Data center infrastructure startup TensorWave raises $350M to help break Nvidia’s AI chip monopoly

Cloud-based artificial intelligence infrastructure startup TensorWave Inc. said today it has closed on a bumper $350 million Series B funding round as it strives to meet demand for an alternative to Nvidia Corp.’s graphics processing units.

The round was co-led by Magnetar and AMD Ventures, the venture capital arm of the chipmaker Advanced Micro Devices Inc. Maverick Silicon, Nexus Venture Partners and Western Frontier also participated in the round, which brings TensorWave’s valuation to $1.55 billion.

The startup is notable for its refusal to have anything to do with Nvidia. It provides access to specialized, high-performance cloud infrastructure such as AI accelerators, renting it to customers who need resources for both training and inference. But uniquely, it relies exclusively on chips and software made by AMD.

TensorWave is quite the outlier, because virtually everyone in the AI industry is scrambling to get their hands made on Nvidia’s GPUs. Nvidia is the world’s top chipmaker for a very good reason: Its GPUs are widely viewed as the most effective for running AI workloads.

But TensorWave Chief Executive Darrick Horton begs to differ. In an interview with the Wall Street Journal, he said he purposely avoids buying Nvidia’s hardware or any of its other products. That’s because he thinks that the company has too much influence over the AI infrastructure market and believes that’s a bad thing for the health of the industry.

“We wanted to figure out how we can solve problems for customers and restore competition to the market,” Horton explained. “I don’t like buying things from monopolies. You don’t have a lot of leverage.”

Las Vegas-based TensorWave was founded in 2023, during the early days of the AI boom, at a time when Nvidia had become “a monopoly by default,” thanks to its status as the world’s largest supplier of GPUs. Back then, Horton said, customers were already eager to diversify into other AI chip suppliers, and that demand has only increased since then.

TensorWave has expanded rapidly in the last three years, and it currently has three operational data centers stacked with AMD’s alternative GPUs. Its facilities are located in Arizona, Florida and Pennsylvania, and each one has 10,000 of AMD’s Instinct processors and delivers computing capacity that’s equivalent to around 14 megawatts of electricity. It’s but a drop in the ocean compared with the massive amounts of compute used by companies like OpenAI Group PBC, Anthropic PBC and Google LLC.

That’s why TensorWave needs more money, so it can expand. Besides its own data centers, it also leases space from other operators and fills them with AMD’s chips and other equipment. It has signed leases for 500 megawatts of capacity, but most of those data centers haven’t yet come online.

Horton said he wants to increase that number to 2 gigawatts in the next year, and will use the funds from today’s round to invest in the chips, power-supply equipment and other gear needed to do that. While building out its capacity, TensorWave has become a key partner of AMD’s, helping the chipmaker to improve its custom software platform ROCm. It’s important work, because a lot of companies have complained that ROCm is harder to use and buggier than Nvidia’s CUDA platform.

However, Horton said his company has helped Nvidia to improve the software so much that these days, “it’s pretty much plug-and-play.” As a result, AMD hopes that it can increase its presence in the market for AI inference, or workloads where trained AI models respond to queries. The chipmaker has positioned its GPUs as a more efficient alternative to Nvidia’s hardware for inference.

TensorWave and AMD aren’t the only companies trying to provide enterprises with an alternative to Nvidia’s chips. Another major rival is Cerebras Systems Inc., which is known for its dinner plate-sized chips that also target inference. It went public last month.

Image: TensorWave