Nearly 900,000 approved AI chips sounds like a massive windfall for Chinese tech giants. It's not that simple.
Data released by the U.S. House Foreign Affairs Committee shows the Commerce Department cleared roughly 879,000 advanced AI processors from Nvidia and AMD for Chinese companies between December 2025 and May 2026. Named recipients in the disclosure include Tencent with 150,000 chips, Alibaba with 75,000, Baidu with 43,000, and BYD with 4,000. On paper, that looks like a major opening for Alibaba, Tencent, and their peers after years of tight export walls.
Look closer, though, and the story flips. Chinese firms asked for more than 3.4 million chips during that six-month window, meaning Washington signed off on well under 30% of what was requested. Even better, an export license gives permission to ship, not a guaranteed delivery on a loading dock. And right now, Nvidia says it has stopped shipping H200 processors to China anyway.
If you've watched semiconductor export battles closely over the past few years, you know headlines about "approvals" almost always lag behind reality on the ground. Policy shifts, supply bottlenecks, and political pushback collide fast. That gap between paper approvals and actual server racks is where the real story lives.
Why 900,000 Approved AI Chips Don't Equal Instant Compute Power
When people read that Washington cleared 879,000 AI chips for Chinese firms, they often picture hundreds of thousands of high-end accelerators plugging straight into data centers in Hangzhou, Shenzhen, or Beijing. Honestly, that's a misunderstanding of how export licensing works.
A license from the Bureau of Industry and Security is a green light under specific conditions, not a completed shipment. Reporting around the disclosure notes that it remains unclear how many approved licenses turned into actual deliveries. Meanwhile, Nvidia stated that approvals carried strict conditions and limited volumes, and that the company has halted H200 shipments to China. Reports also point to roughly 600,000 H200 units sitting stranded in Hong Kong rather than flowing freely into Chinese clusters.
There's also a sharp distinction between asking for hardware and getting it. Alibaba reportedly sought around 950,000 chips and got approvals for 75,000—a fraction of its wish list. Across the entire pool of more than 3.4 million requested units, approvals stopped short of 880,000.
From years of tracking hardware supply chains, I've seen teams treat a permit or a purchase order like deployed capacity. In AI infrastructure, nothing counts until power, cooling, networking, and actual silicon are racked and running. A license without physical delivery is just paperwork.
The Military-List Dilemma Behind Alibaba, Tencent, and Baidu
Here's the part that makes Washington hawks furious. All four companies highlighted in the committee's breakdown—Tencent, Alibaba, Baidu, and BYD—appear on the Pentagon's list of entities with alleged ties to the Chinese military.
Former U.S. official and tech-competition analyst Chris McGuire pointed out that more than 30% of the cleared AI chips went to firms identified with Chinese military links, raising sharp questions about how national security reviews were handled. That critique hits a nerve because advanced accelerators like Nvidia's H200 and AMD's MI325X serve dual purposes: training commercial chatbots and cloud models on one hand, and supporting defense or surveillance workloads on the other.
Context matters here, though, and the timeline isn't black and white. Reporting indicates Tencent was placed on the Pentagon's Section 1260H list in January 2025, whereas Alibaba, Baidu, and BYD were briefly added in February before that list was withdrawn, and then included again in a June update. That means some approvals happened during a window before those later listing updates took effect.
Even with that nuance, the optics are brutal for policymakers. The Commerce Department noted that its H200 licensing approach followed a January 2026 Federal Register rule requiring extra national security conditions for case-by-case review rather than automatic denial. Critics still argue that clearing hundreds of thousands of chips for major Chinese conglomerates undermines the spirit of export controls.
Policy Whiplash and the Offshore Loophole Crackdown
If you want whiplash, look at U.S. chip policy toward China over the past two years. Restrictions tightened under the Biden administration, shifted when President Donald Trump moved to allow H200 exports to China in December, and now face renewed pressure from lawmakers and fresh Commerce Department actions.
Politically, the heat is turned way up. Lawmakers have scheduled hearings and briefings involving Commerce officials and former Nvidia CEO Jensen Huang amid broader scrutiny over enforcement, smuggling concerns, and past policy reversals. Chipmakers like Nvidia and AMD have long argued that heavy restrictions push customers toward domestic rivals and hurt U.S. competitiveness, while critics counter that loosening access hands Beijing valuable compute.
At the same time, Washington is going after workarounds. Chinese tech firms haven't relied solely on direct imports; they've also looked at overseas data center arrangements in places like Malaysia, Singapore, and Thailand to access advanced processors remotely. Reports in November 2026 highlighted Commerce Department moves to tighten rules on computing capacity routed through foreign data centers, a shift Tencent acknowledged would slow development of its next-generation models.
Let's be real: export controls are basically a game of whack-a-mole. Close one door—direct shipments, specific chip tiers, or offshore cloud rentals—and buyers look for another route, while regulators scramble to draft tighter rules.
Domestic Chinese Chips Still Face a Real Gap
A common misconception is that U.S. export limits have already handed Chinese domestic chipmakers an easy victory. In practice, local alternatives like Huawei's Ascend series and Cambricon's Siyuan chips still trail top Nvidia and AMD hardware in performance and ecosystem maturity, as highlighted in congressional commission reporting.
That doesn't mean Chinese firms are standing still. Domestic demand, state backing, and restricted foreign supply push engineers to optimize software, squeeze more efficiency out of available clusters, and test local silicon wherever it fits. Still, bridging a multi-year gap in high-end chip design, high-bandwidth memory, and manufacturing tooling takes time.
For companies like Alibaba and Tencent, the squeeze comes from both sides. Tighter U.S. rules restrict access to top American accelerators and offshore compute, while domestic chips aren't a plug-and-play replacement for every frontier workload.
What Smart Operators Should Do Next
If you're an investor, founder, enterprise buyer, or policy watcher trying to make sense of these numbers, don't trade on surface-level headlines. Take concrete steps to protect your assumptions:
- Separate licenses from actual deployments. Whenever you see big export-approval numbers, check delivery status, shipment halts, and inventory bottlenecks before assuming a company's compute cluster just grew.
- Track regulatory rules alongside hardware roadmaps. Watch Federal Register updates, Commerce Department enforcement, and offshore data-center restrictions—they can change model training timelines faster than a software release.
- Stress-test your AI stack against hardware risk. If you build products or run workloads tied to specific cloud providers or regions, keep model architectures flexible and avoid single-vendor lock-in where possible.
- Read primary disclosures critically. Look past sensational summaries to see who requested what, which specific chips were covered, and whether figures reflect approved caps or shipped hardware.