Founded in 2020, Blumind has developed products based on what’s known as analog computing that it says use anywhere from 100 to 1,000 times less energy than current AI chips.
Already an Insider? Log in
Get Instant Access to This Article
Become an Ottawa Business Journal Insider and get immediate access to all of our Insider-only content and much more.
- Critical Ottawa business news and analysis updated daily.
- Immediate access to all Insider-only content on our website.
- 4 issues per year of the Ottawa Business Journal magazine.
- Special bonus issues like the Ottawa Book of Lists.
- Discounted registration for OBJ’s in-person events.
A startup with a major Ottawa presence has raised millions of dollars in fresh funding to commercialize a cutting-edge, low-power computer chip designed to make artificial intelligence applications more efficient.
Founded in 2020, Blumind has developed products based on what’s known as analog computing that it says use anywhere from 100 to 1,000 times less energy than current AI chips.
The company announced this week it has closed a $20-million series-A funding round led by Cycle Capital and BDC Capital. Existing investors Fusion Fund, Two Small Fish Ventures and Real Ventures also joined the round.
Headquartered in Toronto, Blumind has 16 full-time employees and about half a dozen part-time staffers split between the Greater Toronto Area and Ottawa. Niraj Mathur, the firm’s Toronto-based co-founder and CEO, told Techopia this week the company plans to open an office in Kanata soon as it ramps up hiring in anticipation of getting its first chips to market by 2026.
“Ottawa has been great. We’ve found really strong talent,” said Mathur, adding the firm will be adding more engineers, product development staff and sales and marketing experts to its team over the next few months.
Mathur, who began his career as an engineer at Nortel’s Ottawa campus, says Blumind is solving a key problem: how to power ever-bigger and faster AI applications that are stretching the limits of current computing capacity and energy grids.
While Moore’s Law once held that the number of transistors on a microchip could double every two years, leading to a corresponding spike in computer power and efficiency, Mathur says that’s no longer the case.
“You just can’t shrink (transistor technology) any more,” he explains. “It’s now becoming cost-prohibitive to build these chips in these very, very advanced process nodes. What we’re finding is the power-performance ratio has basically plateaued. We’ve kind of run out of juice on Moore’s Law.”
Yet as usage of generative AI tools like ChatGPT and DeepSeek soars, he says, the need for faster and more efficient chips has never been greater. And that’s where Blumind’s technology, developed by the firm’s Ottawa-based co-founder and chief technology officer John Gosson, comes in.
Today’s digital devices, Mathur explains, use a logic-based computing model that runs in a binary world of ones and zeros, where signals are either on or off.
That logic-based architecture worked well for decades, but generative AI’s need for continuous computations to fuel its inference-based machine learning has put incredible strain on existing computing technologies and the energy grids that power them. As a result, massive data centres are now consuming as much power as entire cities, stretching the grid almost to the breaking point.
“This has rendered our existing chips even more impotent because the (legacy) architecture was never really devised for doing a workload like this,” Mathur says.
By contrast, Blumind’s analog chips operate more like a human brain, he explains, processing information by recognizing patterns in a way that’s much more efficient than existing technology.
“It’s a very different paradigm,” Mathur adds.
Blumind is currently working with several “tier-one” customers to refine its chip technology, though Mathur says he can’t name them.
The company is tailoring its chips for products on the “edge” of the AI ecosystem – that is, things like smart watches and other wearables, products connected to the Internet of Things and next-generation medical devices, rather than data centres.
“We can’t be everything to everybody, so we’ve chosen to focus on applications where energy efficiency is critical,” Mathur says. “That’s where the need is here and now.”
The firm’s investors say they’re excited about the tech’s potential.
“Blumind’s patented approach using analog AI will not only revolutionize edge computing but also help solve the growing energy demand imposed by the adoption of AI,” Cycle Capital founder and managing partner Andrée-Lise Méthot said in a statement.
“We are enthusiastic to support the team, as we believe their highly differentiated technology will play a critical role in the future of AI.”
Blumind is part of a growing field of young companies that are developing new chips and other technologies aimed at overcoming the challenges currently facing AI training and inference.
California-based EnCharge AI, for example, announced in February it raised more than US$100 million in a series-B round to ramp up development of its own analog chips for AI applications. Like Blumind, EnCharge is also channeling its efforts into chips that run AI models “on the edge” rather than for training applications.
Undaunted, Mathur says the more innovation in analog computing, the better.
“This is all still very new,” he says. “Nobody has successfully commercialized an analog chip before. We intend to be the first. But quite frankly, competition is good. This is going to be a huge market, and there are going to be multiple players needed to support it.”
