The Wall Street Journal (WSJ) reported on November 15, 2026 that Megaspeed, a Singapore-based cloud operator, arranged orders for about $12 billion worth of NVIDIA AI chips for Chinese companies. The records the newspaper reviewed cover March 2024 to June 2025, and the paper says it could not determine how many chips were actually delivered or where they shipped from. Separately, Singapore authorities have disclosed that servers purchased by the companies under investigation were transported to Malaysia. The issue raises two questions: where the AI chips ended up, and who was using the computing resources through overseas data centers.
What do the roughly $12 billion in order records show?
According to the WSJ, the records it reviewed include orders for hundreds of thousands of NVIDIA AI chips. The roughly $12 billion figure represents the total value of those orders, not the value of products actually delivered. It remains unclear how many of the ordered chips were handed over, which models were included, or where the corresponding servers were ultimately delivered. Determining the real scale of the transactions would require matching the order records against delivery and shipping records.
Earlier reporting included figures suggesting different transaction sizes. In a November 13 follow-up, The Information referred to its own reporting that Century Litong had imported at least $7.5 billion worth of NVIDIA chips, or 240,000 units, into China since 2023. The WSJ, by contrast, focused on orders Megaspeed arranged between March 2024 and June 2025, so the companies and periods differ. How much the two sets of transactions overlap is not known.
According to Reuters' November 15 report, Megaspeed rebutted recent reports about its business activities and supply chain, saying they contain claims not supported by verifiable evidence. The company said it manages its supply chain to comply with applicable laws, including US export controls, and is conducting an internal investigation with outside legal counsel. It also said former executive Alice Huang left the company effective August 30, 2025, and no longer holds any position there.
NVIDIA also told the WSJ that it continues its efforts to prevent AI chip smuggling.
Singapore confirms transport to Malaysia and investigates suspected re-export
According to a statement released by Singapore's Ministry of Home Affairs on November 12, the country's police and customs authorities opened an investigation in February 2025 into several companies and individuals, including Megaspeed, over suspected offenses such as false declarations in server purchases. The US Department of Commerce's Bureau of Industry and Security (BIS) conducted a voluntary post-shipment verification of Megaspeed on June 24 of that year, and Singapore police also interviewed Huang the same day. In other words, the Singapore investigation began before the US verification.
The ministry said the servers purchased by the companies under investigation did not enter the Singapore domestic market but were transported through the country to Malaysia. Police are examining whether servers declared as destined for Malaysia were subsequently sent on to other countries or regions. Because clarifying the transport route also requires records on shipping origins and final delivery destinations, they have asked relevant countries, including the United States, for information. However, it is unclear which portion of the roughly $12 billion in orders reported by the WSJ corresponds to the servers identified in this investigation.
Malaysia's Ministry of Investment, Trade and Industry (MITI) also issued a statement on November 13. According to the ministry, Speedmatrix, Megaspeed's local affiliate, rents data centers in Malaysia and operates its own servers and AI chips there. MITI has visited the facilities, but said this was routine monitoring and not a finding that Megaspeed's or Speedmatrix's business activities are free of legal problems. It is working with Malaysian customs to investigate whether the AI chips brought into the country were re-exported to other countries and, if so, whether that complied with domestic law.
Under Malaysia's current system, no MITI permit is required to bring AI chips into the country. However, since July 2025, a permit under the Strategic Trade Act 2010 has been required to export or re-export covered high-performance US-made AI chips from Malaysia, or to transship or transit them through the country. Because the order records reported by the WSJ run through June 2025, judging whether individual transactions required a permit would require confirming the actual shipping dates, the products involved, and the form of transport.
As for Huang, Singapore police determined in February 2026 that there was insufficient evidence to establish the false-declaration offenses originally under investigation. However, they concluded that during the investigation she provided information she knew to be untrue in an attempt to downplay her own role, and issued her a stern warning. Huang has been barred from re-entering Singapore. The investigation into Megaspeed and six others is continuing.
Shipping chips to China versus contracting for remote use of overseas GPUs
In considering AI chip export controls, it is necessary to distinguish between the movement of the equipment itself and the remote use of GPUs installed overseas. Even if servers remain installed in a Malaysian data center, customers abroad, including in China, can have the GPUs run computing jobs over a network. It is a mechanism that lets Chinese companies use computing resources located elsewhere without the chips being physically shipped to China.
Citing industry sources, the WSJ reported that NVIDIA's cloud partners receive priority AI chip supply and that Megaspeed's large customers included China's Alibaba. However, it is not known which facilities Alibaba actually used, how much computing resource it used, or for what purpose.
Under US export controls, the location of the buyer and its ultimate parent company, not just the delivery destination, bears on whether a license is required. In guidance published May 31, 2026, BIS distinguishes between its policy of not enforcing the new regulatory requirements added by the 2025 "AI Diffusion Rule" and the export license requirements that existed before it.
Under Section 742.6(a)(6)(iii)(A) of the Export Administration Regulations (EAR), for covered advanced computing items classified as ".a" products, a license is in principle required even for transactions outside the United States if the purchasing entity, or its ultimate parent, is headquartered in Country Group D:5, which includes China, or in Macau. Unless a specified license exception applies, whether a license is needed depends on the counterparty's headquarters location and ownership structure, even if the receiving data center is in Southeast Asia.
