A Singapore-registered firm called Megaspeed spent nearly a year quietly moving almost $2 billion worth of Nvidia’s most advanced AI chips through a Malaysian subsidiary, according to trade records reviewed by The New York Times and reported alongside a wider investigation into how blacklisted Chinese firms keep accessing American semiconductor technology. The shipments, routed through a Malaysian unit called Speedmatrix Sdn Bhd, reportedly numbered more than 200 between June 2024 and June 2025, and much of the underlying hardware traced back to a subsidiary of Inspur, the Chinese server maker the U.S. Commerce Department placed on its Entity List years ago over its ties to Chinese military supercomputing programs.
The Megaspeed-Speedmatrix network is a separate strand of the same NYT investigation that also detailed a $5.6 billion export flow tied to another Inspur-linked entity. Where that story centered on direct trade-data totals, the Megaspeed thread is about geography and structure: a Southeast Asian data-center buildout that let restricted Nvidia hardware sit in Malaysia and Indonesia while, per the reporting, continuing to serve customers in China remotely. It is a distinct mechanism for the same underlying problem regulators have wrestled with since 2023 — an Entity List designation stops a company’s name from appearing on an export license, but it does not stop that company’s hardware, capital, or customer relationships from moving through friendlier jurisdictions.
Don't miss new tech stories on Google
Add Tech Insider once in the Google app and our stories appear in your news suggestions.
What the trade data actually shows
Reporters working from shipping and customs records — the kind of granular trade documentation aggregated by services like ImportGenius — traced a pattern in which Speedmatrix, the Malaysian arm of Megaspeed, placed large, repeated orders for Nvidia data-center GPUs that are barred from direct export to China under current Commerce Department rules. Over roughly twelve months, more than 200 individual shipments moved into Malaysia. The equipment reportedly did not stay in a vacuum: investigators found that a meaningful share of the hardware Speedmatrix received had passed through an Inspur subsidiary earlier in the supply chain, even though Inspur itself has been on the Entity List for years.
What makes this pattern hard to police is that none of the individual steps looks illegal in isolation. A U.S. supplier sells GPUs to a distributor. The distributor sells servers to a systems integrator in Malaysia. The integrator builds out a data center. Customers — anywhere in the world — rent compute from that data center over the internet. Nvidia does not control where cloud tenants are physically located once chips leave its direct distribution chain, and Malaysia is not itself subject to the same license requirements that apply to shipments bound straight for China. NYT reporter Ana Swanson summarized the broader pattern in a public post reviewed alongside the investigation, describing how a company Washington repeatedly blacklisted “reinvented itself” through a network of global subsidiaries and kept shipping advanced chips that ultimately fed Chinese AI development — a description that lines up with what the Megaspeed/Speedmatrix trade data appears to show for the Malaysia route specifically.
Why Malaysia and Indonesia became the transit points
Malaysia has spent the past three years marketing itself aggressively as a neutral data-center hub, courting hyperscalers with tax incentives, cheap land, and — critically — a regulatory posture that has not required the same chip-level export licensing the U.S. imposes on shipments headed to China. That combination made the country genuinely attractive to legitimate cloud operators. It also made it attractive to anyone looking for a jurisdiction where an order for several hundred million dollars of Nvidia silicon would not automatically trigger a Commerce Department license review.
The Commerce Department’s Bureau of Industry and Security has broadened its rules over the past two years specifically to close this kind of gap, adding location-based and end-user-based restrictions rather than relying solely on the Entity List. The official Entity List framework is built around named parties, which is precisely the weakness the Megaspeed case exposes: naming Inspur does nothing to stop a newly formed distributor, buyer, or landlord from filling the same role under a different name, in a different country, with paperwork that satisfies a customs officer even if it would not satisfy an export-control attorney.
Inspur’s role: sanctioned parent, unsanctioned subsidiaries
Inspur’s presence on the Entity List dates back to concerns that the company sought American technology to build supercomputing capacity for the Chinese military. That designation is supposed to require U.S. suppliers to obtain a license — typically denied — before selling controlled items to Inspur or any entity it owns 50% or more of. The complication, laid out across the NYT’s reporting, is that Inspur’s corporate structure spans dozens of subsidiaries and joint ventures worldwide, some with different names and different, less obviously Chinese, ownership registrations. A subsidiary that is not itself individually listed can, on paper, buy chips that its blacklisted parent could not buy directly.
