Chip war: The United States authorizes the sale of H200 to 10 Chinese companies, but Beijing blocks deliveries
🔎 Washington says yes, Beijing says no
On May 14, 2026, an unprecedented situation occurred in the Sino-American tech war: the United States authorized the sale of Nvidia H200 chips to around ten Chinese companies, including Alibaba, ByteDance, and Tencent. But Beijing unilaterally blocked all deliveries. This reversal illustrates the growing absurdity of a chip war where both sides are getting tangled up in their own restrictions.
On the same day, Jensen Huang, CEO of Nvidia, boarded Air Force One in Alaska to join the Trump-Xi summit in Beijing. A powerful image: the head of the world's most strategic chipmaker, a last-minute guest of the US president, literally becoming a geopolitical bargaining chip. The open source LLM war depends directly on these chips. Without H200, Chinese models struggle to compete. With them, they accelerate. Hence this paradox: both sides want to control the flow, and neither wants to be the first to yield.
The Essentials
- Washington authorized 10 Chinese companies (Alibaba, ByteDance, Tencent, and others) to buy Nvidia H200s, but strict conditions on non-military use and the certification of US stocks are blocking actual sales.
- Beijing is unilaterally blocking deliveries, in a context of retaliation following China's previous blocking of H20 imports.
- Jensen Huang joined Trump on Air Force One in Anchorage for the Beijing summit, alongside a delegation of about a dozen US CEOs (Meta, Visa, Cisco, Qualcomm, Micron).
- The H200 is Nvidia's second most powerful AI chip, essential for training large language models.
- This stalemate shows that the chip war is no longer unilateral: China has its own levers to block.
Recommended Tools
| Hostinger | Web hosting for deploying AI apps | Starting at €2.99 (May 2026, check on hostinger.com) | AI developers and startups |
What happened on May 14, 2026
Everything accelerated within 48 hours. On May 12, the New York Times revealed that Jensen Huang had received a last-minute call to join the presidential trip to Beijing. On May 13, Bloomberg confirmed that Huang had boarded Air Force One during the stopover in Anchorage, Alaska. On May 14, details of the H200 authorization leaked to the press.
According to CNBC, the Trump administration approved export licenses for 10 Chinese companies. Benzinga specifies that Alibaba and Tencent are among the beneficiaries, but that Beijing immediately halted physical deliveries.
The parallel with the H20 is crucial. China had already blocked H20 imports — a degraded version of the H100 designed specifically to comply with US sanctions. Beijing's message is clear: if Washington imposes hobbled chips, China refuses to play along. The H200, although more powerful than the H20, remains subject to export conditions that effectively make it partially hobbled in practice.
The 10 authorized companies: who is on the list
Sources agree on the major names but diverge on the complete list. What is confirmed by TrendForce and CNBC:
- Alibaba — cloud leader in China, developer of proprietary models
- ByteDance — parent of TikTok, massive investment in generative AI
- Tencent — cloud and AI, direct competitor of Alibaba
- Lenovo — mentioned in the initial leaks, hardware integration
The other companies have not been formally identified in the available sources. VellaTimes indicates that DeepSeek a reçu un accord conditionnel, but with additional restrictions and a final decision from Beijing still pending.
What stands out is the profile of the beneficiaries. They are all cloud and AI giants. No military labs, no directly sanctioned entities. Washington calculated that targeting these companies would jeopardize American competitiveness without any real geopolitical gain — a reversal from the hardline stance of 2023-2024.
The H200: why this chip is strategic
The H200 is not an ordinary chip. It is the second most powerful in the Nvidia lineup, behind the B200 (Blackwell). It is designed specifically for training and inference of large language models, with 141 GB of HBM3e memory and a bandwidth of 4.8 TB/s.
To understand the stakes, one must look at who depends on it. Models like DeepSeek V4 Pro (overall score of 88, 84 in agentic on the June 2025 benchmark) or Moonshot AI's Kimi K2.6 (84 overall, 88.1 in agentic in self-host) require clusters of H200s to be trained at scale. Without access to these chips, Chinese companies must fall back on less performant domestic alternatives or bypass restrictions through opaque supply chains.
