Nvidia challenger Tenstorrent raises $1.5-billion, valuing Toronto-founded company at $6.2-billion

Artificial intelligence semiconductor maker Tenstorrent Holdings Inc. has raised US$1.5-billion in fresh capital as it steps up efforts to challenge industry behemoth Nvidia Corp NVDA-Q.

The financing, which values Tenstorrent at US$6.2-billion, including the money raised, follows reports earlier this year that U.S. chip stalwarts Intel Corp. INTC-Q and Qualcomm Inc. QCOM-Q had both tried to buy Tenstorrent for upward of US$5-billion.

The Austin, Tex.-based company was founded in Toronto 10 years ago but relocated to the United States in 2023 after its first investor, Jim Keller – one of Silicon Valley’s top chip architects – became chief executive officer. The company’s Toronto office is still one of its largest hubs, where 200 of its 1,000 employees work. Tenstorrent also has operations in Silicon Valley; Tokyo; Belgrade, Serbia; and Bangalore, India.

The deal was led by two American private capital firms, Los Angeles-based Trace Capital, founded by National Football League linebacker-turned-private capital investor Ben Mahdavi, and Innovation X LLC of New York. Other backers in the deal include Agostinelli Group, Cathie Wood’s ARK Invest, Cadence Venture Capital, Maverick Silicon, Eclipse Venture Equity, Protagonist Management LLC, Fidelity Management & Research Co., Morgan Stanley, Sandisk Corp. and T. Rowe Price.

From the archives: AI chip firm Tenstorrent raises $100-million from Hyundai, Samsung

Other notable past investors include South Korea’s AFW Partners, Samsung Securities and Hyundai Motor Group and LG Electronics, plus Amazon founder Jeff Bezos, Export Development Canada and Healthcare of Ontario Pension Plan. Montreal-based Real Ventures was an early investor and generated a windfall return when it sold out last year, making back the entirety of its $88-million third fund.

Tenstorrent is building specialized hardware and software for processing a range of AI workloads, promising to deliver faster speeds than the graphics processing units (GPUs) sold by Nvidia. The upstart says its Galaxy-branded technology is equipped to run large language models that power chatbots and agents and for generating AI video.

Nvidia has become the world’s most valuable public company, surpassing US$5-trillion in market capitalization, by maintaining a tight grip on the market for AI computer chips owing to its in-demand GPUs, which have become the industry standard.

But Nvidia faces several challengers. A handful of AI compute startups are targeting the inference market, which refers to hardware for running AI models, as opposed to hardware for training models. Nvidia’s GPUs can do both but competitors are tailoring their hardware to deliver faster inference speeds. The market for inference, according to experts, is also expected to be much larger. AI company Cerebras Systems Inc., for example, went public earlier this year and has partnered with OpenAI.

Some of the biggest buyers of GPUs, namely OpenAI and Meta Platforms Inc., are also building their own chips designed to work with their AI models, both in partnership with Broadcom Inc. That could weaken Nvidia’s lock on the market and pricing power.

From the archives: AI chip maker Tenstorrent joins herd of Canadian unicorns after Fidelity-led $200-million funding

Nvidia, however, has ample resources. Last December it struck a deal to license technology from U.S. inference chip startup Groq Inc. and hired its executives.

Tenstorrent is looking to scale up manufacturing of its Galaxy servers for the fast-growing market for open-weight large language models that compete with OpenAI and Anthropic. Open-weight models are free for developers to modify, unlike proprietary models. Anyone can download the parameters – numerical values that dictate the model’s behaviour – of an open-weight model and tweak them to have more control over how the model functions.

Tenstorrent is also licensing its intellectual property to others.

“We built one architecture and sell it at every size,” Mr. Keller said in a statement. “The future of compute is open, scalable and owned by the people building it.”

George Bousis, managing partner with Miami-based venture capital firm Protagonist, said in a statement “few companies design the chip and the software that runs on it. Tenstorrent does both, and its video generation results beat industry benchmarks ahead of our expectations. We are excited to double down” on the company.

This is Tenstorrent’s third nine-figure financing; it raised US$693-million in a late 2024 deal that valued the company at US$2-billion, and US$200-million-plus in spring 2021.

Tenstorrent was founded in Toronto by Ljubisa Bajic, a former colleague of Mr. Keller’s when they worked for chipmaker Advanced Micro Devices Inc. (Mr. Keller has previously designed chips for AMD, Apple and Tesla). After Mr. Keller succeeded Mr. Bajic as Tenstorrent’s CEO in 2023, Mr. Bajic left to co-found another Toronto-based AI compute startup called Taalas; it was acquired by AMD in August for an undisclosed sum.