A US startup is borrowing $600 million to buy chips for a Chinese app

Why it matters: the US government restricts exporting its very most advanced AI chips directly into mainland China. It doesn't, however, ban Chinese companies from remotely using compute that's physically located somewhere else, as long as that somewhere else isn't itself restricted.

South Korea is classified as a Tier 1 country under the US government's AI Diffusion Rule framework, meaning it faces no special limits on receiving advanced chips. That single classification is what makes this whole arrangement possible.

• What's restricted: shipping advanced Nvidia chips directly into China.
• What isn't restricted: a company in an approved country like South Korea buying those chips and letting a Chinese company use the resulting compute remotely.
• The result: Xiaohongshu can effectively access chips it couldn't import directly, without anyone physically moving hardware across a restricted border.

Yes, and recently. In late 2025, PaleBlueDot reportedly sought a $300 million loan for a strikingly similar deal, routing Nvidia chip access to Xiaohongshu through a data center in Tokyo instead.

In real life it's a bit like renting a storage unit in a neighboring town because your own town won't let you keep certain things at home, except here the "things" are some of the most sought-after computer chips on Earth, and the storage unit comes with a $300 million financing agreement.

By the numbers: doubling the deal size from $300 million to $600 million, and shifting from Japan to South Korea, suggests this isn't a one-off workaround. It looks more like PaleBlueDot building a repeatable playbook for getting compute to Chinese clients through allied countries.

Bloomberg's reporting names Brookfield Asset Management as one of the potential lenders being discussed for part of the $600 million. Brookfield is a massive global asset manager, and its potential involvement signals that mainstream institutional capital sees this as a legitimate, bankable structure, not some shadowy back-channel arrangement.

Who's affected: for Xiaohongshu, more reliable compute access means better AI features in an app used by hundreds of millions of people for shopping recommendations, lifestyle content, and social search, functions increasingly powered by AI behind the scenes.

The big picture: PaleBlueDot's deal isn't happening in a vacuum. South Korea has become a genuine hub for AI data center investment this year. In July 2026, Naver, Nvidia, and Brookfield separately announced plans to expand a South Korean AI data center from 55 megawatts to 200 megawatts, a project valued at roughly $10 billion.

That expansion is equivalent to around 100,000 Nvidia GPUs. Nvidia committed $1 billion directly to it, with Brookfield offering up to $9 billion as the project's capital partner.

That's a different, larger deal than PaleBlueDot's, but it shows the same underlying dynamic: South Korea's favorable export classification, existing power infrastructure, and willingness to host large-scale AI compute have made it an attractive base for exactly this kind of arrangement, whether the end customer is a domestic Korean company or, as in PaleBlueDot's case, a Chinese one routing compute through it.

Xiaohongshu, known as RedNote outside mainland China, is a major Chinese social platform that briefly became globally famous in early 2025 when a large wave of US TikTok users migrated there during a US ban scare. It's grown into a genuinely serious AI-powered product now, using machine learning heavily for content recommendations and search.

What's next: running those AI systems at scale for a platform with hundreds of millions of active users requires exactly the kind of high-end Nvidia hardware that's hardest for a Chinese company to source directly today. That specific gap is precisely what deals like this one are built to close.

Background: offshore compute access has become a recognized pattern well beyond this one company. Reporting has separately described Chinese AI firms accessing advanced Nvidia computing power through data centers located in Southeast Asia, exploiting the same underlying gap: export control regimes are built around where physical chips get shipped, not around who ultimately gets to use the computing power those chips produce remotely.

That gap didn't go entirely unnoticed by regulators. US officials previously moved to close a related loophole that had let Chinese-owned subsidiaries based outside China buy advanced AI chips directly, after reports suggested hundreds of thousands of chips had been acquired that way.

PaleBlueDot's structure is different, since it doesn't rely on Chinese ownership of the buying entity, but it lands in the same broader category: creative use of geography to reach a market the rules were designed to keep restricted.

Who's affected: every company racing to build AI products in China now has a visible, apparently workable template to follow if they need compute they can't source domestically or import directly into the country. That's a meaningful, public signal to the entire Chinese tech industry, not just to Xiaohongshu specifically.

The catch: based on current reporting, nothing here has been described as illegal. The arrangement appears to work within the specific wording of export controls, which regulate where physical chips can be shipped, not who can remotely access compute hosted in an approved country.

That's exactly what makes this story worth watching closely going forward. It's a live, ongoing example of how sophisticated financing and clever geography can route around a policy's clear intent while still staying technically inside its literal written rules. Whether regulators eventually close this kind of gap, the way earlier loopholes involving offshore subsidiaries were closed, will say a lot about how seriously the current framework treats structures like this one.

• This deal itself won't change anything you use directly. It's about back-end AI infrastructure for a Chinese app, not a product launching in your market.
• It's a useful real-world example of how export controls actually work in practice, and where their limits are, if you've ever wondered how "banned" technology still ends up reaching restricted markets.
• Watch for regulatory response. If US officials view this as undermining the intent of chip export rules, expect scrutiny of similar Korea- or Japan-routed deals going forward.
• This pattern is likely to repeat. A workaround this clean, with mainstream financial backing, rarely stays a one-off.