TensorWave Scores $350 Million to Advance Its AMD Cloud

TensorWave, a neocloud based entirely on AMD infrastructure, has scored $350 million in Series B funding at a $1.55 billion valuation, validating its premise of offering an alternative to NVIDIA for GPUs, particularly for enterprise inference.

“As models grow larger and workloads become more demanding, enterprises need infrastructure with the memory capacity, performance and flexibility to scale without being locked into a single ecosystem,” stated Darrick Horton, CEO and co-founder of TensorWave, in the press release.

The CEO’s reference to enterprises, not hyperscalers, is notable. AMD prides itself on the inference capabilities of its chips. And the key enterprise function for AI these days is inference, which enables AI models to operate with data specific to an organization’s applications.

That TensorWave is growing with enterprise inference demand seems to be borne out by its numbers: In 2025, the company had 8,192 AMD MI325X GPUs in its Arizona datacenter. Now, the company has additional datacenters in Florida and Pennsylvania and boasts 10,000 AMD GPUs in total. Further, the company says it’s scored 2 gigawatts of long-term datacenter capacity to expand its service offerings with AMD’s MI355X chips, which started shipping to TensorWave last August.

AMD Partnership

Overshadowing everything TensorWave does is AMD, whose venture arm co-led with Magnetar both Series A and B funding for the neocloud. The two companies cited in the Series B press release as TensorWave customers—Luma AI and Fireworks AI—have development agreements with AMD.

All of this is to TensorWave’s advantage because working with AMD and its ecosystem give TensorWave a headstart on AMD releases as well as on key improvements to AMD products.

One area needing improvement has traditionally been AMD’s software. Its ROCm environment was long criticized as falling behind NVIDIA’s CUDA software framework in many respects. Now, those criticisms are being reconsidered, as AMD has massively overhauled ROCm. Large-scale, multiyear deals with OpenAI and Meta have helped validate AMD’s progress.

Also in the AMD software area, TensorWave has partnered with Spectral Compute, a company founded in 2018 that offers a toolchain called SCALE. A superset of CUDA and AMD’s ROCm, SCALE compiles CUDA code, untouched, to produce runtimes that suit AMD chips—and other third-party accelerators. It works independently of any NVIDIA runtimes, so it’s not a “translation layer.” Spectrum’s software is also not an emulator—a term that comes with implications of bumpy performance.

Progress Evident in Funding

That TensorWave has moved ahead in furnishing AI infrastructure to enterprises is evident in the figures behind its latest funding. Its valuation of $1.55 billion is more than four times the roughly $400 million valuation reportedly linked to TensorWave’s $100 million Series A in May 2025. The company has now raised about $493 million.

Increasingly, enterprises are looking for alternatives to NVIDIA for a number of reasons, including avoidance of vendor lock-in, cost, and supply chain considerations. TensorWave, it seems, has tapped into this demand successfully.

Futuriom Take: While AMD will likely continue to play second fiddle to NVIDIA in the enterprise market for the foreseeable future, TensorWave provides an alternative that makes sense not just to hypersalers such as Meta and OpenAI but to the vast number of enterprises engaged in deploying AI to streamline operations and boost revenues. TensorWave's latest funding round validates this strategy.