The following table positions the H200 in the current ecosystem:
| Nvidia Chip | China export status (May 2026) | Positioning |
|---|---|---|
| H100 | Banned since October 2022 | 2023 reference, totally blocked |
| H20 | Authorized but blocked by Beijing | Degraded version, rejected by China |
| H200 | Authorized for 10 firms, blocked by Beijing | Second most powerful, gray area |
| B200 | Banned | Latest generation, no Chinese access |
Huang's Air Force One Trip: A Geopolitical Signal
Jensen Huang wasn't invited by chance. According to the New York Times, the call came at the last minute. Huang boarded in Anchorage, joined by a dozen U.S. CEOs. Yahoo Finance confirms the scene: Nvidia's CEO seen climbing the stairs of Air Force One, a near-unprecedented moment for a tech executive.
The delegation included heavy hitters: Elon Musk (Meta), executives from Visa, Cisco, Qualcomm, and Micron. Micron's presence is particularly significant — it's a direct competitor to Samsung and SK Hynix in HBM memory, exactly the key component of the H200.
According to GoldSea, Huang's inclusion suggests the H200 has become a genuine "bargaining chip" — a negotiating token in Trump-Xi discussions. The report notes that Washington hawks have expressed concerns over this instrumentalization of export policy.
Politico, in its analysis Trump's disappearing China hawks, describes a retreat of anti-China voices within the administration. The H200 authorization, initially granted in December 2025 and then materialized through licenses in early 2026 according to Bloomberg, fits into this dynamic of commercial de-escalation.
Why Beijing is blocking deliveries
This is the crux of the problem. Washington is giving the green light, but Beijing says no. Why?
The main reason is the H20 precedent. The United States had authorized the export of a degraded version of the H100, thinking it was making a "reasonable" gesture. China perceived this as condescending: accepting intentionally hobbled chips meant acknowledging American superiority and accepting a status of technological second-class citizenship.
The H200, although more powerful, remains subject to conditions that make it unacceptable to Beijing. According to TrendForce, US rules require proof of non-military use from each buyer, and Nvidia must certify that its US inventories remain sufficient after exports. These conditions are perceived by China as interference in its economic sovereignty.
There is also an industrial calculation. By blocking the H200, Beijing is accelerating the development of domestic chips. Chinese manufacturers like Huawei (Ascend 910C) and Cambricon are gaining market time. Every month without the American H200 is another month for local alternatives.
Finally, it is a negotiating lever. The Business Times reports that Trump brought Huang to Anchorage precisely in the hope of unblocking the situation. Beijing knows that the H200 is the trump card of this summit.
The broader context: Taiwan, weapons, and tech
The Trump-Xi summit is not just about chips. The Guardian points out that Trump had authorized an $11 billion weapons package for Taiwan in December 2025 — but that no shipment had taken place. Xi Jinping was likely seeking commitments on this issue in exchange for trade concessions.
The link between Taiwan and chips is obvious: TSMC, the manufacturer of the H200, is Taiwanese. Any conflict around Taiwan directly threatens the global supply of advanced semiconductors. By including Huang in his delegation, Trump sent a dual signal: to China, that AI is an area where cooperation remains possible; to Taiwan, that the United States remains the guardian of the supply chain.
This triangulation — Washington, Beijing, Taiwan — makes every chip a geopolitical object. The H200 is no longer a commercial product. It is an instrument of diplomacy.
Impact on Chinese AI models
Concretely, what is the effect of this blockage on Chinese AI? To understand this, we need to look at the scores of current models and their compute needs.
DeepSeek V4 Pro (Max) reaches 88 on the general benchmark, but its "High" version caps at 84. Moonshot AI's Kimi K2.6 scores 88.1 in agentic in self-host mode. Z.AI's GLM-5.1 displays 83. These models are competitive but remain behind the American leaders: Gemini 3.1 Pro (92), GPT-5.5 (91), Claude Opus 4.7 Adaptive (90).
Access to the H200 could significantly reduce this gap. More HBM3e memory means wider context windows, training on more massive datasets, and faster inference. Without the H200, Chinese companies must either throttle their models or invest massively in less efficient infrastructures.
The paradox is that China is blocking exactly what could help its companies compete. But Beijing is thinking long-term: dependence on American chips is a more serious strategic risk than a temporary performance deficit.
