AI Minister Evan Solomon, seen departing a Liberal caucus meeting on Parliament Hill in Ottawa on Wednesday, told The Globe that the $300-million Compute Access Fund would not be able to meet every compute need.Spencer Colby/The Canadian Press
When the federal government announced a plan in April, 2024, for Canada to catch up in the artificial intelligence race, there was no time to waste. A key element was a $300-million fund to help companies pay for the cost of building and running AI models.
Ottawa held consultations, published a report, and after the fall of one Liberal government and election of another, announced in June 2025 that the program, dubbed the Compute Access Fund (CAF), was open to applications. Last month, some applicants received rejection notices offering no explanation, while others still in the running to receive between $100,000 and $5-million apiece learned they are moving to another evaluation stage. No funds have been disbursed, and the process could spill into 2026.
This might be the speed of government but it’s woefully behind the speed of AI, where new developments happen weekly. The disparity has left some Canadian AI entrepreneurs frustrated and concerned that Ottawa is moving too slowly. What’s more, some say the program could now be missing a bigger opportunity for Canada to catch up.
“When you have a program that’s a year-and-a-half in the making, you are a dinosaur by the point funds are released,” said Moody Abdul, co-founder of Klarify in Toronto, which makes AI tools for therapists to handle notes and other tasks. “We don’t have the time to do business-as-usual.” Klarify applied to CAF, which is targeted at small- and medium-sized businesses (SMEs), but was rejected.
As AI reshapes hiring, starting pay stagnates at Canada’s consulting firms
Expecting government to move as fast as a speeding industry is a tall order. That’s exacerbated by the fact that AI minister Evan Solomon’s department is not only new but has a lot on its plate, including developing a new AI strategy, a quantum strategy and boosting digital sovereignty.
“Before we start criticizing — oh gosh, who got it and who didn’t — I completely understand and appreciate SMEs that have spent a lot of time on an application,” Mr. Solomon said in an interview. “But I will also say that there was absolutely no way that a $300-million compute access fund will be able to satisfy every single compute need.”
The CAF is part of a $2-billion initiative announced in spring 2024 to stimulate the construction of AI data centres and compute infrastructure in Canada. The government has so far allotted up to $240-million to Toronto-based Cohere Inc., which builds large language models for enterprise users, and committed $42.5-million to University of Toronto for compute infrastructure for the research community. Many applicants to CAF are still waiting to find out whether they will receive funds.
“My initial impression was light, scrappy and fast,” Simon Eskildsen said of the CAF process. Mr. Eskildsen is a co-founder of Turbopuffer in Ottawa, which makes a search engine for companies to connect massive amounts of data to AI models. “Now it’s been months and all we’ve heard is that we’re still in.”
Some worry that CAF is not only moving slowly, but too focused on the wrong thing. The CAF application guide lists a few AI activities eligible for funding. That includes both building AI models (referred to as training) and using existing models as part of software. Think of it this way: OpenAI has built a model called GPT-5, which helps power ChatGPT, an application that can answer questions.
The concern some experts have is that while these foundational AI models are important, that is not where the economic gains will be. These models, along with data centres, will likely be commoditized and companies will have to compete on price, said Daniel Wigdor, chief executive of AI venture studio Axl in Toronto. “They’re all just going to get into a race to the bottom,” he said.
Instead, companies that make use of this infrastructure to build new applications to solve problems will be the winners, he said. The same was true of the internet era. Software companies such as Alphabet Inc. (Google), Meta Platforms Inc. (Facebook, Instagram) and Shopify Inc. proved to be more successful than telecom carriers and network equipment vendors such as AT&T, Cisco and the defunct Nortel Networks.
Canada should instead find ways to support AI software companies – the future Shopifys, not the Nortels. “It’s like looking at electricity and saying we should build a bunch of hydro dams without saying, wait a minute, what are we using electricity to do?” Mr. Wigdor said. “We could be the ones building the applications years ahead of anyone.”
Ottawa launches AI registry showing breadth of tech’s use in public service
Scott Stevenson, CEO of AI legal-software company Spellbook, said bureaucrats have an outdated notion of AI. He has been left frustrated after his interactions with some federal programs, including CAF. Government, he said, is too focused on backing companies that say they are doing intense research and development and building their own AI models.
“The assumptions that these panels make are really incorrect – and based on how AI worked in 2015,” he said, where “you need a bunch of researchers, a bunch of hardware, and you need to burn all this cash training models. It’s a terrible waste of money.”
Spellbook doesn’t need to train models. Mr. Stevenon’s company has secured around 4,000 customers by using existing models from the tech giants to build software that automates tasks for lawyers, such as drafting contracts. Competitors that have built their own models are struggling, dead, or “just sitting on these completely useless models,” he said.
Only a handful of large, well-funded companies such as OpenAI, Google and Cohere are capable of building general-purpose AI models and Mr. Stevenson said that custom models will underperform.
“I don’t think funding training makes a lot of sense,” said Ryan Wilson, founder of legal AI company Walter Software in Vancouver, which does no model training. Doing so may have been a good idea a couple of years ago, but the industry has changed, he said. “There will only be five or six winners in that game.”
“Not every application included model training. That is important to note,” said Deloitte Canada partner Brendan Cooper, who worked with companies to submit proposals to the program. “But it is a pretty prevalent area of focus for a lot of companies.”
Some AI researchers have challenged the industry’s focus on building huge foundational models, arguing that it could be crowding out other approaches and that smaller models still have a role. Vancouver’s Variational AI, for example, is pursuing that route for drug discovery.
Mr. Solomon said it is premature to draw conclusions about the government’s preferences. “The idea that we are overindexing on foundation models versus applications, how would anyone have any idea?” he said. “There’s not been a single announcement. What you probably have is a couple of people who didn’t get it and have complained.”
Last month’s federal budget also contains “signals,” he said, that the government is moving to support companies building AI applications. That includes changes to speed up the Scientific Research and Experimental Development program (SR&ED) and sweeten the tax benefits.
Mr. Solomon acknowledged CAF has moved slowly, but said that’s because it was deluged with applications. “The diligence process on each application took much longer than we thought,” he said. “We want to make sure that these applications are real.”
Mr. Wigdor said other solutions are needed for Canada to build AI application companies. One idea is to rejig SR&ED so that the funds do not go directly to successful applicants, but instead to their customers to subsidize the purchase of their offerings. That would help startups win orders and let the market pick winners. He has also called for public funding of AI institutes focused on commercialization, not just research.
Adam Keating, CEO of St. John’s-based CoLab and a member of the federal government’s AI task force, has also advocated for supporting AI application companies. “That’s what we can do quickly,” he said, adding the government appears serious about moving faster. “As much as there may be some folks frustrated about this right now, I do think they are trying to change things,” he said.