Singapore has found itself at the centre of a high-profile fraud case involving the alleged illegal export of AI-capable servers, potentially containing Nvidia chips, to Malaysia, as Reuters first reported. The case has drawn international scrutiny as authorities investigate whether these transactions violated US export controls aimed at restricting China’s access to advanced semiconductor technology.

The US Department of Commerce first asked Nvidia back in December 2024 to investigate how its AI chips ended up in China despite strict export controls, according to a report by The Information. In response, Nvidia has instructed major distributors like Super Micro Computer and Dell Technologies to conduct spot checks on their Southeast Asian customers. However, smugglers have reportedly found ways to evade detection, including duplicating serial numbers and modifying them within the server’s operating system to bypass restrictions.

Despite these enforcement efforts, Chinese universities and research institutions have continued to procure Nvidia chips through resellers.

The Biden administration at the time intensified its crackdown on China’s access to advanced semiconductors, expanding last year’s ban on high-end AI chips and restricting exports to 140 companies before Donald Trump took office. Nvidia has emphasised that it requires its partners to strictly adhere to export regulations, warning that unauthorised sales harm its business rather than benefit it.

However, the persistence of grey market transactions underscores the challenge of enforcing export controls in a globalised supply chain. 

Early investigations and legal action

The scandal surfaced in late February 2025, when Singaporean authorities charged three individuals—two Singaporeans and one Chinese national—with fraud related to the procurement and shipment of high-performance servers, Channel News Asia reported. These servers, supplied by US firms Dell Technologies and Super Micro Computer, were allegedly exported to Malaysia under false representation about their end-users.

The suspects include CEO of Aperia Cloud Services Alan Wei Zhaolun, the company's COO Aaron Woon Guo Jie and Li Ming, a Chinese national linked to Luxuriate Your Life. 

According to prosecutors, Wei and Woon deceived Dell and Super Micro in 2024 by falsely claiming that the servers would remain with authorised end-users. Instead, the equipment was allegedly redirected to unauthorised parties, with fraud linked to Aperia Cloud Services totalling $250mn. Meanwhile, Li Ming faced separate fraud and cybercrime charges for falsely listing his company as an end-user in 2023, and for conducting unauthorised financial transactions in June 2024, costing $140mn.

Raids

By early March 2025, the case escalated as Singaporean authorities conducted multiple raids, arresting nine individuals. Reports suggested that Singapore was being used as a transhipment hub to bypass US restrictions on AI chip exports to China, as reported by Channel News Asia.

On March 3, Singapore’s Home Affairs and Law Minister K. Shanmugam confirmed that the seized servers likely contained US-controlled AI chips, intensifying concerns over illegal technology transfers. During a court session on March 13, bail was set at SGD$1mn ($750,000) for Li Ming, SGD$800,000 for Alan Wei, and SGD$600,000 for Aaron Woon. Authorities imposed strict conditions on Li, including electronic tagging and exclusion zones around Singapore’s immigration checkpoints.

Malaysia’s response and global repercussions

In the wake of Singapore’s crackdown, Malaysia’s Ministry of Investment, Trade, and Industry followed suit by launching an investigation into whether local laws had been violated regarding the shipment of these servers. While Malaysian authorities have not confirmed any findings linking the country to unauthorised re-exports of Nvidia chips to China, the probe remains ongoing to ensure compliance with international trade regulations.

The US first imposed AI chip export restrictions in 2022 to limit China’s ability to develop cutting-edge artificial intelligence. However, enforcement remains a challenge, with concerns that intermediaries in Singapore and Malaysia may be facilitating illicit shipments.

The scandal has seen intensified scrutiny in the wake of DeepSeek’s January 2025 launch; a Chinese AI platform suspected of using Nvidia’s advanced chips. This development contributed to a $1 trillion sell-off in US tech stocks, underscoring the financial consequences of geopolitical tensions over semiconductor trade.

What happens next?

US authorities are now investigating whether DeepSeek obtained restricted chips through unauthorised channels, though Nvidia has stated there is no evidence of its AI chips being sold to the company via Singapore.

The Singapore Nvidia chip fraud case thus continues to unfold, with authorities tracing financial transactions, scrutinising company records, and determining whether export control laws were breached. Given the high stakes, further arrests or charges could emerge in the coming months.

With Singapore a critical global trade hub, the case is shaping up to be a key test of its legal system’s ability to handle complex international fraud and its commitment to enforcing technology export restrictions amid rising US-China tensions.

Despite the situation, Jenseng Huang, Nvidia’s CEO introduced Blackwell Ultra at the company’s annual GTC conference, an upgraded version of its AI chip family set to launch in late 2025, alongside its next-generation Vera Rubin GPU, expected in 2026, CNBC reported. The company is also maintaining its new annual release cycle, aiming to keep up with the increasing demand for AI processing power from major cloud providers like Microsoft, Google, and Amazon. 

Huang emphasised the escalating computational requirements of AI, noting that Nvidia's latest chips will help data centres handle more complex AI tasks more efficiently. Additionally, Nvidia announced new AI-focused PCs, networking components, and software tools, reinforcing its leadership in AI infrastructure. The event featured major industry collaborations, including a partnership with General Motors for self-driving technology and telecom industry projects for 6G networks.

A key highlight was Vera Rubin, Nvidia’s first custom CPU design, named after astronomer Vera Rubin and based on its Olympus core. The Vera CPU, combined with the Rubin GPU, will deliver 50 petaflops of AI inference performance—more than double the capacity of Nvidia’s current Blackwell chips. Nvidia also unveiled a structural change in its GPU classification: future chips composed of multiple dies will now be considered separate GPUs, with the 2027 Rubin Next model expected to quadruple processing power. This shift highlights Nvidia’s focus on scaling up AI workloads as demand for more efficient reasoning models grows. Huang stressed that "agentic AI"—which enables AI to think and reason—will drive greater adoption of Nvidia's high-performance chips.

Despite concerns over China’s DeepSeek R1 model, which reportedly operates with fewer Nvidia chips, the company views it as an opportunity rather than a threat. Huang argued that new reasoning-based AI models will require higher computational power, ensuring continued demand for Nvidia’s chips. The newly announced Blackwell Ultra will be optimised for these reasoning-based AI models, offering cloud providers a premium AI service that could generate 50 times more revenue than the previous Hopper generation. As the company expands its AI ecosystem with enhanced hardware and software solutions, Nvidia remains firmly positioned at the forefront of the AI revolution.