The pattern is not unique to Inspur. It echoes a case tech-insider.org covered involving Sharetronic’s disclosure of $92 million in banned Nvidia server purchases, and the criminal case against a Super Micro co-founder who was arrested over an alleged scheme to smuggle Nvidia AI chips into China. What distinguishes the Megaspeed thread from those cases is scale and legality-on-paper: nobody has alleged Speedmatrix broke Malaysian law, and the shipments into Malaysia itself were not, on their face, prohibited exports. The exposure sits in what happened to the compute capacity after the servers were racked.
How this differs from the $5.6 billion Aivres story
Readers who followed the earlier reporting on Inspur’s use of its Aivres subsidiary to move $5.6 billion in Nvidia technology may see overlap here, and there is some — both threads trace back to Inspur’s global subsidiary network and both concern advanced U.S. chips reaching Chinese end users despite export controls. But the mechanisms differ in an important way. The Aivres figures describe direct export volume from the U.S. to Southeast Asia across nearly two years, including more than $3 billion in computers built around Nvidia’s Blackwell-generation components. The Megaspeed/Speedmatrix figures describe a narrower, more specific case: a single Malaysian buyer accumulating roughly $2 billion in restricted GPUs over about twelve months, then standing up data-center capacity that investigators say appeared to serve customers back in China remotely, with a separate entity called Maginfra also surfacing in trade records as a route for expensive servers moving back into China.
| Detail | Aivres / Inspur direct export thread | Megaspeed / Speedmatrix Malaysia thread |
|---|---|---|
| Reported dollar value | $5.6 billion (Apr. 2024–Feb. 2026) | ~$2 billion (Jun. 2024–Jun. 2025) |
| Primary mechanism | Direct U.S.-to-Southeast Asia export volume | Malaysian data-center buildout serving remote customers |
| Key named entity | Aivres (Inspur-linked subsidiary) | Megaspeed (Singapore) via Speedmatrix Sdn Bhd (Malaysia) |
| Hardware cited | Blackwell-class systems and Nvidia components | Restricted Nvidia data-center GPUs |
| Reported endpoint | Trade flow into Southeast Asia | Data centers allegedly serving Chinese customers remotely |
| Other linked entity | N/A in this thread | Maginfra, cited as a route for servers re-entering China |
The enforcement gap this exposes
The core problem is structural rather than a single bad actor. Export controls under the Export Administration Regulations are built to stop a named, listed party from receiving controlled technology. They are far weaker at stopping compute capacity itself from crossing borders once it is sitting inside a rented server rack accessible over the internet. A GPU physically located in Johor Bahru or Jakarta is, from a licensing standpoint, no longer subject to the same direct-to-China restriction — but if that GPU’s compute cycles are sold to a customer in Shenzhen through a cloud interface, the practical effect on Chinese AI capability is close to identical to a direct shipment.
This is the enforcement seam that has occupied Commerce Department rulemaking for much of the past two years. The department has repeatedly widened its restrictions beyond simple end-user lists toward broader country and technology-class controls, an approach documented across recent rule changes tracked at the Federal Register’s Industry and Security Bureau page. Even so, semiconductor industry groups like the Semiconductor Industry Association have flagged the tension between tightening controls and maintaining the commercial relationships that fund the next generation of chip R&D — a balance that gets harder to strike every time a new intermediary structure surfaces in the trade data.
Nvidia’s position and limited direct visibility
Nvidia does not sell directly to most of the entities named in this reporting; its chips reach distributors, systems integrators, and cloud operators through multiple layers of the supply chain before reaching an end customer. That distance is both a legal shield and a genuine visibility problem. A company can comply fully with the letter of its own export obligations — checking a buyer against the Entity List, requiring end-use certifications — and still have its hardware end up, several steps downstream, inside a data center that investigators later link back to a sanctioned parent company’s subsidiary network. Nvidia’s own compliance and legal disclosures describe screening processes built around exactly this kind of named-party check, which is the standard the entire chip industry currently operates under — and precisely the standard this investigation suggests is not sufficient on its own.
The broader chip-export enforcement landscape has already produced several parallel cases this year. Samsung and SK Hynix drew scrutiny after four of their Chinese fabs lost Validated End-User status, a separate but related sign that Washington is narrowing the list of arrangements it treats as automatically compliant — a shift tech-insider.org covered in detail around the Samsung and SK Hynix VEU changes. Meanwhile Nvidia’s own data-center hardware pipeline keeps accelerating on the legitimate side of the ledger, with GB300 Blackwell Ultra systems now shipping at volume and pushing quarterly data-center revenue past $89 billion, and hyperscalers like AWS continuing to place enormous orders, including a reported 2 million additional Nvidia GPU commitment. The scale of legitimate demand is exactly what makes a $2 billion diversion easy to lose inside the noise of a market moving hundreds of billions of dollars a quarter.
