Last updated on May 21st, 2026 at 01:01 am
The Three-to-Five-Month Window, the Room Canada Is Not In, and the Bet Canada Has Not Made
By Jennifer Evans, Pattern Pulse AI / B2BNN / Tech Reset Canada
The capability window is now three to five months. The strategic window continues to extend.
Lee Klarich, chief technology officer of Palo Alto Networks, published a blog post on May 13, 2026, with that estimate. The post is titled “A New Era of Security: Frontier AI Defense.” Its core finding is that the timeline Anthropic announced six weeks earlier, when the company restricted release of Claude Mythos Preview through a coordination program called Project Glasswing, has compressed. Klarich’s specific framing: when Mythos first launched, Palo Alto predicted a six-month window before attackers gained access. The company now believes that timeline has accelerated significantly.
Two facts make the new window credible. The first is what Palo Alto found when it ran Mythos and OpenAI’s GPT-5.5-Cyber against its own products in weeks of testing. Seventy-five to eighty-five vulnerabilities. Three weeks of model-based analysis produced coverage equivalent to a year of manual penetration testing. The time from initial access to data exfiltration in AI-supported scenarios collapses to roughly twenty-five minutes. The second is what is already public about adversary use of general-purpose AI models in active operations. The GTG-1002 campaign, a China-linked intrusion operation disclosed by Anthropic, used Claude Code as its operational backbone. Eighty to ninety percent of operations executed autonomously. Reconnaissance, vulnerability discovery, exploitation, lateral movement, credential harvesting, data exfiltration. The capability the three-to-five-month window is being constructed against is not arriving. It has been arriving for over a year.
The window is closing on Canadian critical infrastructure that has not been protected by the coordination architecture being built to defend infrastructure elsewhere.
The argument of this paper is that Canada’s current AI posture is not a strategy. It is a sequence of announcements, procurements, subsidies, and infrastructure decisions made by different parts of the federal government on different timelines against different assumptions about what AI capability is and how Canada should relate to it. The announcements have been described as a strategy by the ministers who made them. They are not. A strategy identifies a binding approach and limitation, makes a concentrated bet against it, and aligns instruments toward the bet. The Canadian pattern of the past twelve months identifies no binding constraint, makes no concentrated bet, and aligns no instruments. The capability window is closing against this absence.
Two problems run through this analysis. They share a diagnosis and they require different responses on different timelines.
The first is defensive. The capability the three-to-five-month Mythos window is being constructed against is already operational in adversary hands and Canadian critical infrastructure is exposed to it now. The defensive response runs on the timeline of the window. The disclosure minimum, the CUSMA capability-sharing ask, the procurement architecture reform, and the procedural framework that lets federal policy operate at the speed of the capability environment are all defensive instruments. They have to be built immediately, because the window is immediate.
The second problem is generative. Canada has not built the sovereign AI capability that would let the country participate in coordination architecture being constructed around frontier models, one that would give Canadian capital somewhere to deploy at growth-stage scale, and that would produce the long-cycle institutional commitment a serious AI position requires. The generative response runs on a ten-to-twenty-year timeline. The bet on extraction AI, quantum, and sovereign edge AI is the generative response. It does not close the window. It is not designed to. It is designed to address the structural absence the window exposes but does not cause. The defensive instruments hold the perimeter while the generative bet builds what should have been built a decade ago. Both have to happen. Conflating them produces the procurement scatter pattern of the past two weeks, where infrastructure announcements, demand-side subsidies, and capability investment have been treated as interchangeable instruments addressing a single undefined problem.
The Room Canada Is Not In
Project Glasswing is the coordination architecture. Anthropic restricted Mythos Preview to roughly fifty organizations rather than releasing it publicly. The launch partners include Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorgan Chase, Linux Foundation, Microsoft, Nvidia, and Palo Alto Networks, plus approximately forty additional organizations primarily in US tech and critical software infrastructure. The model carries one hundred million dollars in usage credits, with an additional four million dollars in donations to open source security organizations. The European Union has been granted access to OpenAI’s GPT-5.5-Cyber through the EU AI Office under a separate framework. The Trump administration is reportedly considering additional government oversight of future frontier models.
The partner list is not a complete map of Canadian exclusion, because some Canadian operations of partner companies are presumably participating. The Linux Foundation operates globally. AWS, Microsoft, Google, and Apple have substantial Canadian presence. The question is not whether code paths used by Canadian customers are being hardened. The question is which governments and which critical infrastructure operators are in the room when the architecture is being designed, when the coordination is being directed, and when the policy framework is being negotiated.