Nvidia's position: caught in the crossfire
For Nvidia, this situation is a logistical nightmare. The company has obtained the export licenses. It has the chips. It has the buyers. But it cannot deliver.
Jensen Huang has every interest in seeing shipments resume. The Chinese market represents a significant share of Nvidia's revenue — the company generated over $40 billion from AI in 2025, and China remains a major customer despite the restrictions. Every quarter that the H200 is blocked translates to hundreds of millions of dollars in delayed revenue.
But Nvidia also cannot cross Washington. The US inventory certification conditions are there to ensure that exports to China do not create a shortage on the domestic market. Nvidia must prove that sales to Alibaba or Tencent do not diminish its ability to serve US customers.
Huang's trip on Air Force One is the manifestation of this impossible position: the Nvidia CEO becomes a de facto diplomat, negotiating between two governments that are using his products as bargaining chips.
What this impasse reveals about the chip war
Three key lessons emerge from this situation.
Firstly, the chip war is no longer unilateral. In 2022-2023, Washington dictated the rules. In 2026, Beijing has developed its own blocking levers. China is no longer merely enduring restrictions — it is responding to them.
Secondly, tech sanctions have become instruments of political negotiation, not security measures. The authorization of the H200 comes at the time of a diplomatic summit. Beijing's blockage is a negotiating counter-measure. National security is the public argument; trade is the private reality.
Thirdly, the AI industry is hostage to this dynamic. Developers, researchers, startups — on both sides of the Pacific — are suffering the consequences of decisions made by a handful of political and corporate leaders. The pace of innovation slows down when hardware becomes a diplomatic tool.
❌ Common mistakes
Mistake 1: Confusing authorization and delivery
The most frequent mistake in media coverage is treating US authorization as an actual sale. Export licenses are a necessary condition, not a sufficient one. Beijing can block them, certification requirements can be impossible to meet, and timelines can stretch on for months. Saying "the H200s will arrive in China" is false. Saying "Washington authorized the sale, but Beijing is blocking it" is accurate.
Mistake 2: Thinking the H200 is Nvidia's most powerful chip
The H200 is powerful, but it is not the top of the line. The B200 (Blackwell) is superior in raw performance. The H200 remains however the most strategic chip in the context of exports, because it is the one located in the gray zone between what Washington is willing to authorize and what China is willing to accept.
Mistake 3: Ignoring the role of DeepSeek
DeepSeek is often absent from geopolitical analyses, because it is not a cloud giant like Alibaba or Tencent. Yet, VellaTimes reports that DeepSeek received conditional approval for the H200. This is significant: DeepSeek is the most visible Chinese model internationally, and its access to American hardware is an issue of prestige as much as performance.
❓ Frequently Asked Questions
Is the H200 superior to the H100?
Yes. The H200 offers 141 GB of HBM3e memory compared to 80 GB for the H100, with a bandwidth of 4.8 TB/s versus 3.35 TB/s. It is specifically designed for large-scale generative AI workloads.
Why is China blocking chips it requested?
Because U.S. conditions (proof of non-military use, stock certification) are perceived as interference. And because the precedent of the H20 — an intentionally throttled chip — created resentment. Beijing prefers to develop its own alternatives rather than accept chips under conditions.
Does Jensen Huang have an official diplomatic role?
No. Huang is a CEO, not a diplomat. But his presence on Air Force One gives him a de facto role as a negotiator. This is unprecedented for a semiconductor executive, and it shows just how much AI has become a state-to-state issue.
What alternatives does China have to the H200?
Huawei offers the Ascend 910C, Cambricon develops dedicated accelerators. These chips are generally less performant than the H200 in terms of memory bandwidth, but they are not subject to any import restrictions.
Does this agreement concern other Nvidia chips?
Available sources only mention the H200. The B200 remains banned from export to China. The H100 and the H20 are in distinct blocking situations. Each chip has its own export regime.
✅ Conclusion
The H200 saga sums up the state of the chip war in 2026: neither the United States nor China totally controls the flow, and AI is becoming a leading diplomatic issue. Washington authorizes, Beijing blocks, Huang boards Air Force One — and Chinese models wait. The next step will depend directly on what is negotiated around the table in Beijing.