Historical context: a pattern that predates this case
Chip-export evasion through Southeast Asian intermediaries is not a new phenomenon; it has simply gotten more sophisticated and higher-value over time. Earlier cases involved smaller volumes and cruder methods — falsified end-user certificates, direct reshipment through Hong Kong, or individuals physically carrying components across borders. The Megaspeed/Speedmatrix pattern represents a more mature version: a legally registered Singapore holding company, a Malaysian operating subsidiary with its own corporate filings, and a data-center business model that has a plausible, legitimate commercial explanation (cloud capacity for Southeast Asian customers) even as the trade data suggests a different practical use.
Regulators have responded to earlier iterations of this problem by expanding controls incrementally — first targeting specific chip performance thresholds, then specific countries, then specific ownership structures. Each expansion has, in turn, pushed the next generation of intermediary structures to become more corporate and more geographically dispersed. The Entity List itself, maintained by BIS and documented in the consolidated entity list, has grown to include thousands of parties precisely because each new addition tends to generate a handful of successor entities designed to operate just outside its reach.
Market and industry impact
For Nvidia and its distribution partners, reporting like this creates reputational and regulatory risk even when the company itself is not accused of wrongdoing. Every high-profile diversion story adds pressure on Congress and the Commerce Department to tighten chip-tracking requirements — proposals that have circulated in various forms, including hardware-level location verification for high-end GPUs, would directly affect how Nvidia and its distributors structure sales contracts. For Malaysian and broader Southeast Asian data-center operators, the story is double-edged: it confirms the region’s importance as AI infrastructure real estate, which is genuinely growing for legitimate reasons tied to power availability and land cost, while also inviting closer scrutiny of every large GPU order routed through the region, legitimate or not.
For Chinese AI developers, the practical effect of these networks — if the trade data holds up to further scrutiny — is continued access to top-tier training and inference hardware despite years of tightening controls, which helps explain why domestic Chinese AI labs have kept pace with frontier model releases even as direct chip sales to China have been restricted. It also complicates the U.S. government’s underlying policy goal: controls aimed at slowing Chinese AI progress lose most of their effect if compute capacity built on the same restricted hardware remains reachable through a rented cloud interface a few hundred miles away.
Competitive comparison: how enforcement approaches differ by chipmaker exposure
Nvidia is not the only chipmaker whose products have surfaced in intermediary-network reporting, but its market position makes it the most frequent subject simply because its data-center GPUs remain the most sought-after training hardware. AMD’s Instinct-series accelerators and Broadcom’s custom AI silicon have drawn far less attention in diversion cases to date, partly because Nvidia still commands the dominant share of high-end AI training deployments and partly because Nvidia’s CUDA software ecosystem makes its chips specifically more valuable to divert than functionally similar alternatives.
| Company | Role in export-control landscape | Recent enforcement exposure |
|---|---|---|
| Nvidia | Dominant AI GPU supplier; primary target of diversion schemes | Repeated subject of trade-data investigations, incl. this case |
| Inspur | Chinese server maker, Entity List since concerns over military ties | Subsidiary network linked to both this case and the separate Aivres $5.6B thread |
| Samsung / SK Hynix | Memory and chip toolmakers with China fab exposure | Four China fabs lost Validated End-User status in 2026 |
| Super Micro (individual case) | Server integrator | Co-founder arrested over alleged Nvidia chip smuggling scheme |
| Sharetronic | Hardware reseller | Disclosed $92M in banned Nvidia server purchases |
What comes next for regulators
Congressional pressure on chip-export enforcement has been building for over a year, and cases like this one tend to accelerate proposals that were already circulating in draft form. Expect renewed pushes for mandatory GPU location-tracking, tighter licensing requirements for large data-center orders in transit hubs like Malaysia and Indonesia, and closer coordination between BIS and counterpart agencies in Southeast Asian governments that have so far had little incentive to police U.S. export rules on their own soil. None of these fixes are simple: location-tracking hardware raises its own privacy and sovereignty objections from buyers, and transit-hub governments benefit economically from exactly the kind of data-center investment this case scrutinizes.
Predictions: where this story goes from here
- Expect the Commerce Department to open a formal review of Megaspeed and Speedmatrix’s export history within weeks of the NYT report, mirroring how past trade-data exposés have triggered follow-up BIS inquiries.
- Look for additional Inspur-linked entities to surface in follow-up reporting, given how many subsidiaries the company’s global structure reportedly includes.
- Malaysian and Indonesian officials will likely face renewed diplomatic pressure from Washington to tighten oversight of large GPU imports transiting their data-center hubs, even without new legislation.
