US Restricts Remote Use of AI Chips, Closing the Cloud Loophole
機械翻訳 / Machine-translated

機械翻訳 / Machine-translated
@aifriends
AI Friends(https://aifriends.jp)のクロスポスト公式アカウント。AIツールの紹介・使い方・できることを、中学生でもわかるやさしい日本語で届けます。
"If we stop selling chips, we can halt AI development." If that was the thinking, it may have been a bit too optimistic. On November 16, 2026, the United States announced new rules to close that loophole. This article provides a straightforward explanation of what was decided and how it relates to Japan.
The Bureau of Industry and Security (BIS) within the US Department of Commerce issued a press release on November 16, 2026.
The title is "Rule Addressing Remote Access to American Super Intelligence Technology."
It is formatted as an interim final rule (a rule that takes effect immediately while public comment is still being collected), and was developed at the direction of the Treasury Department and the National Security Council (NSC).
The purpose is to prevent existing export controls from being bypassed through remote access — in other words, to bring the regulatory burden of using chips over the internet into alignment with that of physically shipping them.
However, the press release itself does not specify the names of targeted countries or detailed obligations. The details below are based on reporting by GIGAZINE. Always consult the original text of the rule for authoritative information.
Previous export controls focused primarily on prohibiting the sale of high-performance AI chips to China.
But the picture changes when Chinese companies rent GPUs (the high-performance components used in AI computation) at data centers in third countries. The chips never enter China — yet cutting-edge AI models can still be trained on them.
To use a smartphone analogy: it's like someone who has been blocked from buying a device instead operating a friend's phone remotely over the internet.
This development didn't come out of nowhere. According to AI industry publication AI Weekly, BIS had already issued related guidance on May 31, 2026.
That guidance established that selling advanced computing equipment to companies with parent companies in mainland China or Macao requires an export license — even if the buyer itself is located outside China. Whether a license is required depends on where the parent company is headquartered, not where the subsidiary operates.
The same article noted that data centers in Southeast Asia had been the primary workaround route. The new rule takes that effort one step further.
The main details as reported by GIGAZINE are as follows:
These rules place responsibility not only on those who sell chips, but also on those who lease access to them. "We're just renting it out" will no longer be a sufficient defense.
GIGAZINE cites an issue involving a Singapore-based company called Megaspeed as background context for these rules.
Megaspeed is a major Nvidia customer in Asia. It is reported to have built data centers in Southeast Asia and provided services to Chinese companies including Alibaba.
US authorities reportedly began investigating around 2024, suspecting that some of the Nvidia chips purchased by Megaspeed may have been illegally diverted to China.
Megaspeed, however, denies any smuggling. The allegations have not been proven at this time. Please keep this in mind when reading related coverage.
GIGAZINE characterizes the new rule as "the US government's first policy-level response to the series of issues surrounding Nvidia."
Comparing the previous regulations with the new rule makes the differences clearer.
China's options are also shifting. Experts note that as controls tighten, Chinese companies may accelerate their transition to domestically produced AI chips.
If that happens, US companies could, in the long run, lose customers and influence in the Chinese market.
Following the announcement, shares of Nvidia, AMD, Intel, and Micron fell. Concerns emerged that the regulations could dampen demand for AI-related hardware.
At the same time, significant uncertainties remain. Regarding the May guidance, AI Weekly noted that it was unclear "how far the rules would apply to Southeast Asian data centers already processing AI workloads for Chinese-linked clients" and "how existing contracts would be treated."
For the new rule as well, many details of implementation will likely become clearer before it takes effect.
Some readers may think this has nothing to do with Japan. But there are several potential points of exposure.
Consider, for example, Japanese telecommunications companies or cloud operators with data centers in Southeast Asia. If their customers include companies with Chinese parent firms, the burden of verification and reporting could increase.
Next, consider Japanese startups developing AI using GPU cloud services in Southeast Asia. If the providers they use become subject to authorization requirements, pricing and contract terms could change.
Japanese manufacturers with subsidiaries in China are another concern. If AI is being used for factory image inspection or demand forecasting, it may become necessary to revisit the source of the computing resources being accessed remotely.
That said, these are possibilities as of now. There are no confirmed reports that Japan is among the countries requiring authorization. The actual impact will need to be assessed based on the full text of the rule and how it is implemented.
A. According to GIGAZINE, it takes effect 90 days after publication in the Federal Register. The BIS press release does not state a specific effective date.
A. Reports name five countries where authorization will be required, and Japan is not among them. However, depending on business partners or the data centers being used, there may be indirect effects.
A. The reported rule concerns restricting remote use of advanced AI chips. There is no confirmed information suggesting that Japanese users will lose access to general AI services.
A. Computing resources used solely for commercial recommendation systems — such as product recommendation displays — are reported to be exempt.
A. Reports indicate that US authorities are investigating on suspicion, but Megaspeed denies any smuggling. No determination of illegality has been made at this time.
As a next step, if you are using overseas GPU cloud services, now is a good time to confirm which country your provider operates in and who that operator is.
This article is a cross-post from AI Friends.