Cohere and Canada’s AI Sovereignty Bet: Can Toronto Compete With Silicon Valley?

Cohere is the test of whether a Canadian company can compete at the frontier of enterprise AI without moving to Silicon Valley, and the Canadian government has decided that the answer matters enough to spend billions on it. This article traces the research legacy of Geoffrey Hinton, Yoshua Bengio and the Vector and Mila institutes, explains Cohere’s enterprise strategy and its funding path from a US$40m Series A to a reported US$20bn transatlantic merger, examines the C$2bn Sovereign AI Compute Strategy and the June 2026 national AI strategy, and asks whether Toronto and Montreal can hold on to their talent. It is part of the Canada Company Stories hub.

Why does Canada have a claim to AI leadership at all?

Canada’s claim rests on the fact that the two most important ideas in modern AI were developed in Toronto and Montreal. Geoffrey Hinton, at the University of Toronto, kept neural networks alive through decades when the field had moved on, and his students produced the 2012 ImageNet breakthrough that launched the deep-learning era. Yoshua Bengio, at the Université de Montréal, built the laboratory that became Mila, and Richard Sutton at the University of Alberta wrote the textbook on reinforcement learning. Hinton and Bengio shared the 2018 Turing Award, and Hinton won the 2024 Nobel Prize in Physics.

The federal government moved early to institutionalise that advantage. The 2017 Pan-Canadian AI Strategy, led by CIFAR, was the first national AI strategy in the world and funded three institutes: Vector in Toronto, co-founded by Hinton, Brendan Frey and Raquel Urtasun with about C$200m of public and private money; Mila in Montreal, formalised that year as a partnership between four universities; and Amii in Edmonton. Shopify and Uber were among Vector’s corporate backers, and the federal government added C$60m for the three institutes in 2021.

The output was real. Canada ranked first among G7 countries in AI research publications per capita, and a Deloitte study found that 6% of the world’s top 0.5% of AI researchers were based in Canada. More than 4,000 people enrolled in or graduated from Canadian AI programmes in the 2022–23 academic year. Mila alone had roughly 1,000 students and researchers and 100 faculty by 2022.

What Canada did not produce, for a long time, was companies. Element AI, Bengio’s commercialisation vehicle, raised about US$340m and was sold to ServiceNow in 2020 for far less than investors had hoped. Hinton’s own start-up was bought by Google in 2013. The pattern of great research, weak commercialisation is the backdrop to everything that follows.

How did Cohere emerge from this ecosystem?

Cohere emerged directly from the transformer paper. Aidan Gomez, as a University of Toronto undergraduate interning at Google Brain, was a co-author of the 2017 paper “Attention Is All You Need”, which introduced the architecture behind every modern large language model. In 2019 he founded Cohere in Toronto with Nick Frosst, another Google Brain researcher, and Ivan Zhang; all three had been at the University of Toronto.

The company chose a different path from OpenAI and Anthropic. Rather than building consumer chatbots, Cohere built models for enterprises that needed privacy, control over where data lived and deployment on their own infrastructure. The Command family of models, the Aya multilingual research models and, from 2025, the North workspace platform all target regulated industries. Oracle integrated Cohere into its Fusion Cloud applications from 2023; Fujitsu co-developed the Japanese-language Takane model; RBC signed up for North for Banking in January 2025; and Dell became the first hardware vendor to ship North on-premises in May 2025. About 85% of revenue comes from private deployments rather than API calls.

The funding followed the strategy. A US$40m Series A led by Index Ventures in September 2021 and a US$125m Series B led by Tiger Global in February 2022 were followed by a US$270m Series C in June 2023 at a US$2.2bn valuation, with Inovia, Oracle, Salesforce and Nvidia participating. In July 2024 the company raised US$500m at US$5.5bn from investors including Cisco, AMD and Fujitsu, and in December 2024 it received C$240m of public money for its data-centre project.

