reliable reporting identifies them. There is also no public confirmation that Beijing has approved sales of this particular LPU product; an order is not equivalent to an import or regulatory clearance.

Market role: If allowed to ship, it could modestly ease China’s shortage of high-end AI compute for serving models, especially inference. But it would face increasingly capable domestic alternatives, including Huawei’s Ascend 950DT, slated for cloud deployment in 2026 and aimed at training and decoding/inference workloads. Evidence is insufficient to make a credible performance or price comparison with Baidu’s hardware or to say the Nvidia chip would decisively outperform domestic options.

H200 is the more immediate route back: Beijing has begun allowing limited H200 imports, with ByteDance and Tencent reportedly receiving roughly 10,000 processors each. That is a limited exception intended to assist leading Chinese AI firms, not a broad reopening of the market.

Broader strategy and risk: The prospective LPU chip appears to be a hedge: Nvidia can target inference with a separately adapted product while it pursues constrained H200 sales and develops China-compliant, lower-power products. The strategy remains exposed to both U.S. export-control changes and Beijing’s discretionary approvals; earlier reports said Nvidia required full upfront payment for H200 orders because of that approval uncertainty.