Against a backdrop of ongoing U.S.‑China technology friction, U.S. President Donald Trump announced on the evening of December 8 via Truth Social a major policy shift: the U.S. government will allow NVIDIA to export its second‑generation high‑end AI chip H200 to China and will impose a 25% tariff on related sales. Trump said in his post that he had personally informed Chinese President Xi Jinping of the decision, who "responded positively." The move marks the first relaxation of the 2023 export ban on AI chips to China and is intended to balance national security with economic interests. NVIDIA's stock jumped nearly 3% in after‑hours trading and its market value topped $4.5 trillion. Media outlets including Reuters and NBC News quickly followed up, calling it a "pragmatic" turn in Trump's trade policy, but it has also drawn bipartisan questions in Congress: will it bolster China's AI military capabilities? At the end of 2025, amid an intense global AI race, the exemption is not only timely for NVIDIA but could reshape U.S.‑China tech supply chains, shifting AI hardware from "embargo walls" to a new balance of "conditional openness."
The H200 is NVIDIA's flagship product based on the Hopper architecture, launched in 2023. Positioned between the H100 and Blackwell B200 in performance, it is designed for AI training and inference and is equipped with 141 GB of HBM3e memory and 4.8 TB/s bandwidth. Since the Biden administration began export controls on China in 2022, the U.S. Department of Commerce (BIS) placed high‑end GPUs such as the H100/H200 on the Entity List, requiring special licenses to prevent potential military use. The restriction led to a roughly 40% drop in NVIDIA's fiscal‑2024 revenue from China, falling from a peak of $15 billion to $9 billion, and its China market share slid from 25% to 15%. In response, NVIDIA developed a cut‑down H20 (about one‑fifth the performance of the H100) and obtained export approval, but H20 was rejected by Chinese authorities, and Chinese firms such as Huawei and Alibaba Cloud turned to domestic alternatives like the Ascend 910B. After Trump took office, he inherited the Biden framework but emphasized the "art of the deal": on November 21 internal Commerce discussions surfaced, and Semafor cited sources saying lifting H200 was a "middle‑path" option in policy circles.
Trump's announcement was concise but consequential: "I have informed Xi Jinping that the U.S. will allow NVIDIA to export H200 products to approved customers in China and other countries, provided national security remains strong. Xi responded positively!" He also mentioned plans to extend similar export permissions to other "great American companies" such as Intel and AMD, with the Commerce Department "finalizing details." Key conditions include a 25% tariff (higher than the prior 15% applied to H20), BIS review of exports to ensure buyers are non‑military entities, and chips incorporating monitoring "backdoors" to prevent resale. NVIDIA CEO Jensen Huang responded on X: "Thankful for the President's leadership; this will help global AI innovation while protecting security." After the announcement, NVIDIA's after‑hours share price rose 2.5% and its market value recovered; analysts forecast a roughly $2 billion rebound in China revenue in fiscal 2026.
The deeper reasons behind the exemption reflect multiple pressures. First, economic considerations: China is NVIDIA's largest market and the ban cost the company more than $10 billion in 2024, which the Trump administration viewed as self‑harm. Trump had pledged on the campaign trail to "let American companies make money," and freeing H200 (about 18 months behind Blackwell) is framed as a security‑conditioned concession—powerful enough for Chinese AI training but not to surpass U.S. capabilities. Second, national security trade‑offs: BIS data reportedly show no evidence of military misuse from H20 exports, and some domestic Chinese chips such as Biren BR100 are said to reach only about 60% of H100 performance. The exemption would allow closer monitoring of supply chains and reduce incentives for a black market in H100s. Third, geopolitical signaling: direct communications between Trump and Xi and a promise not to escalate tech decoupling have been cited. Commerce Secretary ally Howard Lutnick reportedly pushed the case, arguing for balancing exports and embargoes to uphold U.S. standards. Nevertheless, there is significant congressional resistance: Senator Elizabeth Warren criticized the "back‑door deal," and the SAFE Chips Act proposal would ban H200 exports for 30 months.
China's official response was measured but positive. A foreign ministry spokesperson on December 9 said the country "welcomes a pragmatic U.S. move that can help stabilize China‑U.S. tech cooperation." The China Semiconductor Industry Association said H200 will fill a gap in high‑memory AI training but emphasized it "will not rely on a single source." Alibaba Cloud and Baidu have expressed purchase intentions, with the first H200 shipments expected in Q1 2026. Expert views are divided: CSIS analyst James Lewis called the 25% tariff "a clever design to earn foreign exchange while throttling high‑end flows," while the Brookings Institution warned the exemption "could accelerate China’s military‑civil fusion in AI and requires strengthened downstream controls."
The global AI ecosystem impact will be multifaceted. For NVIDIA it could be a "redeeming" move: the exemption may boost 2026 revenue by around 15%, and the stock rebound could help market value claw back toward the $5 trillion mark. For China's AI sector it offers breathing room: the H200's 141 GB memory will accelerate large‑model training and iterations such as DeepSeek V3.2, though domestic chips like the Ascend 910C (planned mass production in 2026) are expected to remain influential. Third, for U.S.‑China relations it is a litmus test: the exemption could ease the "chip war," but passage of the SAFE Chips Act would reverse the change. Global supply chains may benefit if AMD and Intel exports are also allowed, stimulating Asian assembly plants. Risks remain: a 25% tariff could raise costs for Chinese AI development and spur firms like Huawei toward self‑reliance, and the EU and Japan might emulate the approach, creating multilaterally coordinated controls.
Looking ahead, the exemption could open a new era of "conditional openness." The Commerce Department is expected to publish detailed rules in January, and NVIDIA reportedly has stocked 50,000 H200 units. In the AI "arms race," the H200's return is more than a chip story—it sits in a gray zone of U.S.‑China competition where balancing security and prosperity will help shape the 2026 tech landscape. Silicon Valley cheers, Washington is wary, Beijing is cautious: the H200 exemption may be "the beginning of a deal." Will the Chinese government and companies take the offer?