In the past few days, a surprising niche industry has gained global attention: the repair of Nvidia’s most powerful artificial‑intelligence chips in China. Demand for these repair services is rising rapidly even though the United States banned the export of Nvidia’s H100 and A100 GPUs to China back in September 2022. Shenzhen repair shops report that hundreds of these high‑end GPUs—smuggled into China and now worn out from intensive use—are being brought back to life each month. Technicians charge between 10 000 and 20 000 yuan (roughly US$1 400 – 2 800) per chip, running them through battery after battery of software tests and component replacement to keep data‑center servers humming. Some firms have built testing rooms that can accommodate 256 servers for this purpose.
The ban on selling the H100 and A100 to China created a supply vacuum just as the global artificial‑intelligence boom put enormous pressure on hardware. To keep their AI projects running, many Chinese companies turned to underground markets. As a result, thousands of Nvidia GPUs have been running non‑stop for years, accelerating wear‑and‑tear and driving a surge in repair requests. One repair shop now fixes up to 500 Nvidia AI chips per month; another handles about 200 and charges around 10 percent of the chip’s original price for each repair. Industry insiders say some data‑centers rotate parts across servers to keep systems operational.
Another factor: alternatives are scarce and expensive. Nvidia’s newly released H20 chip—a lower‑performance design created to comply with U.S. export rules—costs over 1 million yuan per eight‑GPU server, and the still‑more powerful B200 chip costs over 3 million yuan per server. For AI training workloads, companies still prefer the banned H100. Meanwhile, because the sale of high‑end chips is blocked, Nvidia can’t legally provide warranty repairs in China, leaving unofficial shops to fill the void. The situation has grown so significant that U.S. lawmakers have drafted legislation requiring AI chips to include location‑tracking mechanisms to deter smuggling.
The repair boom underscores how competitive and volatile the chip industry has become. On the same day news broke about the underground Nvidia repair market, Intel’s new CEO Lip‑Bu Tan revealed plans to shrink the company’s workforce by 22 percent and take a disciplined approach to factory construction. The long‑time U.S. chip pioneer has struggled with years of missteps and is now warning that its next‑generation 14A manufacturing process may be shelved entirely unless it secures a major customer. Tan plans to slow construction work on new factories, halt some planned facilities in Europe and Asia, and personally review chip designs. The restructuring highlights how difficult it is for even established players to keep pace with explosive demand for AI chips.
For businesses in North Carolina and beyond, these developments are reminders that the hardware powering artificial intelligence is subject to geopolitical forces and corporate strategy shifts. Proactive planning—through diversification of suppliers, attention to maintenance and flexible workload placement—can help mitigate risks and ensure that AI initiatives continue to deliver value despite a turbulent chip landscape.
Sources: