On May 14, 2026, a piece of news about Nvidia's H200 tightened the already tense battlefield of computing power: the United States had previously approved about 10 leading Chinese tech companies, including Alibaba, Tencent, ByteDance, and JD.com, to purchase H200s, with a maximum of 75,000 units for each company. This high-performance chip, regarded as an upgraded version of the H100, was originally expected to be a key supply to relieve the computing power famine. However, as of May 14, none of these "approved" chips had been shipped to China; the permits lay on paper while the goods remained on the road. According to a single source, the U.S. added conditions to the export license—the chips must transit through the U.S., and a 25% fee would be charged at this stage; this design, seen as a form of price increase and control over the flow, was rejected by Chinese companies, leading to a halt in related purchases, thus creating a deadlock of "approved but not delivered." Intriguingly, on the same night of this trans-Pacific tug-of-war, Nvidia’s stock rose over 2% in after-hours trading, hinting the capital market provided its own interpretation of the situation; at that moment, the fate of the H200 officially became the latest focus in the supply of AI computing power and geopolitical gamesmanship.

Approved but not delivered: Additional terms put exports on the negotiation table

Since the U.S. imposed advanced AI chip export restrictions on China in 2022, Nvidia has been seeking survival in an increasingly tightening gap: first, the H100 was directly blocked, then the A800 and H800, tailored for the Chinese market, were briefly listed before also being subjected to stricter controls. This cycle of "first offering a compromise, then tightening the gate a bit more" has made major Chinese companies gradually accustomed to scrambling for time in a gray area, rather than hoping for genuine policy reversals. In this context, the sudden granting of export permits for the H200—an upgrade of the H100—seemed glaringly conspicuous: about ten Chinese enterprises were named for approval, each with a limit of 75,000 units, including Alibaba, Tencent, ByteDance, JD.com, and Lenovo and Foxconn, which serve as server supply and distribution hubs. This felt like a rare smile, yet it occurred at a point dominated by prohibitions, making it hard not to interpret it as a strategically conditional opening.

According to a single source, these conditions were embedded in the additional terms of the export license: the H200 must "transit" through the U.S. before reaching Chinese customers, and a 25% fee would be deducted at this stage. For the U.S., this was both a re-control over the flow of the supply chain and an additional "tax" on high-end computing power demand; for Chinese companies, this was seen as a bundled package of politics and economics, not only raising costs but also reinforcing the perception of control over the flow of critical chips. The result was that the Chinese side chose not to accept this condition—purchases were halted, orders frozen, and as of May 14, 2026, Nvidia had still not shipped a single H200 to any approved Chinese customer. Formally, "the license has been approved," but essentially this pushed the export itself onto the negotiation table, where the approval document no longer meant delivery but became a bargaining chip that could be magnified or retracted at any moment.

Alibaba Cloud revenue skyrockets: The internal circulation of Chinese AI computing power pressed the accelerator

As the H200 got stuck in the gray area of "approved but not delivered," the billing panels of domestic cloud providers saw rapid increases. According to a single source, from early April 2024 to May 13, 2026, the average daily revenue of Alibaba Cloud's related Token business surged more than fivefold in just over a year, with monthly revenue climbing to several hundred million RMB. The numbers themselves are not complicated: each API call and each segment of large model output combine to form a steeply rising curve, indicating that what is truly suppressed is not the demand, but whether sufficient computing power can be promptly found domestically to meet this demand.

With the supply prospects of the H200 being pulled back and forth by policies, the tech giants named for approval—Alibaba, Tencent, ByteDance, JD.com—found it difficult to pin their AI futures on a permit that might change at any time. Research briefs point out that the process of self-sufficiency in Chinese AI computing power is accelerating, with domestic alternatives including Huawei's Ascend and other local chip platforms; for these giants, "waiting for the ship to come while building boats" has become the new norm: one hand continues to seek Nvidia GPUs while the other simultaneously steps up investments in self-built data centers and local chip ecosystems. The constraints on exports have pressed the originally sluggish "internal circulation of AI computing power" into high gear; local chips and cloud service providers are under multiple pressures of performance, costs, and ecological maturity in the short term, yet this has provided a rare time window to seize discourse power and market share. This passively initiated internal circulation will reshape the winners and losers in China's AI computing landscape in the coming years.

Lenovo and Foxconn: The supply chain caught between red lines and orders

As Chinese cloud providers bet on local chips with one hand while still eyeing Nvidia's latest H200 with the other, Lenovo and Foxconn find themselves in the middle of the chain: they are both listed as approved distributors of the H200 and are key suppliers of AI servers, meant to undertake the "last mile" of bringing American chips into Chinese data centers. Upstream, they must digest the terms set by Nvidia and U.S. regulatory bodies; downstream, they face procurement intentions from approved customers like Alibaba, Tencent, ByteDance, and JD.com for up to 75,000 GPUs each, yet cannot provide a clear delivery schedule in the current reality of "approved but not delivered." According to a single source, the U.S. added a term to the permit stating that "the H200 must transit through the U.S. and 25% fees will be charged"; after the Chinese buyers refused, purchases were halted, and the pace of orders instantly shifted from a fully loaded plan to a plan that could be paused at any moment, forcing Lenovo and Foxconn to continuously rewrite their production scripts between production scheduling, materials preparation, and quotations.

