TORONTO — Tenstorrent is rolling out a package of technology that it says will make it easier and cheaper for cloud services and businesses to run their AI tools, as the hardware firm tries to break into the surging compute market.

It’s the firm’s first major launch since it announced a US$693-million funding round in December 2024, and since it relocated from Canada to the U.S. in November 2023. Founded in Toronto in March 2016, the firm has been led since January 2023 by Jim Keller, a well-known processor designer based in Palo Alto, Calif.

Tenstorrent’s new Networked AI system is “a new way of building AI infrastructure,” said Amr El-Ashmawi, vice-president of strategy and business development.

Nvidia currently dominates AI compute, but Tenstorrent sees room for a new player. The stack that powers its technology includes several kinds of chips and components. That includes graphics processing units (GPUs), central processing units (CPUs), and specialized AI accelerators, which all provide raw processing power. Meanwhile, memory hardware stores data while networking switches to connect everything together. Typically, each of those pieces comes from a different provider and must be bolted together. 

Tenstorrent says Networked AI puts all those features into a single system. It has also picked types of memory and connectivity that are already used widely, so customers can string together its server to form bigger compute clusters. 

The firm claims its Galaxy Blackhole hardware, which is part of the Networked AI system, is well-suited for applications like AI agents and real-time video, and produces responses from open source models faster than the leading products of competitors like Cerebras and Groq. On top of the hardware, Tenstorrent has added software that it claims makes it easier for customers to start running open source models and AI applications.

Put it all together, and El-Ashmawi claims the firm can offer the lowest cost for running generative AI. “If you look at the enterprise market, they’re trying to figure out how to use AI to improve productivity,” he said, “but most importantly, how to get it to improve the bottom line.”

Tenstorrent on Tuesday also announced an initial group of users for its new technology. That includes data-centre giant Equinix’s Distributed AI Hub in Ashburn, Va. “The idea here is to get their customers access [so that] they can implement and deploy AI [in] a much easier, more scalable approach,” El-Ashmawi said. Tenstorrent has also signed up Cirrascale, a mid-sized cloud player that will use its hardware for financial services firms, drug developers and other firms.

Tenstorrent’s largest installation will be for Ai&, a Japanese model developer which says it has secured $2 billion in infrastructure funding. The firm was founded by former Tenstorrent executive David Bennett, a Canadian.

El-Ashmawi says Tenstorrent has also done business in Canada. He declined to identify specific customers, but said they include data-centre operators, as well as government departments and large firms. Despite relocating to the U.S., Tenstorrent also continues to have a significant engineering team in Toronto.