Trump Approves Nvidia H200 Exports to China with 25% Tariff, Keeping Top AI Chips Banned

US President Donald Trump has authorized the export of Nvidia’s H200 AI chips to China, replacing a comprehensive ban with a transactional tariff strategy. Announced late Monday, the move allows sales of the previous-generation hardware subject to a 25% “import tax” collected from Taiwan.

While the H200 gets a green light, the administration is keeping a tight lid on cutting-edge silicon. Trump explicitly confirmed that Nvidia’s powerful Blackwell architecture and the upcoming Rubin series remain strictly off-limits to Chinese buyers to protect U.S. national security.

Shifting to economic leverage defies bipartisan pressure from Capitol Hill. Just days ago, Senators introduced the Secure and Feasible Exports (SAFE) Chips Act, seeking a complete 30-month embargo on advanced semiconductor exports to Beijing.

The Transactional Pivot: Tariffs Over Blockades

Breaking with the “small yard, high fence” doctrine of the previous administration, the White House has opted for a revenue-focused trade policy. Trump’s announcement on Truth Social marks a clear shift, prioritizing economic extraction over absolute denial of technology.

“We will protect National Security, create American Jobs, and keep America’s lead in AI,” Trump stated.

Under the directive, exporters may ship Nvidia H200 processors, chips that are approximately 18 months behind the cutting edge, to “vetted” commercial customers in China. Unlike standard tariffs paid by importers, this 25% fee is described as an “import tax” collected from Taiwan before shipment.

Such a unique structure is designed to capture revenue for the U.S. Treasury. The administration has devised a complex logistical chain to enforce this levy, effectively treating the transaction as a multi-stage import-export loop. According to details reported by Reuters, the H200 processors manufactured by TSMC in Taiwan will not be shipped directly to Chinese clients.

Instead, the hardware must first be routed to the United States. Upon entry, the 25% fee is collected as an import tax. Once on American soil, the chips undergo a mandatory security review by U.S. officials to ensure compliance with technical specifications. Only after clearing this physical inspection and settling the financial penalty are the processors cleared for re-export to their final destination in China.

With this new framework, the U.S. Treasury effectively becomes a stakeholder in the semiconductor trade, collecting a premium on every unit shipped. Nvidia immediately signaled compliance, framing the decision as a strategic win for American manufacturing rather than a capitulation on security.

“Offering H200 to approved commercial customers, vetted by the Department of Commerce, strikes a thoughtful balance that is great for America,” an Nvidia spokesperson said.

Nvidia’s statement carefully avoids mentioning the specific tariff rate, focusing instead on the “thoughtful balance” of the policy. Crucially, the President made it clear that the most powerful hardware remains off the table, ensuring a permanent technology gap.

The President’s statement specifically identified future roadmap items to signal a long-term restriction. This distinction aims to satisfy national security hawks while opening a revenue stream from legacy technology.

“NVIDIA’s U.S. Customers are already moving forward with their incredible, highly advanced Blackwell chips, and soon, Rubin, neither of which are part of this deal,” Trump added.

Legislative Revolt: The SAFE Chips Act

Executive action has immediately collided with legislative intent, specifically the SAFE Chips Act introduced on December 4. Sponsored by Senators Pete Ricketts (R-NE) and Chris Coons (D-DE), the bill proposes a diametrically opposite approach: a statutory 30-month total ban on advanced AI chip exports.

Lawmakers argue that any high-performance silicon, even previous-generation hardware like the H200, accelerates the People’s Liberation Army’s (PLA) AI capabilities.

Such a disagreement sets up a constitutional showdown between the President’s authority to regulate commerce and Congress’s power to legislate national security controls.

Supporters of the bill, including the GAIN AI Act support coalition mentioned in earlier reporting, may intensify lobbying efforts to override the executive order via the NDAA.

While the White House claims alignment on security, legislative hawks view the sale of *any* training-capable chips as a loophole that tariffs cannot close.

“As for the most advanced chips, the ‘Blackwell’ chip, that’s not something we’re interested in selling to China at this time,” White House spokeswoman Karoline Leavitt previously stated, a position that critics argued did not go far enough.

Internal U.S. conflict creates uncertainty for semiconductor firms, who must navigate contradictory signals from the executive and legislative branches.

Market Reality: Too Little, Too Late?

While Washington debates export permissions, the ground reality in China has significantly shifted since the initial bans were imposed. Nvidia’s CEO has been vocal about the damage done to the company’s standing in the region.

“We went from 95% market share to zero percent, and so, I can’t imagine any policymaker thinking that that’s a good idea,” Jensen Huang, CEO of Nvidia, said.

Huang’s lament about losing 95% market share in China reflects a historical baseline that may no longer exist to be reclaimed. On November 5, Beijing mandated that all new state-funded data centers must exclusively use domestically designed and manufactured chips.

Such a mandate effectively locks Nvidia out of the public sector regardless of U.S. policy. Complicating the administration’s revenue goals is the reality of domestic substitution.

Huawei has rapidly filled the void with its Ascend 910C, a processor that benchmarks suggest offers 2x the performance of the crippled Nvidia H20 and price parity with the H200.

“It is foolish to underestimate the might of China and the incredible competitive spirit of Huawei. This is a company with extraordinary technology,” Huang previously warned.

Nvidia’s leadership has consistently cautioned stakeholders against dismissing the capabilities of domestic Chinese firms. The 25% tariff further erodes Nvidia’s competitiveness; adding a premium to an already expensive imported product makes the Ascend series more attractive to private Chinese firms sensitive to cost.

Independent industry analysts have not yet issued formal projections on how the tariff will impact H200 adoption rates, but questions remain whether the “vetted commercial customers” referenced by the White House exist in sufficient numbers to move the needle for Nvidia’s bottom line.

The ‘Import Tax’ Mechanism and Legal Complexities

Mechanically, the 25% fee raises legal and logistical questions regarding international trade law. Describing the fee as an “import tax from Taiwan” suggests a levy imposed on the manufacturing stage (TSMC) rather than the point of sale, potentially complicating relations with Taipei.

Commerce Secretary Lutnick has previously articulated a strategy of controlled obsolescence, arguing that Beijing should only access trailing-edge technology.

“We don’t sell them our best stuff… The fourth one down, we want to keep China using it,” Lutnick said.

Beijing’s official response has been cautious, emphasizing “mutual benefit” while state media outlets criticize the move as “shortsighted” protectionism. If the fee is passed on to customers, the total cost of ownership for an H200 cluster in China could exceed domestic alternatives by 30-40%.

This pricing disparity effectively positions the H200 as a luxury good for entities that specifically require CUDA compatibility, rather than a mass-market infrastructure component.