Nvidia's New China Rule: Pay First for H200 AI Chips Amid Export Uncertainty
Nvidia is now requiring Chinese customers to pay in full, upfront, before receiving its H200 AI chips—a significant policy shift that reflects just how precarious the AI hardware supply chain has become under the weight of US-China trade tensions.
The requirement, first reported by TechCrunch, applies even as regulatory approval for H200 sales to China remains uncertain from both the US Commerce Department and Beijing. In practical terms: Chinese companies want Nvidia's best AI training hardware, Nvidia isn't sure it can legally ship it, and yet the company is asking customers to wire the money anyway.
Why Nvidia Is Changing Payment Terms for China
This isn't standard business practice. Major chip deals typically involve milestone payments or net terms—not full prepayment. The shift suggests Nvidia is managing multiple risks simultaneously: regulatory uncertainty that could block shipments entirely, the possibility that orders get canceled if export licenses fall through, and the sheer demand pressure that gives the company leverage to dictate terms.
The H200 represents Nvidia's current high-end AI training chip, successor to the H100 that powered the first wave of large language model training. Chinese tech giants—including Baidu, Alibaba, and Tencent—have been racing to secure supply as they build out domestic AI capabilities. But the US has progressively tightened export controls on advanced semiconductors, viewing AI chips as strategically sensitive technology.
The Export Control Maze
Nvidia has been here before. In late 2022, the US government restricted sales of the company's A100 and H100 chips to China. Nvidia responded by developing China-specific variants—the A800 and H800—designed to comply with the new rules by throttling certain performance characteristics. Then the US tightened restrictions again in October 2023, effectively banning those workaround chips too.
The H200 occupies an even more precarious position. It's more powerful than the H100, which means it almost certainly exceeds the performance thresholds set by current export controls. Whether Nvidia can sell it to China at all—or whether it would need yet another compliance-engineered variant—remains unclear.
Adding to the complexity: Beijing has its own approval processes and has grown increasingly assertive about controlling technology flows in both directions. Chinese regulators have delayed or blocked various tech deals in recent years, giving them de facto veto power over some transactions.
What This Signals for AI Hardware
The upfront payment demand reveals something important about where power sits in the AI chip market. Nvidia can make these demands because there's no credible alternative. Chinese chip companies like Huawei have made progress on domestic AI accelerators, but nothing matches Nvidia's CUDA ecosystem or raw performance.
For Chinese AI companies, the choice is stark: pay now for chips that might never arrive, or fall further behind in the AI race. Many will pay. The risk of missing out on H200 supply entirely outweighs the financial risk of prepayment.
For the broader industry, this is another data point in the decoupling of US and Chinese tech supply chains. What was once a highly integrated global semiconductor ecosystem is fracturing into parallel systems with limited interoperability.
The Bottom Line
Nvidia asking for money upfront isn't just about managing cash flow—it's about managing uncertainty in a market where the rules keep changing. The company is betting that Chinese customers value H200 access enough to accept terms they'd never tolerate in normal circumstances. So far, that bet appears correct.