Washington has an obvious blind spot in its Nvidia (NASDAQ: NVDA) export restrictions. The United States can block the company's most powerful chips from being shipped directly into China, but that does not necessarily stop Chinese AI companies from using them. If those processors sit inside a data center in another country, a Chinese developer can potentially pay to use the computing power remotely.

Nvidia builds many of the fastest processors used to train AI systems, which is why the U.S. has made chip export controls a major part of its effort to stay ahead of China. Nvidia's GB300 is among the products that cannot currently be sold into China, though some less powerful chips can still be exported there. The rules have been tightened in successive rounds since they were first imposed in October 2022, after Nvidia shipped China-specific chips such as the A800 and H800 that were engineered to sit just under the original thresholds. Chinese companies have reportedly been obtaining access to more advanced computing through cloud services running from other countries, including locations in Southeast Asia.

Chinese AI companies rent Nvidia computing power from data centers outside China

The core problem is that current U.S. restrictions focus mainly on who buys a chip, where the chip is shipped and who physically has it. They do not automatically prevent a company in China from paying to use a machine that is sitting in another country. For AI developers, that distinction matters less in practice: training advanced models requires clusters of thousands of high-end GPUs, so access to computing power, not legal ownership of the hardware, is what largely determines what a lab can build.

Cassia King, a senior researcher on the Compute Policy team at the Institute for AI Policy and Strategy, told CNBC that Moonshot's reported use of a facility in Thailand could still be legal "so long as Moonshot isn't actually buying and owning the physical hardware directly." King also explained that the U.S. system "controls physical AI chips. It does not cover remote access to those chips."

According to CNBC, the White House defended the administration's current approach. "The Trump administration has implemented the most rigorous export control regime in modern history, and is committed to safeguarding America's national and economic security," the official said.

Extending the rules to remote access has already been on the table in Washington. In January 2025, the Commerce Department proposed rules that would require U.S. cloud providers to collect "know your customer" information and to report when foreign clients use American infrastructure to train large AI models - an approach aimed at computing power rather than physical shipments of chips.

The company operating the service also said customers do not own the chips behind the infrastructure or gain physical control over them. "The companies we service do not have ownership, potential future claim or physical access to the chips that power our solutions," the spokesperson said. "Any permitted access to our services, infrastructure or technology is fully compliant with all applicable regulations."

Nvidia builds deeper ties across AI infrastructure as Bank of America holds its rating

The fight over chip access is playing out while Nvidia shares remain up for 2026, although the stock has not climbed as quickly as it did last year. NVDA is trading around $225, after gaining about 39% last year, based on historical pricing data from Yahoo Finance.

Bank of America (NYSE: BAC) analyst Vivek Arya believes the market is worrying too much about the risks that come with Nvidia taking on more financial exposure outside its main balance sheet. That exposure has grown alongside Nvidia's own investment push, which has included multibillion-dollar agreements with AI developers such as OpenAI and Anthropic announced in 2025. Arya kept his Buy rating and left his $350 price target unchanged, which represents a 55.5% gain from the current price at press time.

Bank of America also views Nvidia as more than a company that sells GPUs, arguing that it is getting involved in more parts of the huge AI buildout - a spending push in which the largest cloud providers are collectively investing hundreds of billions of dollars a year in data centers and chips - and securing resources that are already difficult to obtain. The company also brings in money by renting out GPU capacity and generates a large amount of cash.

But if Nvidia keeps using some of its own money to support other companies, projects and infrastructure around AI, investors cannot treat every dollar of free cash flow as money that will eventually come back to them. BofA deals with that by separating Nvidia's cash generation into two parts. About 50% is counted as cash that can go back to shareholders through share repurchases and dividends, while the other half is money that could be used for investments, financing and other spending tied to Nvidia's wider AI business.