- Nvidia will probably face renewed questions from investors and lawmakers about distributor-level compliance screening, even though the company itself has not been accused of directly violating export rules in this case.
- Expect at least one more major intermediary-network story before the end of 2026, given how frequently new trade-data investigations have surfaced this year across Inspur, Sharetronic, and Super Micro-linked cases.
Why the Entity List keeps failing to close this loophole
The Entity List’s core design assumption is that a sanctioned company has a fixed identity a supplier can screen against. Modern Chinese tech conglomerates like Inspur do not operate that way; they function as holding structures with dozens of legally distinct subsidiaries, joint ventures, and minority-owned affiliates spread across multiple countries. Screening software checks a buyer’s registered name against the list. It generally does not — and often cannot, given available public records — trace beneficial ownership through three or four layers of holding companies registered in Singapore, Malaysia, or the British Virgin Islands. Trade groups tracked by agencies like the International Trade Administration have acknowledged this gap in public commentary on export-control modernization, and the U.S. Trade Representative’s office, whose broader trade-policy role is outlined at ustr.gov, has periodically weighed in on how export controls intersect with broader trade relationships with Southeast Asian partners.
Closing that gap would require either dramatically more aggressive beneficial-ownership disclosure rules — politically difficult given how much legitimate foreign investment relies on holding-company structures — or a shift toward controlling the technology itself, through hardware-level attestation or geofencing, rather than controlling the buyer’s identity. Both paths face resistance from different constituencies, which is a large part of why this exact pattern, evasion through a web of subsidiaries in a friendly third country, keeps recurring across multiple Chinese conglomerates rather than being a one-time exploit that gets patched and closed.
Frequently asked questions
What did The New York Times report about Inspur and Nvidia chips?
The New York Times reported that Inspur, a Chinese server maker on the U.S. Commerce Department’s Entity List, continued accessing advanced Nvidia AI chips through a network of global subsidiaries and partner companies, despite formal export restrictions on the company.
Who are Megaspeed and Speedmatrix Sdn Bhd?
Megaspeed is a Singapore-based company whose Malaysian subsidiary, Speedmatrix Sdn Bhd, reportedly purchased close to $2 billion in Nvidia chips restricted from direct export to China, receiving more than 200 shipments between June 2024 and June 2025 according to trade records cited in the reporting.
Is this the same story as the $5.6 billion Aivres/Inspur report?
No. Both stories are part of the same broader NYT investigation into Inspur’s subsidiary network, but they describe different mechanisms and figures. The Aivres thread covers roughly $5.6 billion in direct export volume to Southeast Asia. The Megaspeed/Speedmatrix thread covers a separate $2 billion Malaysian data-center buildout allegedly serving Chinese customers remotely.
Why is Inspur on the Entity List?
U.S. officials placed Inspur on the Entity List over concerns the company sought American technology to help build supercomputing capacity for the Chinese military, which restricts U.S. companies from selling it controlled items without a license that is typically denied.
Did Nvidia break export control rules in this case?
The reporting does not allege Nvidia directly violated export rules. The chips reportedly moved through multiple layers of distributors and integrators before reaching Speedmatrix, illustrating how downstream diversion can occur even when a chipmaker’s own direct sales comply with Entity List screening.
What is Maginfra’s role in this story?
Maginfra is a newer Chinese company that surfaced in trade records reviewed for the investigation as a route through which high-value servers, some tied to the Inspur/Malaysia network, were imported back into China.
Why did Malaysia become a hub for this kind of activity?
Malaysia has aggressively courted data-center investment with tax incentives and cheap land, and its import rules do not carry the same direct-to-China chip-level licensing requirements the U.S. applies to shipments headed straight for China, making it an attractive transit and hosting point for both legitimate and questionable large-scale GPU deployments.
What could the Commerce Department do next?
Likely next steps include a formal review of the entities named in the reporting, potential new restrictions targeting Southeast Asian transit hubs specifically, and renewed pressure for beneficial-ownership disclosure requirements that would make it harder for subsidiaries of blacklisted firms to operate under different names.
Related Coverage
- Inspur Used Aivres to Buy $5.6B in Nvidia Chips [2026]
- Cramer: Nvidia's $13B Deal Locks Out AMD, Broadcom [2026]
- Nvidia Brings DLSS 5 to RTX 40, Skips Full Control [2026]
- HPE Stock Surges 7% as Nvidia Vera CPU Server Debuts [2026]
- Qualcomm Confirms Dual Snapdragon 8 Elite Gen 6 Chips, 5GHz [2026]
