Canadian startups take their place in the chip market

At the Computex conference this week, Nvidia previewed a new generation of chips — something it plans to do every year to hold on to its 80% share of the AI chip business — while the likes of AMD and Intel debuted their own new silicon to try and keep up.

But Canadian startups aren’t letting that keep them from finding a place in the chip market.

  • Taalas makes fully customized chips, building the AI into the hardware for bigger, more efficient models. It emerged from stealth in March with $50 million in funding.
  • Untether AI’s chips are specialized for inference — the term for actually running an AI model after it has been built and trained — while being as energy efficient as possible.
  • Tenstorrent builds computers specifically for AI. They use less power by avoiding the bottlenecks that come with using graphics chips, as Nvidia does.  

Conversation starter: Ljubisa Bajic co-founded Tenstorrent in 2016 before moving on to start Taalas. Tenstorrent is now led by its former CTO Jim Keller, whose Silicon Valley resume includes Apple’s first self-designed iPhone chip, AMD’s Zen chips, and Tesla’s first self-driving car chip.

Zoom out: Bob Beachler, VP of product at Untether AI, says there is still plenty of room for other chip companies to coexist with and complement the multinationals. Not only is Untether AI backed by Intel, but it is also working with chip designer Arm on self-driving car tech.

  • But who's to say they couldn’t also compete? When Nvidia first came on the scene in the 90s, it was taking on AMD, Intel, and Sun Microsystems.
  • “We feel this is particularly true of AI, which is such a different, disruptive workload,” Beachler said. He added the next big change in AI computing “will come from a startup.”

Bottom line: While the big guys are making beefier chips to train better AI models, Canadian startups seem well-positioned to take on other future areas of demand, like inference (as more models finish training and hit the market) and efficiency (as clients begin thinking of the costs of running AI).