The AI infrastructure story this week has been about where the compute goes — Meta in India, OpenAI's reported Ohio megacampus. TensorWave's raise is about whose chips fill those buildings. And the answer it's betting on is: not Nvidia's.

The raise

TensorWave has closed a $350M Series B at a $1.55 billion valuation, led by AMD and hedge fund Magnetar Capital. That's nearly a 4x jump from its roughly $400M valuation a year ago, when the same two backers co-led its $100M Series A. The new money goes to expanding data-centre capacity and buying more AMD hardware.

The bet

TensorWave is a 'neocloud' — a cloud provider renting out AI compute — but with a twist: it runs exclusively on AMD GPUs, not Nvidia's. In a market where Nvidia's chips and its CUDA software ecosystem are near-inescapable, that's a deliberate contrarian position. Founded in 2023, the company pitches itself as the alternative; CEO Darrick Horton says it exists to "restore competition to the market."

Why AMD funding it matters

The most interesting investor here is AMD itself. A chipmaker funding a cloud that runs only its own silicon is a strategic flywheel: it manufactures demand for AMD GPUs and gives customers a credible, ready-built way to actually use them at scale — chipping at the biggest barrier to switching from Nvidia, which isn't raw silicon so much as the software and tooling lock-in around CUDA.

Why it matters

Nvidia's dominance of AI compute is one of the defining facts of this era — and one of its biggest risks, if a single vendor becomes a single point of failure (and pricing power). TensorWave is a bet that the market wants a second option, and that AMD is willing to bankroll one into existence. For everyone building on AI, more competition at the compute layer is the quiet story that could matter as much as any model launch — it's what ultimately sets the price of intelligence.