In a significant policy reversal, the U.S. Commerce Department updated its export criteria in early January to permit Nvidia’s H200 AI accelerator to be sold into China. This approval follows months of negotiations and comes after Nvidia disclosed that the H200 write-down related to prior restrictions totaled $5.5 billion. The greenlight is expected to unlock up to $10 billion in potential annual revenue, according to company guidance, restoring a critical growth avenue in the world’s second-largest AI market. Major Chinese cloud service providers and research institutes are already in advanced discussions to integrate the H200 into inference clusters for natural language processing and recommendation engines, potentially driving multi-year deployment contracts.

At CES 2026, Nvidia unveiled its new Rubin data-center platform, which combines proprietary Vera CPUs, Rubin GPUs and next-generation NVLink networking. Available in rack-scale configurations of 72 GPUs and smaller eight-GPU systems in H2 2026, Rubin is designed to slash AI inference token costs by up to 90% versus the existing Blackwell lineup. Industry forecasts estimate that this cost reduction could accelerate corporate AI rollouts, shifting the ROI calculus: with 95% of enterprise pilots currently failing to generate meaningful returns, the platform’s efficiency gains may drive a surge of long-term contracts from hyperscalers and large financial institutions.

Despite the U.S. export approval, Chinese customs authorities have intermittently detained H200 shipments, citing compliance reviews. These actions underscore ongoing geopolitical risk: Customs agents in Shenzhen reportedly held multiple consignments in the first week of January, delaying delivery by up to two months. Nvidia’s management has warned investors that continued disruptions could suppress revenue growth by up to 8 percent in fiscal 2026. The company is diversifying its supply chain—investing $165 billion in TSMC’s Arizona fab expansion—and has raised GPU prices by 10–15 percent to offset tariff-induced cost pressures.

In its Q3 earnings report, Nvidia posted record revenue of $57 billion, with the Data Center segment contributing $51.2 billion, a 66 percent year-over-year increase. The company’s operating expenses rose 36 percent to $5.8 billion, reflecting elevated R&D and capex spending of $3.2 billion for new fab capacity. On the sell-side, 59 out of 64 analysts continue to recommend shares as a buy, with a consensus one-year price target implying roughly 36 percent upside. Their forecasts hinge on sustained Blackwell and Rubin demand, with full-year revenue guidance of $170 billion for fiscal 2026—30 percent above the prior year—underscoring confidence in Nvidia’s AI leadership.