Malaysia Turns to Huawei For AI Chips, Signaling Shift Away From Western Suppliers

Bloomberg's September 8 report that Malaysia is actively evaluating Huawei's Ascend 910C processor for a 2 billion ringgit sovereign AI programme marks the first time any foreign government has moved this close, publicly, to choosing Chinese AI accelerators over American ones. The decision, if taken, would be the most consequential test of US extraterritorial chip policy since Washington wrote the rules.

The programme carries a price tag of approximately $494 million at current exchange rates, according to Bloomberg's reporting by Faris Mokhtar, Mackenzie Hawkins, and Niki Koswanage . The contract vehicle is Telekom Malaysia, a state-linked operator that won the role through a public tender and already runs Huawei technology inside its cloud infrastructure. Its current AI computing service uses Nvidia chips. The number of Ascend 910C units under consideration has not been disclosed. The deal is unsigned.

Prime Minister Anwar Ibrahim has named the reason plainly. The US Cloud Act allows American courts to compel US technology companies to surrender data stored overseas. "I do not consider that law reasonable, but how does one argue with President Trump?" Anwar said in April, speaking to university students, as reported by Bloomberg . That single sentence is the most honest summary of the sovereign AI dilemma facing emerging-market governments in 2026: the legal architecture of American cloud infrastructure is itself a security argument for Chinese hardware.

The nut paragraph is this: Malaysia hosts one of Southeast Asia's fastest-growing AI data centre ecosystems, built largely by US firms including Microsoft and Oracle. Southeast Asia as a whole is planning four times its current data centre capacity, according to Bloomberg . A $494 million Huawei win inside that ecosystem does not just threaten one contract. It creates the precedent every other government in the Global South is watching for.

The Ascend 910C Is Not a Neutral Choice

Washington did not leave the chip selection ambiguous. The Trump administration specifically named the Ascend 910C last year, warning that purchasing those chips could violate US export regulations. That guidance asserts extraterritorial authority, meaning the US is regulating a transaction between Malaysia and Huawei, two parties outside American jurisdiction. No government has publicly refused that guidance until now.

The 910C is Huawei's flagship accelerator, not the newer Ascend 950 line, which Bloomberg reports remains in limited production . The hardware gap with Nvidia is not disputed. Huawei's accelerators trail Nvidia's by a wide margin, a consequence of years of restricted access to the production tools needed to advance chip design. One Huawei scientist argued in August that Nvidia will eventually encounter the same physical constraints that currently limit Chinese chips. That is a claim about the future. Malaysia is making a procurement decision in the present.

Telekom Malaysia as the Structural Tell

The selection of Telekom Malaysia as the state-linked operator is the analytical detail that investors should focus on. Telekom Malaysia already partners with Huawei and runs Huawei technology in its network infrastructure. Its current AI workloads, built on Nvidia, represent the transitional architecture. If the sovereign AI programme migrates to Ascend silicon, Telekom Malaysia becomes the integration layer between Chinese chip supply and Malaysian government data.

That creates a second-order exposure for every US hyperscaler with Malaysian data centre commitments. Microsoft, Oracle, and Nvidia have built their Southeast Asian expansion on the assumption that sovereign compute would remain within the US technology stack. A Malaysian government decision to route intelligence-adjacent workloads through Huawei hardware does not cancel those investments. It bifurcates the market. Commercial AI on American chips; sensitive government AI on Chinese chips. That split, once institutionalized, is structurally difficult to reverse.

Telekom Malaysia did not respond to Bloomberg's request for comment. Neither did Huawei, the Prime Minister's Office, or Malaysia's Ministry of Digital .

Washington's Two-Track Response and Its Limits

The US response operates on two tracks. A State Department spokesperson told Bloomberg that Malaysia is an important partner and pointed to a 2025 reciprocal trade agreement between the two countries . The spokesperson also warned that suppliers controlled by authoritarian regimes pose manipulation risks, without naming China or Huawei directly. That is the diplomatic track.

The security track is narrower. US officials have told Malaysian counterparts they are concerned about Huawei equipment handling intelligence material, including information shared by foreign governments. Malaysian national security council officials are scheduled to travel to Washington later this month. Separately, Bloomberg reported that Anwar's government is discussing a second procurement process that would allow Nvidia to compete more directly for the contract . Neither Nvidia nor AMD responded to Bloomberg's requests for comment.

The precedent problem for Washington is documented. In May 2025, Malaysia's deputy communications minister announced a national AI system built on 3,000 Huawei Ascend chips. Her office retracted the statement within days, without explanation. The investment ministry subsequently said the project was private and carried no government backing. Some Trump administration officials read that retreat as confirmation that US export guidance carries deterrent force. Sixteen months later, the same chip is back on the table, and the government has not distanced itself .

