Nvidia’s Huang challenges US chip curbs after H20 restrictions | Magica
Huang calls US chip curbs a failure as Nvidia fights for China access
Editorial Team
••📖7 min read
Nvidia chief Jensen Huang argued in May 2025 that US chip restrictions were strengthening Chinese rivals. Filings show a substantial commercial stake, but lost sales and market share alone cannot establish whether the controls failed their security purpose.
Nvidia co-founder and CEO Jensen Huang called US chip export controls a failure at Computex on May 21, 2025.
The April H20 restriction threatened billions in charges after Nvidia had designed the chip for earlier US limits.
China revenue had grown before that restriction; Huang’s strategic warning concerned rival software ecosystems as well as chip sales.
Jensen Huang, Nvidia’s co-founder and chief executive since its inception in 1993, argued on May 21, 2025, that US restrictions on advanced chips were accelerating Chinese competition. Speaking at the Computex technology show in Taipei, the head of the American graphics and AI computing company called export controls a “failure,” according to an account of his remarks.
His intervention followed Washington’s April decision to require export licenses for Nvidia’s H20, a data-center graphics processing unit designed to comply with earlier restrictions. Nvidia faced a large financial charge. Huang’s broader argument was that excluding American suppliers would give Chinese competitors room to establish computing platforms that could spread internationally.
The distinction matters: controls intended to curb China’s military advancement can hurt Nvidia without necessarily failing that security objective. Huang offered a mechanism for long-term damage to American technology leadership, but his company’s losses do not measure the security benefit purchased in exchange.
Nvidia’s company-reported H20 licensing notifications and Commerce’s separate AI Diffusion Rule rollback. Chart: Magica. Source: US Commerce Department.
The H20 restriction put another China product at risk
Successive US measures had already narrowed Nvidia’s access. Its fiscal 2025 annual report describes licensing requirements imposed on its A100 and H100 chips in August 2022, followed by expanded restrictions in October 2023. The company then introduced products specifically for China that did not require export licenses.
JetBrains IDEs 2026.3 bundle Air for running coding agents and reviewing their changes, while the standalone Air app is in maintenance mode. Existing provider subscriptions remain usable, but billing, security controls and execution options vary by integration.
H20 was a lower-powered version of other Nvidia chips built to meet the earlier limits. In its April 15 filing, Nvidia said the US government had notified it on April 9 that H20 exports to China, including Hong Kong and Macau, and certain other destinations or companies required licenses. The requirement also covered circuits meeting specified H20 bandwidth characteristics. On April 14, officials told Nvidia it would remain in effect indefinitely.
The government’s stated reason, as relayed in Nvidia’s filing, was the risk that the covered products could be used in, or diverted to, a supercomputer in China. Designing a chip to meet one set of thresholds had not secured continued access under the next.
The filing forecast up to approximately $5.5 billion in charges for H20 inventory, purchase commitments and related reserves for the fiscal quarter ending April 27. In a May interview, Huang described $5.5 billion as already written off and said Nvidia had walked away from $15 billion in sales. He supplied no time window for the sales figure. His description should not be confused with the filing’s forecast or treated as a final reported expense.
Company-reported revenue by customer billing location, before the April 2025 H20 restriction. China includes Hong Kong; figures cover all products. Chart: Magica, from Nvidia’s fiscal 2025 annual report. Source: Nvidia fiscal 2025 annual report.
China sales grew while their weight in Nvidia declined
Huang had a substantial commercial interest in reopening access. Nvidia’s February earnings release reported worldwide fiscal 2025 revenue of $130.5 billion, up 114%, with data-center revenue reaching $115.2 billion, up 142%. AI infrastructure had become the core of its business.
The annual report provides a more specific baseline for its China exposure:
Reported measure
Fiscal 2024, ended Jan. 28, 2024
Fiscal 2025, ended Jan. 26, 2025
Revenue billed to China, including Hong Kong
$10.3 billion
$17.1 billion
Total worldwide revenue
$60.9 billion
$130.5 billion
China and Hong Kong share of total revenue
16.9%
13.1%
The shares are calculated from the filing’s unrounded figures. Geographic revenue reflects customer billing addresses; shipping and end-customer locations can differ. These figures cover all Nvidia products and precede the April H20 measure, so they do not measure China’s AI-chip market or the restriction’s eventual impact.
They show why falling share needs a denominator. Nvidia also said its China data-center revenue grew in fiscal 2025, while China’s proportion of worldwide data-center revenue remained well below levels before the October 2023 controls.
At Computex, Huang offered a different comparison: he said Nvidia’s share of China’s AI-chip market had fallen from almost 95% at the beginning of the Biden administration to 50%. The reported remarks supplied no measurement method. That was Huang’s market-share claim, separate from the company’s geographic revenue disclosures. His estimate in the interview of a roughly $50 billion annual China market was likewise a market opportunity estimate, not Nvidia’s booked revenue.
The asset at stake was a computing platform
Nvidia’s position rests on more than selling processors. It introduced CUDA in 2006, opening its GPUs’ parallel processing capabilities to applications beyond graphics. The annual report describes CUDA as the programming foundation of a platform spanning chips, networking and software, and says more than 5.9 million developers worldwide used CUDA and its other software tools.
That history explains Huang’s warning: an installed hardware base attracts developers; their applications make the platform more useful, encouraging further adoption. In the interview, he agreed that a Chinese alternative to CUDA was the long-term threat. In his account, lost sales could also mean losing the developers who shape future demand.
