Tenstorrent

THE PROBLEM

AI demand is growing faster than the industry’s supply of affordable, power-efficient compute. At the same time, most accelerated computing remains tied to proprietary hardware and software stacks, limiting customers’ ability to adapt infrastructure to their own workloads.

THE APPROACH

Tenstorrent is building an alternative: AI accelerators and systems, licensable RISC-V CPU and Tensix AI cores, and an open-source software stack. Its Networked AI architecture unifies compute, memory and networking, scaling from a licensable core to Galaxy superclusters over standard Ethernet. Customers can buy systems or license IP for custom silicon, giving the company multiple paths to market.

WHY THIS TEAM

CEO Jim Keller helped lead the DEC Alpha, AMD K7/K8 and Zen, Apple A4/A5 and Tesla self-driving chip programs. Co-founder Milos Trajkovic has worked across the Tensix processor from architecture through post-silicon validation, while chief CPU architect Wei-Han Lien leads RISC-V and chiplet architecture. Few teams have comparable hands-on experience turning new architectures into working silicon.

MOMENTUM

Tenstorrent closed an oversubscribed $693 million-plus Series D in December 2024 at a $2 billion pre-money valuation and later acquired chiplet-interconnect specialist Blue Cheetah. In June 2026, it announced its largest deployment yet—more than 120 Galaxy systems with Japan’s ai&—alongside TT-Ascalon S. Reuters also reported that Qualcomm had discussed an $8–10 billion acquisition; those talks remain unconfirmed and may not lead to a transaction.