The Espionage Dimension Raises the Stakes

The Malaysia deliberations land inside a broader pattern of technology confrontation that institutional investors cannot separate from their semiconductor exposure. On September 7, Belgian prosecutors confirmed they have held a 52-year-old Belgian-Chinese researcher in pre-trial detention since May on suspicion of passing gallium nitride chip-making secrets to a Chinese company . The suspect worked at BelGaN, the Oudenaarde fab that a consortium of Chinese investors acquired in February 2022 and that declared bankruptcy in July 2024. Prosecutors are investigating four charges: espionage, participation in a criminal organization, misuse of company assets, and unlawful disclosure of trade secrets. A second suspect, whose identity prosecutors have not confirmed, remains at large .

One day later, on September 8, the FBI, the National Security Agency, and the Cybersecurity and Infrastructure Security Agency published a joint advisory alleging that major Chinese technology firms are conducting what they described as "industrial-scale" theft of American AI trade secrets . The advisory focused specifically on distillation, a technique that involves training a weaker AI model on outputs from a more advanced one. American officials and executives have argued that the volume and targeting of Chinese distillation campaigns against US frontier AI models constitutes systematic appropriation of proprietary technology .

Three stories, one week, one directional signal: the contest for AI capability is no longer confined to export control lists. It has moved into procurement tenders, bankruptcy investigations, and federal criminal advisories simultaneously.

Investment Positioning: Follow the Bifurcation

For institutional capital, the Malaysia decision creates three distinct exposure categories.

First, Nvidia's Southeast Asian revenue pipeline carries new concentration risk. The company has not commented publicly, and its Malaysian data centre partnerships remain intact. However, a sovereign procurement decision favoring Huawei, even on a single programme, establishes a competitive reference point that other governments in the region will cite. The four-times capacity expansion planned across Southeast Asia becomes a contested market rather than a captive one.

Second, Huawei's strategic value increases non-linearly with this contract. What the company lacks, as the source analysis makes clear, is a sovereign customer willing to be identified publicly. Egypt is a bid in progress. Malaysia would be a signed, public choice. The reputational and commercial value of that precedent exceeds the contract value itself.

Third, the BelGaN espionage case is a reminder that the European semiconductor supply chain, specifically advanced materials like gallium nitride, carries undisclosed IP transfer risk inside deals that were framed as rescue acquisitions. Investors evaluating distressed European chip assets should treat the BelGaN structure as a due diligence template, not an anomaly .

The Plocamium View

The market is reading the Malaysia story as a geopolitical headline. It is actually a pricing signal for the cost of US cloud jurisdiction.

Anwar's Cloud Act citation is not diplomatic rhetoric. It is a legal risk assessment by a head of government with access to his country's intelligence posture. When a prime minister publicly asks how one argues with the US president over data sovereignty, he is describing an asymmetric legal relationship that no amount of Nvidia chip incentives resolves. The hardware is not the problem. The legal architecture that travels with the hardware is.

Our thesis: the sovereign AI market, governments building national compute infrastructure for sensitive workloads, is structurally different from the commercial AI market. In the commercial market, Nvidia's performance lead is decisive. In the sovereign market, the Cloud Act is a competitive liability that Huawei does not carry. Malaysia's 2 billion ringgit programme is the first public test of that distinction, but it will not be the last.

The second-order play is not in Malaysia. It is in the 15 to 20 governments across the Middle East, South and Southeast Asia, and Africa that are at earlier stages of the same sovereign AI deliberation. Huawei has been courting Egypt. DeepSeek is planning a large Huawei chip order for a new data centre . The FBI advisory on distillation tells us that Chinese AI capability gaps are closing faster than export controls anticipated. If the performance gap narrows and the legal liability gap widens, the sovereign AI market tips toward Chinese silicon at a pace that current US policy cannot arrest through guidance alone.

Washington's counteroffer, packages of AI hardware, software, and financing for the Global South, is the correct response. It is also an admission that guidance without incentives does not hold.

The Bottom Line

Malaysia has not signed the contract. A second tender may yet route the sovereign AI programme to Nvidia. The last time this surfaced, in May 2025, the government reversed course within days. What is different in September 2026 is the direction of travel: it is on the record, the security council is flying to Washington to defend it, and the government is explicitly treating the decision as commercial rather than geopolitical.

The $494 million contract value is a rounding error in global semiconductor capital flows. The precedent is not. The first government to publicly choose Chinese AI accelerators for a sovereign programme will have created a reference point that every subsequent procurement committee in the developing world will cite. Institutional capital should price that optionality into Nvidia's emerging-market revenue forecasts now, not after the ink dries.