Tenstorrent Raises $1.5B Series E for Open AI Compute

The Tensix core gives Tenstorrent a commercial choice: build it into a computer the customer can buy, or license it to a chip designer making something of its own. That choice runs through the company's new $1.5B Series E, which funds the manufacturing and deployment of Galaxy servers alongside further AI and CPU intellectual-property development.

Announced on October 19, 2026, the closed round values the Santa Clara AI computing company at $6.2B post-money. Trace Capital and Innovation X led the financing, backing CEO Jim Keller's effort to reach customers at several levels of the computing stack. For enterprises and chip designers, the significance is the range of ownership decisions Tenstorrent wants to support, from running a model on a purchased system to building a processor around licensed technology.

A computer sale and a silicon relationship

Tenstorrent's product range connects cards and workstations with racks and superclusters. Its licensing business reaches a different customer: an organization that wants AI or CPU building blocks inside its own silicon. The company identifies Samsung and LG as licensees, and names TT-Ascalon RISC-V CPU IP alongside its Tensix AI cores.

Those relationships carry different responsibilities. A buyer choosing a finished machine needs to evaluate deployment, software support and the behavior of its workloads. A chip designer choosing IP has to consider how that technology fits a much longer design process. Tenstorrent's commercial argument stretches across both decisions, making the boundaries between product engineering and customer engineering especially consequential.

The new money supports volume manufacturing and global commercial expansion, according to the announcement. For a company offering this much of the stack, expansion also means maintaining a coherent relationship between the components it licenses and the computers it sells. Customers approaching from different directions still need a clear answer about which parts they control and which parts their supplier will support.

That is the useful reading of the round for infrastructure buyers. Tenstorrent is financing several ways to acquire compute, with different levels of involvement for the customer. The purchase can begin with a workstation or with the architecture of a future chip; the resulting engineering relationship will look different in each case.

The compiler belongs in the buying decision

Tenstorrent's TT-Forge page makes the software work visible. The MLIR-based compiler stack is in public beta and provides paths from machine-learning frameworks toward execution through TT-NN and TT-Metalium. TT-XLA addresses JAX and PyTorch projects, while TT-Forge-ONNX provides another frontend for model conversion.

For an engineering team, that stack connects a proposed hardware purchase with the models it actually needs to run. Framework support is a starting point for that evaluation. The team's own operators, numerical requirements and deployment configuration still belong in the conversation, especially when a compiler is evolving in public. Open-source access gives that team something concrete to inspect and adapt as it works through the fit.

This puts a practical meaning around Tenstorrent's openness pitch. A developer can participate in the software surrounding the accelerator, while a silicon customer can license IP for a different level of customization. Neither relationship should be collapsed into a promise that every workload moves effortlessly, or that commercial chip IP is free because it uses an open instruction set.

Tenstorrent announced Galaxy Blackhole availability at scale in April 2026. The Series E now directs further capital toward scaled deployment. Buyers evaluating that progression will need to connect the available systems with their software workload and operating requirements; the relevant performance is the performance of the application they intend to keep running.

A longer history than the latest valuation

Tenstorrent's founding record names Ljubisa Bajic, Ivan Hamer and Milos Trajkovic, who started the business in 2016. The same historical announcement records Bajic stepping back from a full-time role in 2023. That chronology matters when older descriptions of the company circulate with executive titles attached.

The current leadership page identifies Jim Keller as CEO and Keith Witek as COO. Trajkovic remains a co-founder and Senior Fellow leading systems engineering and foundational software. Keller's background includes major chip work at AMD, Apple and Tesla; Witek's responsibilities include operations, licensing and strategic deals. Their respective remits help explain the technical and commercial breadth that a systems-and-IP business has to sustain.

The earlier Series D announcement, dated December 2, 2024, reported more than $693M raised at a $2B pre-money valuation. Samsung Securities and AFW Partners led that round. The distinction between that pre-money figure and the new $6.2B post-money figure should travel with any valuation comparison, rather than disappearing inside a headline about growth.

Among the Series E participants are ARK Invest, Fidelity Management & Research Company, Morgan Stanley, Sandisk, T. Rowe Price, Tether Investments and Protagonist. Their backing places more capital behind the manufacturing and software work connecting Tenstorrent's architecture to different kinds of customer. As Galaxy deployment expands, those customers will approach with their own models, design choices and expectations about how much of the machine they want to own.