Tenstorrent Raises $1.5B at $6.2B Valuation, Stays Independent in Nvidia-Challenger Bet
Jim Keller's chip startup Tenstorrent rejected buyout interest from Intel and Qualcomm, raising $1.5B instead.
Chip startup Tenstorrent announced a $1.5 billion round at a $6.2 billion valuation — more than double the $2.6 billion it fetched in late 2024. In a market where Nvidia challengers usually end in acquisition, CEO Jim Keller is staying independent.
The dramatic backdrop
Five months ago, Bloomberg reported acquisition interest from Intel and Qualcomm at valuations of up to $5 billion. Keller told Bloomberg the company chose to stay independent and raise more capital instead. One of his proof points: Tenstorrent's Galaxy servers can run DeepSeek models.
The round was led by Trace Capital (the new firm of Quiet Capital co-founder Ben Mahdavi) and Innovation X, with Cadence Ventures, Eclipse, Fidelity, SanDisk and T. Rowe Price participating.
Open silicon for open models
The new money goes toward chips tailored for serving open-weight AI models and video generation. Tenstorrent builds open, Nvidia-alternative AI silicon — a bet on compute diversity at a moment when data-center power and GPU supply both stay tight.
The timing lines up with a broader fracture in Nvidia's moat. This same week, AWS released TorchNeuron, letting PyTorch applications run on Amazon's Trainium chips with a single line of configuration — no code rewrite required. Since roughly 85% of AI applications are written in PyTorch, moves like this attack Nvidia's software lock-in, CUDA, not just its chips.
Tenstorrent's pitch converts a technical trend into a commercial one: as open-weight models like DeepSeek's become production-grade, enterprises want cheaper, more flexible silicon to run them — and don't want to be locked into one vendor's hardware, software and pricing.
What this means for your business
GPU costs remain one of the largest line items in any serious AI deployment, and the Tenstorrent raise plus the AWS move both point the same direction: the compute stack is opening up.
Three takeaways. First, keep your AI workloads portable — avoid architectures that lock you into CUDA-only features, so you can bid compute across providers. Second, when evaluating inference partners or cloud credits deals, ask about support for alternative silicon (Trainium, Tenstorrent's roadmap, AMD) — price competition is coming and flexibility is leverage. Third, track open-weight models seriously: the better they get at your tasks, the more options you have to run them on cheaper, purpose-built hardware instead of flagship Nvidia GPUs.
The era of "default to Nvidia and never think about it" is ending. For businesses, that means compute is becoming negotiable — build your AI stack so you can take advantage of that.