Getting a high-end GPU is still the annoying part of building with AI. The companies that reserved capacity years ago are fine. Everyone else is in a queue.

That is the problem GMI Cloud says it is raising money to fix. On Wednesday the company announced a $668 million Series B, led by ARCHIV, with Nvidia participating.

The money is meant to add GPU capacity in the U.S., Taiwan, and the rest of the Asia-Pacific region, and to scale the platform that runs models once they are trained.

Founder and CEO Alex Yeh put the figure a little more loosely in his own post, calling it "over $660 million," and he named more of the room: DSC Investment, Trend Micro, KB Investment, Kyobo Life, KT Corporation, and others. The Information reported the same $668 million number.

If GMI is new to you, the product is easier to explain than the funding. You rent Nvidia GPUs. The site lists H100s, H200s, and Blackwell, and it is built more for running models for users than for a training-cluster keynote.

Yeh says he started the company in 2021 with one question: who actually gets access to compute? His answer, then and now, is that the biggest buyers lock it up early. The rest wait.

He also says the industry is bad at hitting its own dates. More than a quarter of expected data center capacity missed its completion date last year, he wrote. GMI's counter is that every cluster it has committed to has come online on schedule, helped by ties into Taiwan's supply chain. That is his account, not an outside audit.

Two growth numbers sit next to the funding. Revenue the company says it already has under contract is more than nine times what it was at the end of 2025.

The platform now processes about 4 trillion tokens a week. A token is a small piece of text a model reads or writes, so the figure is there to say people are actually using it.

Yeh did not name customers. He described them. A lab behind a widely used open-source AI agent. An inference service for developers. A router in front of thousands of models. A studio making production AI. A cybersecurity company watching millions of users. Next, he said, come teams decoding DNA and designing new materials.

This is a different size of round from the last one people wrote up. In October 2024, TechCrunch reported an $82 million Series A led by Headline Asia, with Thailand's Banpu and Taiwan's Wistron involved. That took total capital raised then to about $93 million. A $668 million Series B is what you get when GPU demand is still ahead of supply.

We have been looking at the same squeeze from the other direction. DeepSeek and Huawei are open-sourcing programming tools so developers can write for Ascend chips without living inside Nvidia's CUDA world. GMI is betting the other way: stay on Nvidia, take Nvidia's money, and put more of those GPUs in more places.

Yeh closed by saying the company has something special coming, and told people to watch the GMI account. No details yet.