The answer for Canada is that no Canadian government agency, no Canadian critical infrastructure operator, and no Canadian cybersecurity firm appears in the published partner list. The Communications Security Establishment is not a Project Glasswing partner of public record. The Canadian Security Intelligence Service is not. The Royal Canadian Mounted Police is not. Innovation, Science and Economic Development Canada is not. Public Services and Procurement Canada is not. No Canadian provincial cybersecurity authority is. No Canadian-headquartered cybersecurity firm is. Five Eyes intelligence-sharing arrangements give CSE channel access to frontier model threat intelligence that non-Five-Eyes Glasswing partners may not have. The substantive question is whether that downstream channel access translates to participation in architecture design, coordination direction, and policy framework negotiation that determine how the defensive coordination operates and whose critical infrastructure it covers.
Five Eyes access does not give Canadian critical infrastructure operators access to Mythos. It does not give Canadian cybersecurity firms participation in partner coordination. It does not give Canadian provincial authorities a seat in the room where critical infrastructure protection priorities are set. What Five Eyes channel access provides at the operational layer for industry, provincial deployment, and the broader critical infrastructure surface is not publicly specified. The defensive coordination architecture being built around frontier model capability has been built without published Canadian participation at the federal layer, the provincial layer, and the private sector layer.
This is not a complaint. It is the operational fact the next federal AI strategy has to address.
The Chinese Parity Picture
The window framing implies that capability reaches adversaries on a delay measured in months. The proliferation evidence says the delay is shorter than the framing acknowledges.
Zhipu AI released GLM-5.1 in April 2026. The model is a 744-billion-parameter mixture-of-experts architecture. It runs eight-hour autonomous execution loops. It handles approximately 1,700 autonomous steps in a single session. It is the first frontier model trained entirely on Huawei Ascend chips without any Nvidia hardware. It is released under MIT license, which means it is openly downloadable, modifiable, and self-hostable, with no enforceable mechanism for the developer to control downstream use. It scored 68.7 on the CyberGym benchmark across 1,507 real cyber tasks.
The UK AI Security Institute published its evaluation of GPT-5.5 on May 1, 2026, with comparison results across several frontier models. GPT-5.5 scored 71.4 percent on Expert-tier cyber tasks. Mythos Preview scored 68.6 percent. Opus 4.7 scored 48.6 percent. CyberGym and AISI Expert are not the same benchmark, so the numbers are not directly comparable, but the operational capability range is. A Chinese open-weight model is operating in the same offensive cyber range as Mythos Preview, under MIT license, on hardware that is not subject to US export controls.
Alignment removal on open-weight models is a documented, well-understood, and operationally simple procedure. An uncensored GLM-5.1 is not Mythos at frontier scale. The capabilities emerging at frontier scale today reach the near-frontier scale of open-weight models within one or two model generations. The proliferation curve is steeper than the original Anthropic six-month window assumed and steeper than the revised three-to-five-month Palo Alto window (shrinking daily) implies. The capability is globally distributed and operational now for many adversary use cases.
What Canada Did In the Window
Three days. Three federal announcements. None coordinated with the others. All anchored in Vancouver. None governed by a published federal AI strategy.
On May 11, the federal government announced a two billion dollar partnership with TELUS to build sovereign AI infrastructure across three British Columbia data centres. The compute stack inside the buildings is NVIDIA. The networking is NVIDIA. The systems software is NVIDIA. The buildings, the cooling, and the electricity are Canadian. The technology that produces the AI capability is American. The framework Paper 1 of this series introduced distinguishes hosting sovereignty from capability sovereignty. The TELUS announcement is hosting sovereignty marketed as capability sovereignty.
On May 12, Ontario Auditor General Shelley Spence released a report finding the province’s AI strategy lacks several key components. Three percent of Ontario Public Service staff had completed the responsible AI training. Sixty percent of AI sites accessed by OPS staff between April and August 2025 were flagged as unsafe by the government’s own Microsoft Defender. The Ministry had not blocked access to those sites. Sixty percent of approved medical AI scribes recorded a different drug than was prescribed in clinical testing. Nine of twenty fabricated information and made unsolicited treatment suggestions. The deployment layer at the provincial level operates without controls.
On May 13, AI Minister Evan Solomon announced sixty-six million dollars through the AI Compute Access Fund to subsidize compute access for forty-four Canadian small and medium-sized enterprises. The fund pays sixty-seven percent of eligible costs for Canadian cloud-based compute and fifty percent for non-Canadian. The subsidy structure routes most of the public money to AWS, Azure, and Google Cloud, the cloud platforms the funded companies are already using at the scale they need to operate. The minister confirmed on the record that the new BC data centres are not connected to the funded projects.
