Last December, Steven Woods got a “Yes” that had been more than a year in the making—or perhaps his whole career, depending on how you count. The computer scientist had always worked in and around the field of artificial intelligence. Now an investor, he was set to lead a statement deal to back one of the field’s buzziest startups. 

The technology couldn’t be hotter. “Every VC in the world is saying, ‘AI, AI, AI,’” said Woods, a partner and CTO at Montreal-headquartered Inovia Capital, a mainstay financier of private Canadian software companies. To show its bonafides in the space, the firm on Monday is publishing its thesis on generative systems, outlining the opportunities and challenges it sees for systems that produce text, code and images at a prompt.

Across 20 predictions, it anticipates ever-more knowing and personalized tools, as well as growing competition in the space, as large tech firms build on early wins and startups try to dethrone them.  

Inovia has invested in AI before, backing the likes of legal-research platform Ross Intelligence, drug-discovery toolmaker BenchSci and autonomous equipment developer Clearpath Robotics. While plenty of its portfolio firms were building products infused with machine learning, AI wasn’t “talked about at every single board meeting of every single company, [as it is] today,” said partner Karamdeep Nijjar. 

The firm’s growing focus on the technology tracked developments both out in the world—read: ChatGPT—and in-house. Waterloo, Ont.-based Woods joined Inovia in July 2021, after over a decade heading Google’s Canadian engineering corps. His right hand Kory Jeffrey followed in March 2022. Their three-person team wrote the white paper and serves as an internal AI brain trust, while Inovia’s other investors explore the opportunities and impacts in the individual verticals they cover like retail, travel and health care. 

That statement deal Woods had been working was Toronto-based Cohere’s US$270-million Series C. Inovia led the round, announced in June, with participation from Oracle, Nvidia and Salesforce.

It was a major undertaking. Around the time Woods moved to Inovia, he checked in with Cohere co-founder Nick Frosst to get a sense of how the startup was going. (Years earlier, Woods and Jeffrey had played a part in Frosst going to work for deep learning pioneer Geoffrey Hinton at Google Brain.) The technology was more advanced than he’d imagined. “I saw some demos that, at the time, for me were quite shocking,” recalled Woods, positively. 

Informal conversations eventually turned into an investment opportunity when it came time for Cohere to raise its Series C. It was not an average deal. “It was a very large round at a very high price—it’s not typical of what we usually do at Inovia,” said Nijjar. But Cohere, he noted, was “one of the world’s most important technology companies being built in our backyard that aligned with our thesis.”

Still, the partners had questions, many of them about competition in the space. Some wondered whether a firm building large language models (LLMs) could survive between cloud platforms and applications developers. “Google, Microsoft, Amazon—aren’t they going to do something like this?” said Jeffrey. “Yes, of course they are.” All three have, but they’re also offering clients models from other providers, which generates cloud revenue for the big firms.

Partners also asked about OpenAI, which seized the public imagination and possibly a lot of market share when it launched ChatGPT in November 2022. The San Francisco-based firm remains a leader, Woods acknowledged. But businesses can’t simply plug an OpenAI API into a product and expect results. Cohere has worked with partners like Oracle to optimize models for enterprise clients’ needs, he noted. “That put a degree of reality on this for our partners.” 

The last part of diligence involved others on the Inovia team spending time with Cohere CEO Aidan Gomez and president Martin Kon to understand their approach to building the business, which by then was scaling revenue rapidly, according to Nijjar. 

In late December 2022, Inovia’s partners signed off on leading the round, out of its US$325-million Venture Fund V. Having made the commitment, Woods and Jeffrey said, the firm’s investment teams started looking for other generative AI opportunities in their own verticals.

Tools like ChatGPT are built atop a stack of technology layers. At the bottom is “bare metal,” Jeffrey said—chips and other hardware. Above that come cloud platforms, which provide the compute power and infrastructure to train and run AI systems. Inovia has yet to invest in either. 

Rather, the firm starts writing cheques at the next layer, the model-builders. Companies like Cohere and OpenAI turn huge amounts of data, compute and engineering expertise into programs that can respond to questions, produce paragraphs of passable text and real-ish photos, or search through and summarize documents. Beyond its LLM pick, Inovia is also looking at startups building domain-specific models using proprietary data. For example, it just invested in Montreal-based Reliant AI, a stealth startup founded by ex-employees of Google DeepMind to build generative AI for knowledge workers. 

Above the models are firms selling developer tools for tasks like managing prompts and ingesting data; Inovia’s white paper predicts model builders will eventually launch their own offerings in this space. At the top of the stack sit applications, the generative products and services that are wowing businesses and browsers. (OpenAI’s ChatGPT is an application, and GPT-4 is the model under it.) Inovia has done some seed-stage deals for tooling companies out of its new US$25-million Discovery Fund, while its venture funds have backed application developers.

Jeffrey and Woods cite Spellbook as an example of the new kinds of companies generative AI has enabled. The Toronto firm had been building legal-automation products for a few years when it launched an LLM-based tool last summer to help lawyers draft and review contracts. It was initially a marketing play, but sign-ups skyrocketed. “It was very clear that we had hit this wild inflection point, and that lawyers were loving this product,” said CEO Scott Stevenson. Inovia participated in Spellbook’s $10.9-million round in May. 

In a competitive AI funding market, founders say Inovia’s team and technical expertise make them an attractive addition to cap tables. The firm has convened meetings of CEOs, CTOs and other executives at portfolio companies to share practices for code-generation tools, copyright infringement concerns and other generative issues.

Woods brought Inovia “this real ability to quantify the risk in making a technology bet,” said Tomi Poutanen, CEO of Toronto-based Signal 1 and co-founder of Radical Ventures. Historically, Poutanen noted, most Canadian investors have shied away from funding novel technologies. 

Health-care providers use Signal 1’s AI tools—first developed at St. Michael’s Hospital in Toronto—to monitor and manage patients. When Poutanen was deciding whether to launch, he called Woods at Google, which had published research on similar technology. “He gave me a prediction that there was a market, but nobody else was doing it,” Poutanen said. After Woods moved to Inovia, the fund joined Signal 1’s April 2022 US$10-million seed round. 

Most VCs acknowledge there’s at least some hype in the hot AI funding numbers, and that investors with limited knowledge of the sector will get burned. The size of the rounds keeps growing, as startups spend big on expertise and compute. “If properly applied [generative AI] can actually change the economics of an industry,” said Woods. But “we don’t want to over-index.”

The technology is evolving quickly. Inovia’s white paper predicts that the current transformer-based LLMs will keep getting better in the near term, and eventually be beaten by newer approaches. It also forecasts that companies offering generative AI products and services will engage in pricing wars, even as compute remains expensive and scarce. 

The authors expect to be wrong more often than right with their 20 predictions. “This stuff changes so ridiculously rapidly,” said Woods. For now, there are deals to be made.