Nvidia faces antitrust, China chip limits

France’s competition authority said on Thursday it is nearing the end of its antitrust investigation into Nvidia, adding a European regulatory front to the chipmaker’s already complicated position in the global AI market. On the same day, reports citing people familiar with the matter said China plans to let some of its biggest AI companies buy limited quantities of Nvidia’s H200 chips after months of tighter controls. Those developments arrived as Nvidia broadened ties with other AI hardware and software groups, including d-Matrix and Hugging Face. Together, the moves show how Nvidia’s business is being shaped at once by regulators, export controls and partnership deals. ### How far has the French antitrust case progressed? Paris-based Autorité de la concurrence said the Nvidia investigation is close to completion, according to Reuters. Umberto Berkani, the authority’s general rapporteur, told reporters on July 9 that the agency was nearing the end of the investigation into alleged anti-competitive practices. Reuters reported that the end of the investigation does not necessarily mean Nvidia will face sanctions. (money.usnews.com) France’s watchdog has been examining Nvidia’s role in AI-related markets as governments and regulators look more closely at concentration in advanced chips and the software stack around them. Reuters did not report a final timetable for a decision, and the authority did not say on Thursday whether it would bring formal charges or close the matter without penalties. (money.usnews.com) ### What is China reportedly preparing to allow? China is planning to let leading domestic AI companies buy a limited number of Nvidia H200 chips, The Information reported, and Reuters and Bloomberg summarized the report on July 8 and July 9. Reuters said Chinese officials had told Alibaba, ByteDance and DeepSeek in recent weeks that they may soon receive permission to buy some H200 chips. (money.usnews.com) The proposed approvals would still leave total purchases below 200,000 chips, according to Reuters’ account of The Information report. Reuters also said companies would need to request approval and specify the number of chips sought and the reason for the purchase. That would amount to a controlled channel for a small pool of advanced Nvidia hardware rather than a broad reopening of the market. (msn.com) ### Why does the H200 matter more than an ordinary chip sale? Nvidia’s H200 is one of the company’s high-end AI chips used for training and inference workloads, making any relaxation in access important for Chinese model developers that have faced shortages. Reuters said the possible approvals would offset a supply gap caused by strong demand for AI chips. Bloomberg’s summary of The Information report also described the move as an easing of restrictions on sought-after U.S. technology. (msn.com) Chinese companies have been navigating both U.S. export limits and domestic controls on which Nvidia products can be purchased. The reported H200 permissions suggest that any reopening would be narrow, company-specific and subject to case-by-case approval, based on the reporting cited above. (msn.com) ### Why is Nvidia partnering with companies that could become rivals? The Information reported on July 8 that Nvidia has been responding to a growing list of AI server-chip competitors by partnering with them instead of trying only to displace them. Its report cited d-Matrix as the latest example, saying Nvidia and the startup were combining their hardware in a new system to run AI models. (msn.com) That approach gives Nvidia another way to stay embedded in AI systems even as customers test alternatives for inference and lower-cost deployments. The Information described the strategy as a response to rising competition in AI server chips. ### Where does Hugging Face fit into this picture? Nvidia and Hugging Face said this week they are adding Nvidia robotics models and frameworks to LeRobot, Hugging Face’s open-source robotics library. (theinformation.com) Nvidia said the additions include Isaac GR00T 1.7, Isaac Teleop, datasets and workflows, with Nvidia Cosmos 3 integration planned later. The announcement was focused on robotics software rather than data-center chips, but it showed Nvidia expanding ties with a widely used open-source AI platform. Nvidia’s blog said the new integrations are intended to give developers shared tools for collecting robot data, training models and deploying workflows. July 9 leaves three separate next steps in view: France’s competition authority has yet to say whether it will file charges, Chinese officials have not publicly announced any H200 approval process, and Nvidia said further LeRobot integrations, including Cosmos 3, are still planned. (blogs.nvidia.com) (money.usnews.com)