Tenstorrent hits $6.2B valuation to scale open-weight AI chips

Tenstorrent, the inference-focused chip startup led by veteran processor designer Jim Keller, has raised $1.5 billion at a $6.2 billion valuation. That's more than double the $2.6 billion it was valued at in late 2024, and it comes five months after Intel and Qualcomm reportedly showed acquisition interest at valuations of up to $5 billion.

Trace Capital and Innovation X led the round, with Cadence Ventures, Eclipse, Fidelity, SanDisk and T. Rowe Price also in. Keller told Bloomberg the company chose independence after demonstrating its Galaxy servers could run DeepSeek models. The money goes toward chips built to serve open-weight AI models and video generation.

We like this bet. The AI hardware market has been one company deep for too long, and inference is where costs actually pile up once models leave the lab and start answering real requests. A chip company that explicitly targets open-weight models is building for the part of the ecosystem that anyone can download, modify and run. That's good for developers and for buyers who don't want every workload priced by a single vendor.

The proof point Keller cited is telling. Running DeepSeek models on Galaxy servers is a demonstration aimed at the open-weight world, where Chinese labs have been among the most aggressive releasers. That alignment is commercially smart. It also ties Tenstorrent's fortunes to a model supply that's politically contested in Washington. If restrictions on Chinese models tighten, the most compelling demos could become harder to sell to certain customers.

Walking away from a possible acquisition is the kind of decision that looks brilliant or reckless only in hindsight. Independence means Tenstorrent keeps control of its roadmap. It also means it has to win on its own against Nvidia's software lock-in, which has been the real moat, not just the silicon. A fat round buys time; it doesn't buy developers. I'm watching whether the software and tooling story keeps pace with the hardware.

The broader picture: Tenstorrent sits in a crowded wave of Nvidia challengers, and TechCrunch reports Etched has been fielding offers at $40 billion or more. Valuations across this category are climbing fast. Some of these companies will matter enormously. Others will find that beating Nvidia on a benchmark is far easier than beating it in a procurement meeting.

Tenstorrent is about ten years old and positions itself as an inference specialist rather than a training-chip competitor. Open-weight models are those whose parameters are published for anyone to run, as opposed to closed models served only through a company's own API.