This uncertainty has already reflected in financial reports and market expectations. In a certain year’s first quarter, Foxconn reported revenue of 2.12 trillion NTD, slightly below the market expectation of 2.15 trillion NTD, but posted a net profit of 49.92 billion NTD, exceeding the expected 48.43 billion NTD. Public opinion quickly shifted attention to the AI server business regarded as the "growth engine," but how much profit this business actually contributed remains to be further verified. Export rules tighten and loosen alternatively, and the H200 remains unshipped; these contract manufacturing giants, originally skilled in “scalability and standardization," are forced to learn to split production capacity under multiple scenarios: part is contingent on whether compliant export routes can be reopened, while another part is reserved for new orders that local computing solutions might bring. Between the red lines of U.S. regulation and the surge in Chinese demand, each gear shift in capacity by them is a bet on the direction of the next computing power contest.

Clouds of war over Iran and crypto pull: Global policy noise superimposed with technological unease

In the first quarter, the GDP growth rate of the UK returned to 0.6%, marking the fastest figure in a year, seemingly stepping out of the shadow of recession. However, the research brief provided data while pointing out the source of the shadow: the aftermath of the Iranian war drags energy prices and geopolitical risks that refuse to dissipate, while at home lies a political game lurking with a changing prime minister. For capital that values the match of timing, this means that behind the seemingly repaired economic curve in Europe is a line that could retract at any moment due to changing Middle Eastern situations or shifts in Downing Street.

At the same time, signals from the Middle East are inherently contradictory. Iranian Foreign Minister Araghchi publicly emphasized that "there is no military solution to the issue of Iran," and that the Iranian people will not yield to threats or pressure, declaring their "preparedness to fight for freedom and territory" while simultaneously stressing their willingness to resolve issues through diplomatic means. This posture of "preparation for war while seeking peace" locks regional tensions at a high level but leaves paths for both conflict and de-escalation in suspense. Across the Atlantic, according to Politico, U.S. Senator Tim Scott plans to veto a dozen amendments related to crypto on the grounds of "miswording," again demonstrating the reality that rules can be rewritten at any time by technical details and political calculations in the realm of crypto and broader tech regulation. The research brief views this series of fluctuations as a microcosm of the global policy environment’s complexity: when the clouds of war over Iran, European political situations, and U.S. legislative pulls intertwine, AI and crypto, as technological assets, are being forced to reprice policy risks amid geopolitical shocks and regulatory back and forth.

New Cold War in AI computing power: Who is losing the window of opportunity

In the deadlock of "approved but not delivered" for the H200, the export permit has been crafted into a repeatedly adjustable bargaining tool: the U.S. approved about 10 Chinese enterprises to purchase the H200, with each company capped at 75,000 units, yet as of May 14, 2026, shipments have not been made. According to a single source, conditions were also added, requiring transit through the U.S. and charging a 25% fee, with the time cost being borne by both sides of enterprises and developers. High-performance chips remain tightly controlled, and Chinese tech companies are forced to accelerate their bets on local computing power, pushing local platforms like Huawei's Ascend and self-built data centers into the spotlight; according to a single source, from early April 2024 to May 13, 2026, Alibaba Cloud’s average daily Token revenue increased over fivefold, with monthly revenue reaching several hundred million RMB, indicating that the local AI ecosystem growing in the gap is actively filling the blocked computing power. For U.S. manufacturers like Nvidia, the research brief suggests this round of restrictions may ultimately weaken their dependence on the Chinese market and rearrange revenue structures, but this causal relationship still needs time for verification. The capital market has already placed its bet—in the night the H200 news became public, Nvidia's stock price rose over 2%. As the UK economy sways between a 0.6% quarterly growth rate and political risks, Iranian Foreign Minister denies a "military solution" while emphasizing both battle and diplomacy, and U.S. senators repeatedly grapple with crypto regulation clauses, the AI computing power market is being pushed toward a more fragmented and regionalized track by these noises. The real competition lies in which countries and companies can maintain the resilience and continuity of the computing power network amidst institutional uncertainty and supply chain shocks.

Join our community, let’s discuss together and become stronger!
Official Telegram community: https://t.me/aicoincn
AiCoin Chinese Twitter: https://x.com/AiCoinzh
OKX benefits group: https://aicoin.com/link/chat?cid=l61eM4owQ
Binance benefits group: https://aicoin.com/link/chat?cid=ynr7d1P6Z

免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。