More than 10,000 delegates described in the source as the central nervous system of the AI industry assembled in Montreal this week at the fourth annual All In conference, the largest of its kind in Canada, to discuss where AI research, development and deployment are headed next. Attendees included thought leaders from government, big tech, startups, safety groups and scientific institutions.
Canada leads in research, lags in AI adoption
Canada’s pioneering researchers — including Yoshua Bengio, Geoffrey Hinton and Ilya Sutskever — helped spark the revolution behind many current AI technologies. Hugo Larochelle, scientific director at Mila, said Canada still “punches above its weight on discovery-driven AI research.” He added: “Our expertise in a lot of areas has no comparison [elsewhere in the world], including with multi-agent systems, continual learning, in the application of AI in sustainability and the environment and for scientific review,” he said during the conference.
The country’s research leadership, however, makes what All In speaker Rick Glumac called an “adoption problem” more pronounced. Statistics Canada reports that 12.2 per cent of Canadian firms used AI to produce goods or deliver services in 2025, a level that remains well below adoption in the U.S. and other countries. Conference commentary pointed to a general shortage of venture capital in Canada and the possibility that promised productivity gains from AI may be overstated.
Safety and the $300 million commitment to LawZero
Safety was a major theme at All In. Canadian Minister for Artificial Intelligence and Digital Innovation Evan Solomon and Germany’s federal minister for digital transformation and government modernization, Karsten Wildberger, jointly committed nearly $300 million in funding to support LawZero, a Canadian non-profit AI safety organisation.
Yoshua Bengio, LawZero’s scientific director and an AI safety advocate, warned about recent incidents in which AI agents “escaped, deceived and conspired against their creators” at companies including OpenAI and Anthropic. “The smarter the AI, the more useful it is, but also the more dangerous if it doesn’t do what it’s supposed to do,” he said. He criticised industry leaders who said they did not have plans to prevent such behaviour: “When I hear things like this, I want to raise my hand. Because I do have a plan.” The plan discussed at the conference involves building models that understand the world without having goals of their own, allowing them to provide knowledge and insight without taking independent action.
AI sovereignty and international collaboration
Many sessions emphasized that AI sovereignty will require international collaboration to reduce dependence on the U.S. and China, which dominate the race to build larger models. On the conference’s first day, Toronto-based Cohere and Germany’s Aleph Alpha announced they had merged to create what they say is the largest frontier AI company outside Silicon Valley, valued at about $20 billion. A.J. Bhadelia, Cohere’s VP of governmental affairs, said their goal is to ensure “that the world has another option for AI business applications if powerful players like the U.S. and China decide to shut their models off to other countries.”
Research partnerships were also highlighted. Mila announced a partnership with a German academic institution, the Research Center for Artificial Intelligence. Larochelle said: “I think you can see in this conference a lot of promising ties that we’ve made and will make with that country,” naming “responsible AI, safety, multi-agent systems and AI for energy” as areas for collaboration.
About 40 other countries were represented at All In. The United Arab Emirates had a large booth, and Singapore hosted a lunch promoting its bid to be Asia’s AI hub. The International Centre of Expertise in Montréal on Artificial Intelligence and Animikii Indigenous Technology announced a partnership with Sandy Bay Ojibway First Nation to give Indigenous groups control over how their data is collected, stored, stewarded and used with AI. Animikii CEO Jeff Ward said: “Each community wants to govern their data in unique ways that are specific to their cultural context and traditional government structures. We can customize this for nations, communities, non-profits and community organizations so they can build how they want to and not have to put their data into other people’s buckets – colonial data buckets, for example. Sovereignty means sovereignty. Sovereignty means choice in how your data is governed and designed.”
AI is showing up in unexpected places
Startups and industry showed AI applications beyond software and research. WindoSmart presented what it says are the first major innovations in skyscraper-cleaning technology in more than 40 years: AI-enabled soap-and-water-spraying drones that can distinguish windows from walls and scan buildings for damage and decay. “The AI application allows for full automation,” said WindoSmart founder Mitko Piperkov. He acknowledged job losses for rope-access window washers but said the technology “limits risk at all times” and creates new ground-based technician roles: “You’re still going to need technicians on the ground.”
Other exhibitors included an AI-enhanced webcasting platform offering simultaneous subtitling in eight languages and an “AI-powered autonomous trailer unloading” solution. Entrepreneurs pitched applications across mining, construction, property management and finance, including multiple “hedge-fund-in-your-pocket” concepts. French cosmetics company L’Oréal showcased AI-driven tools to analyse skin and hair cells to model signs of ageing and propose personalised care routines.
Robotics, data centres and public concern
Robotics remains more challenging than digital AI, but robotic arms, drones and inspection machines featured on the trade show floor. Nvidia’s Liila Torabi, director of engineering for Isaac Sim, said the robotics shift will be gradual: “I think the efficiency of production lines, warehouses and factories – that’s the first line that will change,” she said. “The other part is going to change in bits and pieces. Like, nobody has a fully autonomous car now. But all the cars coming out today have autonomous parking or at least some sensors. It’s not going to be a fully automated world immediately.”
Physical infrastructure for AI is also visible in communities. Earlier in September the federal government and the Federation of Canadian Municipalities announced a set of principles to guide “responsible data centre development,” saying data centres must create lasting local benefits; not shift electricity costs to Canadians; minimize water use and environmental impacts; be transparent about local impacts; and bring strategic value to Canada. Despite those principles, vandals spray-painted the words “Burn the data centres” on the side of Mila’s building a day before the conference began.
Craig Tavares, president and COO of Buzz HPC, which builds data centres in Canada, urged community engagement and stressed technical measures such as closed-loop cooling to reduce water use. “Noise and vibration is another concern,” he said. “But if your window is open in the summer, you’ll hear your neighbour’s AC before you hear a data centre.” He warned that slow community processes could shift compute needs abroad: “Slowing down AI forces us to rent computing capacity elsewhere,” he said. “It’s a national priority. I don’t think the question should be whether we build data centres, but whether we can build them responsibly.”
