The Deep Tech

BetaKit Most Ambitious

Science is the new moat

Boris Wertz on why he invests in AI, robotics, and Web3.

When Boris Wertz started his marketplace company in 1999, creating a software platform was hard. It took a significant amount of time, money, and infrastructure to create even a minimum viable product, and the prospect of success—let alone profitability, IPO, or acquisition—was far from guaranteed.

Now an investor with his BC-based Version One Ventures, Wertz has watched over the years as the barriers to entry have been removed from the software landscape.

“With cloud compute, efficient marketing, and great developer frameworks, there’s almost no moat left around starting a software company,” he said. “Everybody can replicate it.”

In search of real differentiators, Wertz has focused his investments on AI, Web3, and robotics—three deep tech verticals with the potential to make big, unique impacts.

“Building in atoms still has a real differentiator, if you nail it,” he said. “It’s much harder to replicate, it takes much longer, and fewer people can do it.”

Across the country and around the world, there’s an insurgent emphasis on actually making things.

Countries are focused on rebuilding their physical infrastructure and shoring up their defences, both at the border and on the digital battlefield. Companies are looking for epoch-changing competitive advantages, enabled by the abilities of AI, the speed of quantum, and the power of compute.

All of this requires deep tech, a sector defined by its ability to tackle complex challenges in science and engineering. It is hard, time-consuming and expensive. But it holds the promise to fully eradicate old ways of doing things, in fields as diverse as finance, medicine, manufacturing, and agriculture.

At Version One, Wertz is particularly interested in robotics, healthcare, defence, and space.

Founders in these fields are “off the charts” smart, he said, and have often dedicated their lives and their careers to a singular focus of study.

“You don’t just wake up one day and say, ‘I’m going to build a protein sequencing thing,’” he added.

Deep tech requires the ability to imagine future capabilities and impacts that might not materialize for years. In this context, Canada has a track record of success. Wertz notes that Canadians shaped almost the entire field of AI, and sees a similar ability for the country to define innovation in drug discovery, vertical-specific robots, and Web3.

“The currency for me is just the token, the economic expression of that ecosystem,” Wertz said of crypto. “Ultimately, what’s interesting is that it enables a new ownership layer on top of the internet. That’s a powerful idea.”

Investing your time and your money in deep tech is not for the faint of heart. Breakthroughs that are years in the making are met with requests to do more, sooner.

“There’s not a single important, deep technology that came faster than people thought,” he said. “They all take time.”

The difference today, he notes, is the urgency of the work, and the excitement it is generating—not just in labs, but in boardrooms, governments, and investor portfolios.

“I think, for the first time in a long time, we’re back in building mode,” Wertz said. “And Canada is very good at being at the forefront of crazy ideas.”

Putting the prize in enterprise

Cohere (Toronto, ON)

To get enterprise to adopt AI, you have to train AI to act like enterprise.

As Canada’s leading performer in the global AI market, Cohere is making a play to be the go-to AI for large-scale enterprise customers. They’re building LLMs for the boardroom.

Co-founders Nick Frosst, Aidan Gomez, and Ivan Zhang forged their connections through the University of Toronto and Geoffrey Hinton’s team at Google Brain. Their company has raised and earned a ton of money by Canadian standards, securing a $500-million USD Series D at a $5.5 billion valuation last year, with annualized revenue hitting $70 million USD at the beginning of 2025.

But as giants like OpenAI and Anthropic raise rounds in the tens of billions, Cohere has rejected the “bigger-is-better” approach to AI models.

Cohere has developed its LLMs in the specific context of enterprise work. To give the company an edge in its quest to partner with global firms in Japan and the UAE, Cohere’s Aya model supports more than 100 under-served languages. To ease enterprise fears about AI adoption, its program, North, provides a secure, all-in-one AI workspace platform that maintains the highest standards of privacy, security, and control.

Its Command A model, unveiled last year, has showcased faster performance and better energy efficiency than leading models. It also runs with twice the context length, meaning it can sift through the larger documents produced by enterprise customers.

The company has released a stream of high-quality products, showcasing Generative AI for business with finance, manufacturing, security, and healthcare. Its clients today include the Royal Bank of Canada and Japanese tech giant Fujitsu.

Reality hits the road

Waabi (Toronto, ON)

You don’t want AI to hallucinate behind the wheel of a 20-tonne transport truck.

Self-driving cars have long been the promise of artificial intelligence, but safety risks remain the biggest roadblock to widespread commercialization. In the $900-billion long-haul trucking market, addressing this issue is seen as key to addressing a looming labour shortage.

Along this road, Waabi CEO Raquel Urtasun is speeding up the delivery of “provably safe” AI systems for self-driving trucks. Her company has also created a standard by which its Silicon Valley competitors can evaluate the accuracy of their own simulators.

“Just as we have safety standards for vehicles, we must establish clear and measurable standards for the simulators on which their safety depends,” Urtasun said.

Urtasun came to Toronto to lead Uber’s self-driving research team, and quickly established herself as a vital element of Canada’s AI community, teaching at the University of Toronto and co-founding the Vector Institute with Geoffrey Hinton and Deep Genomics founder Brendan Frey.

At Waabi, which she founded in 2021, Urtasun has developed a simulator that trains self-driving systems off the road, providing infinite high-risk scenarios, so it learns from its mistakes in a safe environment. Waabi Driver can operate in 3D environments and make decisions based on how objects move and change over time, in any setting.

