Jim Keller's Tenstorrent raises $1.5 billion at a $6.2 billion valuation

What Tenstorrent announced

On 19 October 2026, Tenstorrent announced that it has closed a $1.5 billion Series E at a $6.2 billion post-money valuation. The AI chip company, led by chief executive Jim Keller, says the money will fund volume manufacturing and global commercial expansion, as chip designers, governments building sovereign infrastructure, and inference platforms look to own their AI compute instead of renting it.

The raise more than doubles the company's value. TechCrunch reported that Tenstorrent was valued at $2.6 billion in late 2024, and that the new round comes five months after the company reportedly drew acquisition interest from Intel and Qualcomm at valuations of up to $5 billion. Tenstorrent's release does not mention those talks.

Keller summed up the pitch in one line in the release: “We built one architecture and sell it at every size.” He added that the future of compute is “open, scalable, and owned by the people building on it.”

Who put money in

The round was led by Trace Capital and Innovation X. Trace Capital is a new firm, founded in 2026 by Ben Mahdavi, who previously co-founded Quiet Capital. Innovation X, founded in 2016 by Roger Monteforte, is a late-stage venture firm that says it manages more than $4.5 billion across its funds.

Tenstorrent lists a long set of new and existing investors alongside them: Agostinelli Group, ARK Invest, Brett Barna, Cadence Ventures, Eclipse, Fidelity Management & Research Company, Maverick Silicon, Morgan Stanley, Protagonist, Sandisk, T. Rowe Price, Tether Investments, Troluce Capital, Virtu Financial and XTX Markets. Earlier backers named in the release include Bezos Expeditions, Samsung, LG Electronics and Hyundai Motor Group.

Two names stand out. Virtu Financial and XTX Markets are trading firms, and SiliconANGLE reports that Virtu already runs Tenstorrent Galaxy servers to power on-premises AI trading agents. Sandisk, a memory maker, is a supplier-side investor. Tenstorrent does not break down how much each investor put in.

What the money is for

The release gives three uses for the money. Tenstorrent will deploy its Galaxy servers at scale, keep serving open-weight models for language and video generation, and advance the AI and CPU designs it licenses to companies that build their own chips. SiliconANGLE adds that some of the funds will go to research in the RISC-V ecosystem and to hiring around the world.

The focus on open-weight models is the commercial point. Tenstorrent is not training its own frontier model. It wants to be the hardware that runs models others have released, such as DeepSeek's, which customers can host themselves. TechCrunch reports, citing Bloomberg, that Keller said the company chose to stay independent and raise more money after showing that its Galaxy servers could run DeepSeek models.

Protagonist co-founder George Bousis, quoted in the release, said Tenstorrent's video generation results “beat industry benchmarks ahead of our expectations.” The release does not say which benchmarks those were or publish any results.

What Tenstorrent sells

Tenstorrent sells compute at every level of the stack, from a single PCIe card or desktop workstation to racks, superclusters and licensed chip designs. Its chips use Tensix AI cores alongside CPUs based on RISC-V, the open instruction set that anyone can build on without paying for a licence to the instruction set itself.

On the hardware side, SiliconANGLE describes Blackhole PCIe cards with RISC-V cores and GDDR6 memory, a liquid-cooled TT-QuietBox desktop workstation built on those cards, and Galaxy servers that can be networked into large clusters. It says Equinix offers Galaxy systems to enterprise customers. The company also licenses its TT-Ascalon RISC-V CPU and Tensix AI cores to chip designers, and names Samsung and LG among them.

That spread is unusual. Most challengers sell either chips or servers or a cloud service. Large cloud companies are going the other way and building chips only for their own data centres, as with Microsoft's Maia 300 and Amazon's Trainium.

The open software bet

Nvidia's strongest advantage has long been its software, which most AI code is written for. Tenstorrent's answer is to open-source its own stack. SiliconANGLE lists three parts: TT-Forge, a compiler that takes models from frameworks such as PyTorch; TT-Metalium, a software development kit for writing custom kernels; and TT-LLK, for low-level kernel tuning. As our explainer on AI chip metrics notes, raw compute counts for little if the software cannot use it.

The same week brought similar moves from bigger rivals. Amazon said it is opening TorchNeuron so PyTorch code runs on Trainium without a rewrite, and Modal said it plans to put NVIDIA's Vera CPUs behind its agent sandboxes. Each company is trying to make it easier to move existing AI code onto its hardware.

Tenstorrent's release says the open stack lets developers run frontier models locally and lets companies deploy models on their own premises with what it calls industry-leading latency and throughput. It does not publish figures to support that claim.

Why it stayed independent

TechCrunch describes Tenstorrent as a ten-year-old company, and the release lists offices in Santa Clara, Austin, Toronto, Belgrade, Tokyo and Bangalore. Keller is one of the best-known chip architects in the industry. The release credits him with Apple's A4 and A5, AMD's Zen and Tesla's Full Self-Driving chip.

The reported takeover interest from Intel and Qualcomm earlier in 2026 was the alternative to this round. Selling at up to $5 billion would have folded Tenstorrent into a larger chipmaker. Raising at $6.2 billion keeps it independent and puts a higher price on it, but it also raises the bar the company now has to clear in revenue and shipments.

Tenstorrent is one of several AI hardware start-ups attracting large sums. TechCrunch notes that chipmaker Etched was reported earlier this month to be fielding investment offers at valuations of $40 billion or more.

Limits and open questions

The release leaves out the numbers that would show how the business is doing. Tenstorrent does not disclose revenue, how many Galaxy servers it has shipped, its order backlog, or which foundry and process node it will use for volume manufacturing. It also does not explain the T. Rowe Price footnote attached to that firm's name in the investor list.

Its performance claims, from industry-leading latency to video benchmarks beaten ahead of expectations, come without published results. The Intel and Qualcomm interest is a report, not something Tenstorrent has confirmed. AI Lookout has not tested Tenstorrent hardware.

Running open-weight models also ties Tenstorrent to what other labs choose to release. If the best models stay closed, or new open models are tuned first for Nvidia hardware, Tenstorrent's software team has to keep catching up.

What it means for AI hardware buyers

For companies and governments that want to run open models on hardware they own, the round makes Tenstorrent a better-funded option, with a product range from a single card up to licensed silicon and an open software stack. The investor list, which mixes asset managers, trading firms that already use the hardware, and a memory supplier, suggests buyers as well as financiers are betting on it.

The next tests are practical: shipping Galaxy servers in volume, publishing performance data that buyers can check, and keeping the software ready for each new open model as it arrives. A $6.2 billion valuation buys time to do that, not proof that it will work.