Nvidia Faces Policy, Supply Strain

Nvidia is pushing on two fronts at once: Jensen Huang wants to keep selling AI chips in China even as the company looks abroad for more power-hungry data-center gear. (economictimes.indiatimes.com) (asiae.co.kr) Huang said on an April 15 podcast that U.S. export curbs have not stopped China’s artificial-intelligence push and have instead helped local rivals such as Huawei. He argued Nvidia should keep Chinese developers on its CUDA software platform rather than push them toward domestic alternatives. (economictimes.indiatimes.com) The policy fight comes after two years of shifting rules on China-bound chips. Reuters reported in August 2025 that Washington started issuing licenses for Nvidia’s H20 chip after reversing an April 2025 ban, while exports of Nvidia’s other advanced AI chips to China remained restricted. (finance.yahoo.com) China is still a meaningful market even after those curbs. Nvidia said in May 2025 that H20 brought in $4.6 billion in first-quarter sales and that China accounted for 12.5% of company revenue in that period; by February 2026, Chief Financial Officer Colette Kress said Nvidia had still generated no revenue from newly approved H200 sales to China. (finance.yahoo.com) (cnbc.com) At the same time, Nvidia’s supply problem is no longer just about getting enough graphics processors out of factories. South Korea’s Asia Business Daily reported on April 22 that Korean power-equipment makers are aligning with Nvidia’s roadmap for 800-volt direct-current data centers, a design that would change how electricity moves inside AI server sites. (asiae.co.kr) That voltage shift points to a bottleneck outside the chip package itself. The Korean report said companies including LS Electric, HD Hyundai Electric and Hyosung Heavy Industries are preparing for higher-voltage distribution, direct-current conversion gear, transformers and circuit breakers as AI facilities draw more power. (asiae.co.kr) The backdrop is Nvidia’s scale. The company reported $215.9 billion in fiscal 2026 revenue on February 25, with $62.3 billion in fourth-quarter data-center revenue alone, numbers that show how much of the AI buildout now depends on power systems, networking and site design as well as semiconductors. (nvidianews.nvidia.com) Nvidia has said export controls also carry a direct cost. Reuters reported the company disclosed a $5.5 billion charge tied to H20 restrictions in April 2025, then later said the curbs would cut $8 billion from a July-quarter sales outlook before some licenses resumed. (finance.yahoo.com) Chinese competitors are advancing while Nvidia waits on policy. Kress said in February that Chinese rivals, backed by recent initial public offerings, “have the potential to disrupt the structure of the global AI industry over the long-term,” as Nvidia still lacked fresh China revenue from approved H200 shipments. (cnbc.com) So Nvidia’s China debate and its South Korea power outreach are part of the same buildout. One side runs through export licenses and Washington; the other runs through substations, transformers and how to feed electricity to ever-larger AI clusters. (finance.yahoo.com) (asiae.co.kr)