Tenstorrent raises $1.5B at $6.2B valuation to rival Nvidia

Tenstorrent, the AI chip company run by veteran processor architect Jim Keller, has closed $1.5 billion in Series E funding at a $6.2 billion post-money valuation. The company announced the round on 19 October. It's one of the largest bets yet on a challenger to Nvidia that sells both hardware and the designs inside it.

Who is paying

The round was led by Trace Capital, a new firm founded this year by Quiet Capital co-founder Ben Mahdavi, and Innovation X. Other new and existing investors include ARK Invest, Fidelity, T. Rowe Price, Morgan Stanley, Sandisk, Tether Investments, Cadence Ventures, Eclipse, Virtu Financial and XTX Markets. Tenstorrent was already backed by Bezos Expeditions, Samsung, LG Electronics and Hyundai Motor Group.

What the money is for

Tenstorrent says it will use the cash to:

• deploy its Galaxy servers at scale, serving open-weight models for both LLMs and video generation;
• advance the IP it licenses: its TT-Ascalon RISC-V CPU and Tensix AI cores, which chip designers including Samsung and LG build into their own silicon.

The company sells "at every level of the stack": a single card or workstation, full racks, superclusters or a licence. Its software stack is open source, which it says lets developers run frontier models locally and enterprises run them on-premises.

"We built one architecture and sell it at every size," Keller said. "The future of compute is open, scalable, and owned by the people building on it."

Why it stayed independent

TechCrunch notes that the round comes about five months after Intel and Qualcomm reportedly showed interest in buying Tenstorrent at valuations of up to $5 billion. Keller told Bloomberg the company chose to stay independent and raise more money after showing its Galaxy servers could run DeepSeek models.

Why it matters

Keller led work on Apple's A4 and A5 chips, AMD's Zen cores and Tesla's self-driving chip, so his company's bet on open RISC-V designs and open-weight models gets attention. More competition for AI inference hardware could, over time, mean cheaper ways to run models without renting Nvidia GPUs. The build-out isn't smooth everywhere, as Microsoft's exit from Nscale's Loughton site showed. If you'd rather run open models on your own desk today, our guide to choosing a laptop for local AI explains what hardware you need, and how much RAM is enough.