Along the way, Urtasun struck a deal to incorporate NVIDIA DRIVE Thor chips into Waabi vehicles and invented a new measurement for reality that compares how AVs drive in simulated scenarios versus real-world scenarios. Right now, the company can build digital twins of the world with 99.7 percent accuracy.

A data centre that’s faster than yours

Xanadu (Toronto, ON)

Christian Weedbrook’s grandmother used her bank book to record transactions long after the technology existed to check her balance digitally.

He described her habit to evoke the slowness of what was once considered standard, so we might consider how quickly something can become outdated.

“Imagine if we didn’t have email and you had to write letters,” he said. “Imagine the slowness of communicating ideas.”

Weedbrook believes that our current computing power will one day be remembered as quaintly slow.

His company, Xanadu, builds quantum computers that can solve problems at unimaginable speeds. He imagines applying this power to drug discovery, material design, and the creation of next generation batteries. Governments are paying attention because of quantum’s potential to crack encryption.

This year, Xanadu was accepted into the US Quantum Benchmarking Initiative, launched by the Defense Advanced Research Projects Agency, or DARPA, to determine whether quantum can achieve utility-scale operation by 2033.

Weedbrook believes it can, and Xanadu published a peer-reviewed paper in the journal Nature this year announcing it had achieved a breakthrough in networking quantum computers together. Unlike other quantum systems, Xanadu uses light to generate qubits, which allows the system to operate at room temperature, facilitating better network design.

Weedbrook now plans to start the equivalent of the cloud revolution by building quantum data centres that sell access to premium compute, with the first in operation by 2029.

“The two big challenges remaining for the industry are the improved performance of the quantum computer and scalability,” Weedbrook told BetaKit. “Xanadu has now solved scalability.”

Domesticating the vaccine supply chain

AbCellera (Vancouver, BC)

One of Canada’s biggest biomedical breakthroughs was brought to you by the US Department of Defense.

In 2018, University of British Columbia spinoff, AbCellera, was asked by the US government to apply its state-of-the-art capabilities in human antibody discovery to model how vaccine countermeasures could be quickly developed in the case of a viral pandemic.

We all know what happened next.

AbCellera, founded by Carl Hansen, was in its second round of DARPA-backed trials when the COVID-19 pandemic struck. Eli Lilly took the company’s antibody treatment, bamlanivimab, to market in 2020 and AbCellera went public.

Since then, Vancouver’s bio-tech anchor continues to pursue antibodies capable of preventing and treating various diseases including cancer, metabolic and endocrine conditions, and autoimmune disorders.

And while AbCellera’s stock price has seen a drop from its pandemic peak, it has built a clinical-grade biomanufacturing plant in Vancouver, with funding from provincial and federal governments. The building stands as a high tech reminder that Canada could do more than just discover cures: we could actually make them here.

“It wasn’t a priority before COVID,” AbCellera’s vice-president of business development said of Canada’s lagging biomanufacturing capacity. “Now it should be.”

Fault-tolerant quantum computing

Photonic (Vancouver, BC)

Fault tolerance is one of the biggest barriers to useful quantum computing, as cascading errors can create catastrophic failures of computation.

But earlier this year, BC-based Photonic claimed a breakthrough in quantum error correction that it says could dramatically reduce the number of quantum bits—the basic unit of information used to encode data—required for computations.

The breakthrough comes in the form of a new family of Quantum Low-Density Parity Check codes (QLDPC) that Photonic is calling “Subsystem Hypergraph Product Simplex” codes, or SHYPS.

Such codes have long been used to reduce qubit overheads, but Photonic is the first to discover how to use them to perform quantum logic.

“Unlocking the quantum logic of high-performance QLDPC codes has been the holy grail of quantum error correction R&D for decades,” said Photonic co-founder and chief quantum officer Stephanie Simmons. “And one of the obstacles to cost-effective quantum computing at scale.”

With great compute comes great responsibility

Sasha Luccioni

Hugging Face

Data centres that power generative AI models are power and water-hungry, and are expected to become the fifth-highest energy consumer in the world sometime next year.

Sasha Luccioni is hungry, too, for the type of data that tracks the water consumption and carbon emissions of global AI companies.

After studying under AI godfather Yoshua Bengio during her postdoctoral degree at Université de Montréal, Luccioni is now the AI and Climate Lead at Hugging Face, a New-York based company that compiles open-source AI research, models, and datasets.

Like many of Canada’s leading computer scientists, Luccioni has made artificial intelligence her life’s work. But her job is to ensure it doesn’t destroy the world.

Part of the challenge lies in the fact that Big Tech companies don’t readily share their carbon footprint. Google acknowledged in 2023 that its greenhouse gas emissions rose by 48 percent since 2019, but little data is available from other tech companies about emissions and water usage from chip manufacturing and data centre performance.

At Hugging Face, Luccioni and her team build tools like CodeCarbon, which estimates the carbon emissions created by running AI programs. Her team also built the AI Energy Score, a ranking system to compare the efficiency of open-source models.

Her goal is to encourage more companies to open-source their data, so researchers like her can understand the true environmental costs of AI.

Accurate emissions data could impact public perceptions around the technology, and prompt policymakers to set an energy-efficiency threshold for an AI model before a company is allowed to operate a data centre on Canadian soil, for example.

Luccioni believes that ranking AI models on energy scores could also encourage positive competition for compute efficiency. Today, global companies have little immediate incentive to slow emissions.

“I profoundly believe that open source is the way forward for AI so that we avoid this monopoly of power,” said Luccioni. “It’s really important that AI stays accessible so we know both how it works and when it doesn’t work.”

Feature image courtesy Unsplash.