$300 Million Smuggled, 100,000 Rented, $8 Billion Sold and Leased Back: One Week in the Life of an AI Chip

Three stories about Nvidia's AI chips broke within 48 hours this week. A California business owner was arrested, accused of smuggling more than $300 million worth of servers to China. Tencent, one of China's largest tech companies, was reported to have rented about 100,000 advanced chips in Oracle data centres in Southeast Asia. And Amazon was reported to be in talks to sell about $8 billion of its Nvidia chips to investors, then rent them straight back.

Each story on its own is a business headline. Together, they show what AI chips have become: something governments police like weapons, companies rent across borders, and investors finance like buildings.

What happened

The arrest. On October 1, federal agents arrested Greg Lui, 38, of San Gabriel, California, the owner of Earthmade Computer Inc. in nearby City of Industry. A grand jury indictment returned on September 29 alleges that between 2023 and 2024 Lui and unnamed co-conspirators bought export-controlled servers containing Nvidia AI chips and sent them to buyers in China without the required US licences. According to the Justice Department, they gave US manufacturers false paperwork about where the equipment would end up, shipped the servers to Malaysia and Singapore, where no licence was needed, and then re-exported them to China. Prosecutors say Earthmade received more than $176 million from two Malaysia-based freight companies between January and October 2024. Lui faces three counts: conspiracy to violate export controls, outbound smuggling and conspiracy to commit money laundering. These are charges, not convictions.

Nvidia is not accused of wrongdoing. It said the case shows "smuggling is a losing proposition", and that its work with law enforcement has led to prosecutions. Bloomberg, in an investigation published the same day, reported that officials are asking why the company missed red flags as its chips keep reaching China, and that Nvidia has removed more than half of its previously approved Asian AI-chip buyers after tightening its checks.

The rental. The Financial Times reported on September 30 that Tencent has signed a five-year deal worth about $7 billion for access to roughly 100,000 advanced AI chips in Oracle data centres across Southeast Asia, with about 30% paid upfront. It would be Tencent's largest overseas leasing deal with a US cloud provider. Reuters said it could not independently verify the report, and neither company had confirmed it. The legal point is the one that matters: US rules stop Chinese companies buying the most advanced chips, but do not currently stop them renting computing power that sits outside China.

The sale and leaseback. The Financial Times also reported that Amazon has discussed moving about $8 billion of Nvidia Grace Blackwell chips into a separate company, a special-purpose vehicle, which would sell debt to outside investors and lease the chips back to Amazon. The chips, thousands of them across more than a dozen US data centres in five states, would stay where they are. Amazon would lighten its balance sheet while still using them. Amazon declined to comment, and the talks may change. The context is scale: Amazon expects to spend about $220 billion on capital projects this year, mostly chips and AI data centres.

Why it matters

Export controls leak at the edges. The Lui case, as alleged, is a simple pattern: buy in the US, ship to a country without a licence requirement, ship on. Southeast Asian transit points keep appearing in these cases. Each prosecution shows the controls being enforced, and also how much effort it takes.

Renting is the legal workaround. If the Tencent report is accurate, a Chinese company can get large amounts of advanced computing without a single chip crossing into China. That is within current rules, but it is exactly the kind of arrangement that tends to attract new rules.

Chips are becoming a financial asset. Amazon's reported plan treats GPUs like office buildings: assets you can sell to investors and lease back. That only works if investors believe the chips keep their value. Grace Blackwell is due to be followed by Nvidia's Vera Rubin generation, and Amazon argues each generation should stay useful for at least five years. If that is wrong, someone is left holding hardware worth much less than expected. The same scarcity is driving record results elsewhere in the supply chain: memory maker Micron just reported a $54 billion quarter.

What is still unclear

• Whether the Tencent and Amazon deals are final. Both come from FT reporting. Neither Tencent nor Oracle has confirmed theirs, and Amazon's talks are ongoing.
• Whether Washington will close the rental route. There is no announced rule on remote access to chips hosted abroad.
• How Lui's case ends. He has been charged, not tried, and the co-conspirators are unnamed.
• How long AI chips really last. Amazon's five-year view is a company argument, not an industry standard.

What it means for your business

1. If you resell or ship computer hardware, know your end user. US export controls follow US-origin technology when it is re-exported, wherever you are. Resellers and integrators in Canada and Francophone Africa who handle servers with advanced GPUs should document who the final customer is and where the equipment will be used. False end-user paperwork is at the heart of the Lui charges. Do not be the unknowing middle link.

2. You probably should be renting, not buying. If the biggest companies in the world prefer to lease computing power, a small business almost certainly should. Cloud access to GPUs lets you pay for what you use and avoid owning hardware that ages quickly.

3. Expect compute prices to stay firm. When Tencent pays billions to rent capacity abroad and Amazon is looking for new ways to finance its chips, demand is not easing. Budget for AI usage costs to stay level or rise, and shop around between providers.

4. Read the location line in your cloud contract. Where your data is processed is becoming a political question as well as a technical one. Know which country your AI provider runs your workloads in.

Questions You Should Be Asking

• Does any hardware we buy, sell or ship fall under US export controls, and who checks?
• Are we buying servers we will struggle to use fully, when renting would do?
• In which countries does our cloud provider process our data, and could that change?
• What happens to our AI costs if computing prices rise 20% next year?

What To Watch Next

Any new US rule on remote access to advanced chips held abroad, which would hit deals like Tencent's directly. Confirmation or denial of the Tencent and Amazon reports. And Nvidia's response to the pressure Bloomberg described: tighter checks on buyers are the cheapest way for it to show Washington it is taking smuggling seriously.

Sources

• US Department of Justice: California man arrested for smuggling more than $300 million in export-controlled computer servers to China
• Bloomberg: Man charged by US with illegally shipping Nvidia chips to China
• CBS Los Angeles: LA County man accused of smuggling $300 million of computer servers to China
• Courthouse News: California man charged with smuggling $300 million in restricted AI hardware to China
• The Register: Californian accused of shipping $300M worth of Nvidia chips to China
• DigiTimes: California tech exec arrested in US$300M Nvidia AI server smuggling case
• Bloomberg: Nvidia faces questions over China AI chip smuggling cases
• The Next Web: Tencent leases 100,000 AI chips from Oracle in a $7bn deal, FT reports
• Investing.com (Reuters): Tencent leases 100,000 chips from Oracle for $7 bln, FT
• TrendForce: Tencent reportedly signs $7B deal to lease 100,000 AI chips from Oracle in Southeast Asia
• TechRepublic: Tencent reportedly turns to Southeast Asia for $7B Oracle AI chip deal
• Yahoo Finance: Amazon seeks to offload $8 bln of Nvidia chips to investors, FT
• Finimize: Amazon looks to move $8 billion of Nvidia chips off balance sheet