In addition, in a policy statement dated May 13, 2025, BIS explained that exports, re-exports, and in-country transfers of advanced computing items to infrastructure-as-a-service (IaaS) providers may also require a license under certain conditions. Specifically, this applies where, under the "knowledge" standard defined in the EAR, the exporter is aware that the items will be used to train AI models for entities headquartered in Country Group D:5, such as China, or in Macau, and that those models will be used for military-intelligence end uses or users or for weapons of mass destruction.
For that reason, an investigation of AI chip transactions needs to confirm separately the location of the purchasing entity and its ultimate parent, the actual delivery destination, and the users and uses of cloud-based access. The physical shipping route alone cannot reveal every export-control issue.
According to the WSJ, the US Commerce Department is considering a plan to tighten export controls on advanced AI chips to three countries: Malaysia, Indonesia, and Singapore. Under the reported plan, exporting high-performance AI chips such as NVIDIA's Blackwell series to the three countries would require a license, with a condition prohibiting remote use by Chinese customers. No formal rule text or effective date has been confirmed, and the measure is described as under consideration.
How NVIDIA counts China revenue changes the figure by about $5.4 billion
To understand the reality of AI chip transactions routed through overseas data centers, it also helps to read NVIDIA's regional revenue figures carefully. In the company's disclosures, China and Hong Kong revenue differs significantly depending on whether it is based on the customer's billing location or the headquarters location of the direct customer.
| NVIDIA regional revenue basis | China and Hong Kong | Consolidated revenue |
|---|---|---|
| Customer billing location | 8,291 | 90,805 |
| Direct customer headquarters location | 13,644 | 90,805 |
In millions of US dollars. The period covered is the six months ended July 27, 2025.
In its quarterly report (10-Q) for the period ended July 2025, filed with the US Securities and Exchange Commission (SEC), NVIDIA broke down regional revenue by customer billing location. On that basis, China and Hong Kong revenue for the six months ended July 27, 2025 was $8.291 billion.
In its quarterly report (10-Q) for the period ended July 2026, by contrast, NVIDIA aggregates by the headquarters location of the direct customer and reports China and Hong Kong revenue of $13.644 billion for the same six months ended July 27, 2025. The difference is $5.353 billion, or about $5.4 billion. Under either basis, consolidated revenue is unchanged at $90.805 billion; what differs is the counting method, that is, which country or region the revenue is assigned to.
In the July 2025 report, Singapore accounted for 22% of billing-location-based revenue for the quarter and 21% for the six months. NVIDIA explained that customers centralize billing processing in Singapore while most products are shipped elsewhere. It also said that for both the quarter and the six months, more than 99% of revenue from export-controlled data center compute products billed to Singapore came from orders by US-based customers.
On the headquarters-location basis for the six months ended July 26, 2026, China and Hong Kong revenue was $12.43 billion, about 7.0% of consolidated revenue of $177.837 billion. NVIDIA notes that a direct counterparty may be a distributor or equipment maker, and that its headquarters location may differ from the location of the end customer and the product's shipping destination.
In short, regional revenue figures show the result of allocating sales to regions according to the basis NVIDIA adopted, not the physical delivery destination of the chips or the location of the end users. Likewise, these financial documents do not allow one to identify the quantity actually delivered under the roughly $12 billion in orders Megaspeed arranged, or the amount booked as NVIDIA revenue.
US Congress also demands preservation of export records, with equipment movement routes in focus
On November 11, US Senators Elizabeth Warren and Chris Coons sent letters to the Commerce Department, the State Department, and the White House asking them to preserve records on exports of advanced AI chips to China. The covered materials include documents and communications on export license decisions, investigations of diversion, and regulatory enforcement. The two senators asked for written confirmation by November 17 that the relevant offices have been instructed to preserve the records.
Earlier, on November 8, NVIDIA's board announced an independent investigation following reports that then-CEO Jensen Huang had been targeted by a foreign intelligence operation. The company said at the time that it had no reason to believe national security or confidential information had been compromised. On November 11, it then accepted Huang's resignation as president, CEO, and director, effective immediately.
According to a Form 8-K filed with the SEC on November 12, Chief Financial Officer Colette Kress will serve concurrently as interim CEO. The filing does not state the results of the independent investigation or the reason for Huang's resignation.
To tie the orders reported at about $12 billion to the servers Singapore authorities are tracing, it would be necessary to cross-check purchase orders, delivery notes, and shipping records and to confirm product model and serial numbers. Matching these against customs declarations and export license records would make it possible to trace which equipment was handed to which company, when, and by what route. If US license documents can be confirmed, the conditions applied to the counterparty and the transport route could also be understood.
If the servers remain installed in Malaysia, cloud service contracts and usage records would be needed to confirm which companies rented the computing resources, which facilities ran the workloads, and when they were used. Whether equipment physically moved and who used the GPUs remotely from abroad are separate questions.
If the re-export records Malaysian authorities are examining can be linked to the same equipment as the export license records US authorities have been asked to preserve, they could provide clues for verifying which of the reported orders were actually delivered and how far the equipment's subsequent movements were authorized.
