AI access dispute widens

Nvidia Chief Executive Jensen Huang is pressing Washington to ease artificial intelligence chip curbs on China just as access to the chips themselves gets more expensive. (bloomberg.com) (trendradarx.com) Huang said on a podcast posted April 15 that the United States should keep “research dialogue” with China on artificial intelligence safety, arguing that treating China only as an adversary would backfire. He tied that warning to Anthropic’s recent “Mythos” model and said China already has the talent, power supply and data-center capacity to build advanced systems. (bloomberg.com) (benzinga.com) The policy fight sits on top of a rules fight in Washington. The Commerce Department’s Bureau of Industry and Security published its “Framework for Artificial Intelligence Diffusion” on January 15, 2025, expanding controls on advanced computing chips and adding new controls on some artificial intelligence model weights. (federalregister.gov) At the same time, Reuters reported on April 15 that Chinese officials have held initial talks with solar-equipment suppliers about possibly restricting exports of the most advanced solar-panel manufacturing technology to the United States. Reuters said the discussions are early-stage and would widen retaliation beyond semiconductors into clean-energy machinery. (finance.yahoo.com) That would extend a trade dispute that already covers the hardware used to train artificial intelligence models. The same export-control system that limits high-end chips can also reshape where cloud companies build data centers, which customers get priority access and how much startups pay to rent computing time. (federalregister.gov) (stblaw.com) Cloud rental prices show that squeeze in real time. Recent market trackers cited by several April 2026 reports show one-year Nvidia H100 rental pricing rising from about $1.70 per graphics processing unit hour in October 2025 to roughly $2.35 by March 2026, while some Blackwell rentals climbed from $2.75 to $4.08 an hour in about two months. (stockminded.com) (edgen.tech) Reports circulating this week say Amazon, Microsoft, Google and Meta have been raising prices or steering customers toward longer contracts for Nvidia capacity. Those reports point to the same bottleneck: the biggest cloud operators control much of the supply of the graphics processing units used to train and run large artificial intelligence models. (trendradarx.com) (edgen.tech) Nvidia has argued for months that broad curbs cost United States companies sales without stopping Chinese development. Huang said in 2025 that earlier controls were a “failure,” and his new call for dialogue shows the company is still pushing against a strategy built around denial of access. (nbcnews.com) (bloomberg.com) For artificial intelligence companies outside Big Tech, the result is a narrower path: fewer export destinations, tighter cloud supply and higher hourly prices for the chips that matter most. Huang is asking for more dialogue across the Pacific just as the market is making access more expensive on both sides. (bloomberg.com) (trendradarx.com)