The team has also been reinforced with people from larger technology firms. In August 2025 Cohere hired Joelle Pineau, formerly head of AI research at Meta and a Mila faculty member, as chief AI officer, and François Chadwick, formerly of Uber and Shield AI, as chief financial officer. The company has grown to more than 500 employees with offices in Montreal, New York, San Francisco, London, Paris and Seoul.

What did the 2025 funding round and revenue numbers show?

The 2025 round showed that investors still believed in a sovereign, enterprise-focused alternative to the American frontier laboratories. On 14 August 2025 Cohere announced an oversubscribed US$500m round at a US$6.8bn valuation, led by Radical Ventures and Inovia Capital, two Canadian firms, with existing investors AMD Ventures, Nvidia and Salesforce Ventures and new investor the Healthcare of Ontario Pension Plan. A US$100m extension in September 2025 lifted the valuation to about US$7bn.

Revenue had finally begun to scale. Annual recurring revenue was about US$62m at the end of 2024, reached about US$100m by May 2025 and US$150m by October 2025, and was reported by CNBC at US$240m in February 2026, ahead of the company’s own US$200m internal target. That implies growth of roughly 290% during 2025, with quarter-on-quarter growth above 50% through the year, according to one analyst estimate. Gomez said in October 2025 that the company could go public “soon” and that he was targeting profitability by 2029.

Those numbers remain small next to the American frontier labs, whose annualised revenues are measured in the billions and whose latest rounds are in the tens of billions. Cohere’s bet is that the enterprise and government segment values control, auditability and data residency enough to pay for a smaller, specialised vendor. The sharp growth in 2025 suggests that at least some of that market exists.

What is the Aleph Alpha merger and why does it matter?

The Aleph Alpha merger is Cohere’s attempt to become the default sovereign-AI vendor for both North America and Europe. Talks were disclosed on 10 April 2026 and the deal announced on 24 April; a definitive business combination agreement followed in September 2026, subject to regulatory approvals including from the German government and possibly the European Union, with closing expected later in 2026. Financial terms were not disclosed, but the Financial Times reported a combined valuation of about US$20bn, roughly C$27bn, with Cohere shareholders holding about 90%.

Aleph Alpha, based in Heidelberg, was Germany’s best-funded AI laboratory and is partly owned by Schwarz Group, the retailer behind Lidl and Kaufland. Schwarz has committed €500m, about C$800m, to lead a Series E round for the combined company, and Cohere’s models will be delivered through STACKIT, the Schwarz Digits sovereign cloud designed to keep data outside the reach of the US CLOUD Act. The combined group keeps the Cohere name and Gomez as chief executive; Aleph Alpha’s Ilhan Scheer becomes chief operating officer and co-founder Samuel Weinbach chief revenue officer. Global headquarters remain in Toronto, with a European headquarters in Berlin and the Heidelberg office retained for research.

The Canadian reaction was relief rather than alarm. Gomez had said in June 2025 that Cohere was not for sale and that an exit taking the company out of Canada would only happen if it failed. Pineau said in April 2026 that the company would always remain headquartered in Canada, and a source told BetaKit that it would stay majority Canadian-controlled with its intellectual property in Canada. AI minister Evan Solomon called the combination a big moment for Canadian AI. Patrick Pichette of Inovia, a Cohere investor, described it as the kind of strategic move that defines generational companies.

Strategically the deal bundles two things customers in regulated sectors want: a model developer and a sovereign deployment stack. The combined company targets finance, defence, energy, manufacturing, telecommunications and healthcare in Canada and Europe, and has partnerships with SAP, which launched an EU AI Cloud with Cohere, and with OpenText, which announced an agentic-AI partnership with Cohere in September 2026 that is expected to reach clients in early 2027. That OpenText link, examined in the OpenText and Waterloo story, is a rare example of two Canadian software companies combining data and models domestically.

What is the C$2bn Sovereign AI Compute Strategy?

