Tenstorrent raises $1.5B at a $6.2B valuation

By VK

Published 20 Oct 2026, 12:53 UTC

Reporting from TechCrunch

Jim Keller's chip company closed a Series E led by Trace Capital and Innovation X to scale Galaxy servers for open-weight AI inference.

What it means for founders

Another inference supplier now has volume money. If you serve open-weight models such as DeepSeek or Kimi, ask your inference hosts whether they plan to offer Tenstorrent capacity, and run your own prompts before trusting any vendor's tokens-per-second numbers.
Open software makes a trial cheap. Because the stack is open source and the hardware comes in card and workstation sizes, a small team can test locally before committing to racks. Still budget engineering time to port and tune a serving setup built around GPUs.
Video generation is a stated target. Tenstorrent names open-weight video alongside language models, and says its systems beat GPUs on that work. Teams with large GPU bills for video should ask for trial access and measure cost per clip themselves.
Watch for pricing and partners. The useful signals next are published prices for Galaxy systems or hosted tokens, new cloud and inference partners, and third-party benchmarks on the models you actually run.

The story

Tenstorrent, the AI chip company led by veteran processor designer Jim Keller, has closed a $1.5 billion Series E at a $6.2 billion post-money valuation, it announced on Monday, October 19. Trace Capital, launched this year by Ben Mahdavi, who helped found Quiet Capital, led the round with Innovation X. Fidelity, T. Rowe Price, Morgan Stanley, ARK Invest, Eclipse, Cadence Ventures, Sandisk and Tether Investments were among the other backers.

The new price is well over double the roughly $2.6 billion attached to its Series D in late 2024. TechCrunch notes the raise lands about five months after Bloomberg reported that Intel and Qualcomm had shown interest in buying the company, at prices as high as $5 billion.

Why Tenstorrent stayed independent

Keller told Bloomberg the company chose to stay on its own and raise more money once its Galaxy servers had shown they could run DeepSeek models, according to TechCrunch. DeepSeek itself is nearing a $12 billion round.

The money is earmarked for deploying Galaxy servers at scale to serve open-weight language and video generation models, and for improving the Tensix AI cores and RISC-V processor designs that Tenstorrent licenses to chipmakers such as Samsung and LG. "We built one architecture and sell it at every size," Keller said in the announcement. The product line spans everything from one accelerator card or a workstation up to full racks, superclusters and custom IP, and its software stack is open source and published on GitHub.

The pitch against Nvidia

At its TT-Deploy JP event in June, Tenstorrent claimed its Galaxy Blackhole superclusters served Kimi K2.6 at 900 tokens per second per user and DeepSeek-R1-0528 at more than 400, and said the systems can sit beside an existing GPU fleet over standard Ethernet or run alone. Those figures are the company's own. Its largest deployment so far is with Japan's ai&, which runs more than 120 Galaxy systems for inference served inside the country.

Tenstorrent is one of a crowd of hardware startups challenging Nvidia's lead in AI chips, Etched among them. The round also arrives amid heavy spending on AI compute more broadly, including Lambda's plan to raise up to $4 billion.

What we don't know yet

Prices. Neither the announcement nor the coverage gives figures for Galaxy servers, cards or hosted inference.
How the speed claims hold up in independent testing, and which GPUs they were measured against.
Revenue, order backlog, or how many Galaxy systems the new money will put into service and when.