How NVIDIA Helped Alibaba Circumvent US Chip Sanctions
New reporting alleges NVIDIA's sales team actively brokered relationships between Alibaba and Alice Huang — a figure at the center of a multi-billion-dollar AI chip smuggling ring — and even recommended a Singapore-based loophole after US export controls took effect.
The Introduction That Changes Everything
Until now, the most damaging revelations about NVIDIA’s ties to Chinese sanction-busting came from investigations into how chips physically moved — shell companies, shipping manifests, customs records. What a new DeInformaticom report makes concrete is something narrower and arguably more consequential: NVIDIA executives didn’t just sell into a market that later proved porous. They actively brokered the very relationships that would feed that porosity.
In 2022, before any AI-specific export restrictions took hold, NVIDIA’s sales team connected a prospective customer — Alibaba — with Alice Huang, who was then structuring a semiconductor leasing business. Huang used that Alibaba relationship to secure bank financing, which she then deployed to buy NVIDIA chips. The deal looked legitimate on both sides: a chip maker securing a major account, a reseller building a client base. Then came August 2022, when the United States began restricting A100 exports to China. The same NVIDIA sales team that had made the introduction offered Huang a new play: stop trying to import chips directly. Lease them instead, through a Singapore entity called MegaSPEED, which was linked to Huang’s network.
That was the pivot point. The US restriction targeted direct shipments from NVIDIA to China. It did not, at the time, prohibit a Chinese company from accessing those same chips via a foreign subsidiary or leasing arrangement. The sales team’s recommendation wasn’t a clumsy misstep — it was a precise exploitation of a regulatory blind spot that would become a $4 billion loophole.
How the Workaround Actually Worked
The architecture was elegant because it required no fabrication, no forged documents, no bribery. It worked entirely within the letter of existing rules. Alibaba needed GPUs for its data centers. China had restrictions on direct import of advanced chips. But nothing barred a Singapore-registered company from owning and operating those chips and offering computing power to anyone willing to pay — including a Chinese firm located inside China.
Huang’s vehicles — MegaSPEED in Singapore, and later Shanghai Century Litong, a company closely tied to her family — became the conduit. By 2023, Litong had obtained NVIDIA’s “priority cloud partner” designation, meaning it received preferential allocation of restricted hardware. From 2023 through 2025, DeInformaticom reports, billions worth of sanctioned semiconductors flowed into China through this structure. Some ended up in Alibaba’s and Tencent’s data centers. Others were absorbed into the broader ecosystem of Chinese AI training workloads.
The mechanics matter because they reveal the scale of the gap. This wasn’t a handful of rogue employees quietly shipping boxes through Hong Kong. This was a structured, volume-based distribution channel that benefited from NVIDIA’s own prioritization system — the same system designed to reward strategic partners with early access to new products.
Who Benefits When Rules Have Holes
Jensen Huang built NVIDIA’s dominance on two claims: technical supremacy and compliance rigor. His public posture has consistently framed NVIDIA as a responsible actor navigating an impossible geopolitical landscape — selling what it can, refusing what it can’t, losing enormous revenue rather than violate American law. The report reframes that narrative uncomfortably. If NVIDIA’s sales team was not merely turning a blind eye but actively designing workaround pathways, then the company’s compliance infrastructure wasn’t weak — it was selectively applied.
There is a difference, legally and morally, between failing to police your customers and helping them build the infrastructure that lets them bypass your regulators. The former is negligence. The latter is complicity by architecture.
Who wins here? The immediate winners are the same ones that have always won from fragmented enforcement: Chinese cloud providers, AI labs, and research groups that otherwise would have been locked out. They gain compute access at scale without triggering the diplomatic incidents that direct procurement would create. Huawei, by contrast — the company the sanctions were partly designed to strangle — still faces hard barriers. The beneficiaries aren’t the intended targets of US policy; they’re the companies sitting just outside the prohibition net.
Why the August 2022 Timing Is Crucial
The sequence is important. The introduction of Alibaba and Huang predated the A100 restriction. The switch to the leasing model postdated it. That means NVIDIA’s sales team understood the regulatory shift and responded by redirecting their own customer onto a compliant pathway — one that achieved the same commercial outcome through a legal backdoor rather than violating the rule outright.
This isn’t a story about bad actors moving in secret. It’s a story about how well-resourced companies can navigate regulatory change while preserving revenue streams their governments would prefer they abandon. The compliance team drew a line. The sales team found a route around it. Both were acting rationally within their own incentive structures. The result was the erosion of the policy they ostensibly shared.
DeInformaticom’s reporting also notes that the Alibaba-Huang introduction was not unique — NVIDIA had similar connector relationships beyond MegaSPEED. The article frames this as evidence of a pattern under Jensen Huang’s leadership rather than an isolated incident. That framing is significant because it shifts the question from whether one employee crossed a line to whether the organizational culture incentivized crossing it.
The Enforcement Gap That Won’t Close Itself
The fundamental problem exposed here is structural, not accidental. US export controls on semiconductors have always drawn a boundary around direct transfers — a shipment from Santa Clara to Shenzhen triggers scrutiny. They do not, and cannot easily, govern the behavior of chips once they leave US jurisdiction. A GPU installed in a Singapore data center belongs to a Singapore entity. How that entity allocates compute to downstream users falls outside the reach of US Commerce Department licensing regimes.
That gap existed before this report. It exists now. But the report shows how quickly a motivated seller can turn a theoretical gap into a commercial strategy. The sales team didn’t discover the loophole by accident. They recognized that their existing customer relationships — already embedded in Chinese tech — could be rerouted through a structure that happened to sit outside the restriction zone.
Policymakers have talked about closing this gap for years. The challenge is real: restricting compute access through offshore intermediaries requires either extraterritorial enforcement authority that the US currently lacks, or a complete redesign of how semiconductor licensing works — perhaps tying allocations to end-user verification rather than transaction origin. Neither is politically simple.
What Comes Next
The most urgent question isn’t whether NVIDIA violated the law — that determination rests with prosecutors and the Commerce Department. The more pressing question is what this means for how America enforces its own technology policy going forward.
If the report’s characterization holds — that senior leadership knowingly facilitated these arrangements — then NVIDIA’s long-standing public stance as a compliance-first company deserves serious reconsideration. Not as punishment, but as calibration. Investors, partners, and competitors need accurate information about whether the company’s risk management reflects its public commitments or its revenue incentives.
For Alibaba and Tencent, the implications are operational as much as reputational. Their supply chains for advanced compute were partially built on a structure that may now be considered tainted. Future sourcing decisions will carry added scrutiny regardless of whether legal exposure is ultimately established.
And for Washington, the takeaway is blunt: you cannot design sanctions that bite only on paper. When the people you’re regulating have the expertise and the relationships to find the seam between what’s prohibited and what’s merely discouraged, the prohibition becomes advisory. That doesn’t mean sanctions are useless. It means they need to be sharper — covering the circuits, not just the shipments.
The NVIDIA-Alibaba-Huang thread is a single line in a much larger tapestry of US-China technological decoupling. But it’s a useful one, because it shows exactly how that decoupling frays at the edges: not through dramatic violations, but through quiet commercial ingenuity exercised by companies that profit from both sides of the divide.