Huawei, the Chinese technology company competing in AI chips, was identified as a beneficiary of the pressure in a separate account of Huang’s criticism. This was established competition: Nvidia’s annual report already listed Huawei among its processor rivals and among companies developing their own accelerated-computing hardware and software. It also identified competing internal platforms at Chinese cloud companies Alibaba and Baidu.
Huang’s own interview limits the claim that restrictions created Chinese progress. He praised Huawei’s capabilities and Chinese researchers and said he had expected China to advance regardless. Restrictions could encourage customers to substitute rival systems without being the original cause of those systems’ development.
Substitution also need not mean equal economics. Huang argued that lower performance per watt requires more power and infrastructure for a given computing output, even where national energy supplies are abundant: each data center still has a finite power budget. This was the vendor’s argument, rather than an independently measured cost comparison between H20 and Huawei chips. Customers denied access to Nvidia could accept higher costs to obtain a usable alternative.
The next policy must reconcile access with diversion risk
The Trump administration’s May rollback addressed a separate measure. On May 13, the Commerce Department announced that it would rescind the AI Diffusion Rule, issued in January to govern international access to advanced AI technology. Its compliance requirements had been due to take effect on May 15. The department instructed enforcement officials not to enforce it and said a regulation formalizing rescission and a replacement rule would follow.
That announcement did not reverse H20 licensing. Commerce simultaneously warned about risks from using specific Huawei Ascend AI chips. China’s commerce ministry called the US warning bullying and protectionism and said organizations or individuals helping enforce such measures could violate Chinese law, according to the report on its response.
Huang welcomed wider access to American technology, describing partnerships for AI infrastructure in Saudi Arabia and Abu Dhabi in his interview, while pressing for competition in China. Washington’s pending replacement policy therefore faced two tests: whether it could prevent access to sensitive computing capacity, and whether restrictions would push customers toward viable rival hardware and software.
Evidence on diversion, the availability and operating costs of substitutes, and their developer adoption would help resolve that trade-off. Nvidia’s lost business establishes a commercial cost; it cannot alone show whether that cost bought a security benefit.
Amazon’s shares and convertible notes could be worth $271.8 billion at the top of Anthropic’s proposed IPO range. Existing accounting gains, a 12% ownership cap and Anthropic’s computing commitments shape what that paper value means.
Reported demand has Anthropic’s bankers considering a price above its filed $70–$78 range. The combined financing could exceed $100 billion, but existing cash, uneven operating results and long-term computing commitments complicate the valuation test.
The British AI chip designer has secured private and UK government backing before its planned 2027 commercial launch, putting a multibillion-dollar valuation on its promise of faster, cheaper inference.
Zhang Yiming reportedly told staff that ByteDance needs years to challenge leading US AI labs, even as the TikTok parent prepares a personal agent for release within weeks. Its consumer reach and video strengths have yet to translate into frontier language models.
Helsing has unveiled a military intelligence platform promising European control over data, models and operations. Germany's evaluation of roughly 30 European systems makes procurement and demonstrated performance the next tests.
Public safety leaders warn that synthetic voices and background sounds make hoax emergency calls harder to assess. A voluntary survey exposes preparedness gaps, but does not establish a national AI-swatting rate or measure growth over time.
Fortinet's November security disclosures include AI-assisted findings, but some affected software branches still await fixes. Earlier testing put token costs at about $540 per true positive, excluding the broader work of reviewing findings and securing customer deployments.
Tamil Nadu told the Madras High Court that accounts containing about 2.22 million rupees had been frozen after its library-procurement order, while a criminal case was still being prepared. The dispute tests safeguards ordered in 2024; AI involvement remains unconfirmed.
Unions withdrew objections to Google’s proposed $10 million purchase of Spirit Airlines data after securing employee-record exclusions and independent privacy review. The agreement follows earlier passenger-data protections; court approval and verification remain ahead.
Baird cut Nvidia to Neutral and its price target to $200 from $500. Its case combines slower forecast AI spending growth with China-linked inventory exposure and uncertainty over founder and CEO Jensen Huang’s tenure.
ByteDance is reportedly accelerating plans to combine its Spell project with Doubao. Existing travel bookings reveal a transaction-fee business, while the phone assistant's history shows how access to other apps can constrain execution.
The Office for Students wants more AI-free assessment and better AI skills teaching, backed by existing quality rules. Its research raises doubts about unsupervised coursework, but universities say assessment reform is already under way.
Finland has recognised Tesla’s provisional Dutch approval after changes to branding and speed controls. Drivers remain responsible, and the national decision does not commit Finland to backing EU-wide authorisation.
Grok Bot can connect to internal MCP servers through Tailscale without exposing them to the public internet. User-based authorization preserves network controls, but administrators must configure how narrowly the assistant can reach into their systems.
Daylight’s Goldman Sachs Alternatives-led Series B backs security agents supported by human engineers. The young company reports rapid growth, but has not disclosed the economics behind its automation promise.
Primordial Labs raised $18 million to widen deployment of Anura, its military command software. Existing contracts and demonstrations give it a foothold, but integration counts do not establish sustained use or lower operating costs.
Northwestern Polytechnical University bought two Nvidia H200 processors for 874,700 yuan and awarded five contracts for rented Nvidia computing, procurement records show. Separate US restrictions on the defense-linked institution remained in place despite January’s conditional opening for H200 exports to China.
The North Dakota Republican says allegations involving a suspected Chinese intelligence operative make Huang vulnerable to coercion. Nvidia’s board is backing its founder while independent directors investigate, with investors facing both security and leadership risks.
Scott Bessent says Washington and Beijing are establishing a channel for serious AI incidents, following their September agreement. Its usefulness depends on unresolved reporting procedures and participation by the companies that detect model failures.
