Tenstorrent raises $1.5 billion to expand open-weight AI inference chips
Tenstorrent, the AI chip startup led by microprocessor engineer Jim Keller, has raised $1.5 billion at a $6.2 billion valuation. The company is focused on inference hardware and plans to use the capital to scale chips designed to serve open-weight AI models and video-generation workloads.
The valuation is more than double the $2.6 billion Tenstorrent received in late 2024. The financing arrives five months after reports that Intel and Qualcomm had shown acquisition interest in the 10-year-old company at valuations of up to $5 billion.
New capital backs an independent path
Trace Capital, a firm launched by Quiet Capital co-founder Ben Mahdavi, and Innovation X led the round. Other backers included Cadence Ventures, Eclipse, Fidelity, SanDisk and T. Rowe Price.
Keller told Bloomberg that Tenstorrent opted to remain independent and raise additional funding after demonstrating that its Galaxy servers could run DeepSeek models. The company is positioning that result as evidence that its systems can support a class of models increasingly relevant to organizations seeking deployable, modifiable AI weights.
Inference is the commercial focus
Tenstorrent is targeting inference, the stage at which a trained model processes prompts or other inputs in production. Its planned investment is directed at chips tailored to serving open-weight models as well as video generation, rather than being framed as a broad effort to replace every component of the AI hardware stack.
The company joins a wider group of AI hardware startups seeking to challenge Nvidia's dominant position. That group includes Etched, another AI chip maker that TechCrunch reported was receiving investment offers at valuations of $40 billion or more.
What the funding signals
The transaction gives Tenstorrent substantial resources to build and scale its product strategy while remaining independent. It also shows that investors continue to fund alternatives aimed at the growing operational demand for AI inference.
For businesses planning AI deployments, the practical implication is to evaluate inference infrastructure against the models and workloads they expect to serve, including open-weight and video-generation requirements, rather than treating accelerator choice as a one-size-fits-all decision.