The Sovereign AI Compute Strategy is the federal government’s answer to the fact that Canada had world-class researchers and almost no domestic capacity to train large models. Announced in Budget 2024 with C$2.4bn of AI measures and formalised later that year as a C$2bn, five-year programme, it has three parts: an AI Sovereign Compute Infrastructure Program of up to C$705m for public compute, an AI Compute Challenge of C$700m to finance commercial data centres, and a C$300m AI Compute Access Fund offering grants of C$100,000 to C$5m to Canadian firms, whose first call for proposals closed on 31 July 2025.

Cohere was the first beneficiary. On 5 December 2024 Ottawa announced up to C$240m toward a C$725m project in which Cohere and the American cloud provider CoreWeave would build a data centre in Canada using Nvidia’s latest GPUs, with Cohere as anchor tenant and other Canadian companies gaining access. Gomez said the facility would allow Cohere to train its next models in Canada for the first time. The location was not disclosed at launch, and the government said the centre would come online in 2025.

Budget 2025, tabled in November 2025, added C$925.6m over five years for large-scale, domestically controlled data centres for AI and cloud, of which about C$800m had been earmarked in 2024, and allowed the Canada Infrastructure Bank to invest in AI infrastructure. It also created an Office of Digital Transformation inside government and told Shared Services Canada to work with National Defence and the Communications Security Establishment on a made-in-Canada AI tool for government-wide use. The infrastructure financing side of this is covered in the Canada Infrastructure Bank story.

Private capital has followed. Bell has built a Bell AI Fabric business, with a 300 MW facility in Saskatchewan whose construction partners were named in May 2026, a sovereign-AI agreement with Cohere, Hypertec and BUZZ HPC in June 2026 and a Cisco partnership in September 2026. Telus has pursued a parallel sovereign-AI factory strategy. The telecom side of that build-out is discussed in the Rogers-Shaw and Canadian telecom story.

How does the government work with Cohere directly?

The government works with Cohere through a non-financial partnership to deploy AI across the public service. On 19 August 2025 Solomon, as Minister of Artificial Intelligence and Digital Innovation, and Joël Lightbound, Minister of Government Transformation, Public Works and Procurement, signed a memorandum of understanding with Cohere to explore AI deployment in government operations and to build Canada’s commercial capacity to use and export AI. Solomon’s office described it as early-stage collaboration with no financial component and no formal deadlines.

The context is Prime Minister Mark Carney’s promise to use AI to make the public service more efficient while capping its size and asking most departments to find spending cuts of up to 15%. Carney had met Gomez and UK Prime Minister Keir Starmer together in Ottawa on 15 June 2025, and a joint Canada-UK statement that day referenced cooperation on AI. The signal was that Cohere would be treated as a national champion and a preferred supplier, a stance that invites the obvious criticism of industrial policy picking winners.

Solomon’s ministry, created in May 2025, is itself new. His first year was spent consulting on a refreshed national strategy, which slipped from a promised 2025 launch into 2026, and signing memoranda with industry on infrastructure. The AI and Data Act, Canada’s attempt at comprehensive AI legislation, had lapsed in January 2025 when Parliament was prorogued, and the government has since preferred targeted measures on deepfakes and surveillance pricing to a single omnibus bill.

What does the June 2026 national AI strategy promise?

The national strategy, titled “AI for All” and released on 4 June 2026, promises a shift from research funding toward adoption, sovereignty and champions. It has six pillars: protecting Canadians and democracy, empowering Canadians, powering shared prosperity through adoption, building a sovereign AI foundation, scaling Canadian champions, and building trusted partnerships. Priority sectors are health and life sciences, energy and natural resources, transportation, agriculture, and manufacturing and robotics. More than 11,000 Canadians took part in the consultation.

The talent measures include AI training for one million post-secondary students and 3,000 educators, expansion of the Canada CIFAR AI Chairs to nearly 200 researchers, an expanded Global Talent Stream for immigration, and a target of up to 90,000 AI-related jobs for young Canadians, with a projection of more than 250,000 AI-relevant jobs by 2031. The sovereign-foundation pillar promises a public AI supercomputer, large-scale data centres with private capital crowded in, and a “Build-Partner-Buy” approach that admits Canada cannot build everything domestically.

Critics note that the sums are modest next to American and Gulf investment, that the strategy’s targets stretch to 2034, and that the Canadian Tech Growth Fund’s C$500m would buy a small stake in a single company at Cohere’s reported valuation. Supporters reply that the purpose is not to match the United States but to ensure that Canadian data, models and talent are not entirely dependent on it, a theme that runs through the wider Canada-US trade dispute.

Is Canada losing the AI talent it trained?

Canada is losing a meaningful share of its AI talent, and the evidence hardened in early 2026. A report by Mila and Bain & Company published in January 2026 found that Canada is home to about 10% of the world’s top AI researchers but attracts less than 2% of global AI venture investment, and that two-thirds of high-potential Canadian-led AI start-ups are headquartered outside the country. The same month the Canadian Press reported that the three national institutes had warned Ottawa the previous summer that the global war for AI talent had reached a “fever pitch”.

The pull is financial. American laboratories pay packages measured in millions, and even mid-level AI engineering roles in Silicon Valley pay several times a Toronto salary; recruiters have long noted that two Toronto engineers cost roughly one Bay Area hire. Remote work lets Canadians take American jobs without moving, and American firms have opened Canadian offices to hire locally. The retention record is mixed: about 92% of Vector-affiliated master’s graduates stayed in Ontario between 2018 and 2023, but the researchers who leave are disproportionately the senior ones who could found companies.

Montreal’s position is distinct. Mila is the largest academic deep-learning laboratory in the world, Bengio has remained in the city and in 2025 launched LawZero, a non-profit focused on safe AI, and Cohere opened a Montreal office in 2025. Mila Ventures, led by Stéphane Marceau, is exploring a venture fund to commercialise the institute’s research, which would address the complaint that Canadian discoveries are turned into companies elsewhere. The city’s broader start-up ecosystem is mapped in the Toronto entry of the City Startup Atlas and its Montreal counterpart.

How does Canada’s approach compare with other mid-sized countries?

Canada’s approach is closer to the European model of state-backed national champions than to the American market-led model. France has Mistral, Germany had Aleph Alpha until the Cohere deal, the United Kingdom has a sovereign-compute programme and a safety institute, and all three have tried to combine public compute with a domestic model developer. Canada’s twist is that its champion is also now Germany’s, which could make the Cohere group the sovereign vendor for two G7 countries at once.

The constraint is scale. Even after the Budget 2025 and June 2026 additions, Canadian federal AI spending is a few billion dollars over five years, while the largest American laboratories raise more than that in a single round. The Canadian pension funds, which manage well over C$2 trillion and are examined in the Maple Eight story, are the obvious domestic source of patient capital, and the Healthcare of Ontario Pension Plan’s entry into Cohere’s 2025 round was read as a signal that they are beginning to participate.

What can founders and investors learn from Cohere and Canada’s AI bet?

The first lesson is that a differentiated segment beats a head-on fight. Cohere could not outspend OpenAI on consumer models, so it chose enterprises and governments that value control over novelty. That choice looked slow in 2023 and looked prescient by 2026, when sovereignty became a purchasing criterion across Europe and Canada and the company’s revenue grew nearly fourfold in a year.

The second lesson is that staying home can be a strategy rather than a sacrifice. Cohere’s Canadian domicile unlocked C$240m of federal compute money, a government partnership, pension-fund investment and the political goodwill that allowed it to acquire Germany’s national champion with official blessing on both sides. A company that had moved to San Francisco in 2021 would have had none of those advantages, and the pattern of founders who stayed is examined in the Tobi Lütke story.

The third lesson is for policymakers. Research excellence without commercial capital produces talent for other countries; the Mila and Bain numbers prove it. The June 2026 strategy’s shift toward equity stakes, commercialisation funds and procurement is an overdue correction, but the amounts are small and the timelines long. Whether Toronto and Montreal can compete with Silicon Valley will be decided less by Ottawa’s spending than by whether Cohere, and the next five companies like it, choose to build their futures in Canada. The Shopify story shows it has been done once; Cohere is the attempt to do it again in the technology that matters most.

Frequently Asked Questions