On the same day, CityNews Toronto reported that the federal government has spent more than eight hundred million dollars on AI contracts and licensing since 2023, in response to an order paper question from Conservative MP Jagsharan Singh Mahal. The figure includes a three hundred fifty million dollar contract with Dayforce to replace the Phoenix pay system. Dayforce is Ceridian rebranded, a payroll and human capital management platform acquired by Thoma Bravo on February 4, 2026, for $12.3 billion. The figure also includes a two hundred forty million dollar equity investment in Cohere. The Communications Security Establishment and the Canadian Security Intelligence Service declined to provide the requested information. The RCMP said it did not have the information available in a centralized database.
Five facts, three days. A foreign-vendor procurement marketed as sovereign infrastructure. A provincial deployment failure documented by audit. A demand-side subsidy that defaults to American hyperscalers. A payroll platform classified as AI. (I don’t pretend to understand the rationale or explanation for that reclassification.) And a national security disclosure refusal.
The three-to-five-month shrinking window is operating against this. The federal AI strategy that would coordinate any of it is nine months past its original timeline.
The Quarter That Documented What That Costs
On May 14, 2026, the Canadian Venture Capital and Private Equity Association published its Q1 report.
Nine hundred thirty-six million dollars in venture capital across one hundred four deals. Seventy-seven percent less capital than the previous quarter. Forty-one percent fewer deals. The smallest deal count for a quarter since 2017. The Globe and Mail reported the most striking number on the front of the business section: one growth-stage deal in the first quarter, worth approximately one million dollars. The growth-stage layer of Canadian venture capital, the layer that funds companies as they scale from product to revenue to global presence, deployed one million dollars in three months across the entire country.
The structural picture is sharper than the quarterly numbers. In 2025, sixty percent of all venture capital investment in Canada was led by US parties, the highest share since 2017. Two organizations, the Business Development Bank of Canada and Radical Ventures, accounted for seventy percent of total Canadian VC fundraising. Sixty-two other Canadian funds collectively raised one billion seven hundred million dollars across the year, an average of seventeen million dollars per fund. Andreessen Horowitz raised fifteen billion dollars across five new funds in the first nine days of January 2026, roughly ten times Canada’s entire 2025 venture capital output. One firm. Nine days.
The CVCA is pressing the federal government to deploy the seven hundred fifty million dollars committed in Budget 2025 for early-stage venture capital, with the recommendation that the money go to funds capable of leading large investment rounds. The framing in today’s reporting from the CVCA is that Canada lacks growth-stage capital. The framing is internally coherent. It also misdiagnoses the problem.
The reason Canadian growth-stage venture investment has collapsed is not that Canadian VCs lack capital to deploy. It is that there is no sovereign AI capability layer in Canada at scale that Canadian venture capital can deploy into in growth-stage rounds without the company moving to the United States for its capability infrastructure. The capital is following the capability. The capability is not here. The capital is going where the capability is.
Capital follows capability. This is the through-line. What we are hearing about now is that physical infrastructure assets including airports and ports are now for sale. In the absence of a sovereign capability layer to generate new wealth, the economic alternative defaults to the liquidation of existing physical assets. We are told that Canada is considered the safest bet for infrastructure investments in the world. But selling off assets, as we have already learned many times (the Wheat Board, 53% of Hydro One, Eagle Creek Renewable Energy at a loss among other examples) is not an investment option. It is a one-time divestiture. Public assets once sold are rarely deprivatized. Fire sales are not builds.
What Capital Following Capability Actually Means
“Let’s build,” Prime Minister Carney has said recently in statements and social media posts about AI data centres and more. Ok. But let’s be strategic about it, because the venture-based-building environment has changed. AI has changed it, and we need to be clear about where our actual competitive advantages lie.
When a Canadian AI startup raises a seed round, the round can be deployed in Canada because product development and early team building can happen here. When the same company tries to raise a growth-stage round to scale, the round requires alignment with a capability stack that the company can demonstrate operational integration with. In 2026, the capability stack at the model layer is OpenAI, Anthropic, Google, and Meta. At the compute layer it is AWS, Azure, Google Cloud, CoreWeave, and Nvidia. At the application layer it is the integrators and platforms aligned with the model and compute layers. None of those companies are Canadian. The closest thing to a Canadian capability layer participant is Cohere, which sits inside the triumvirate this series has documented as the Cohere-CoreWeave-Palantir arrangement that defines Canada’s current de facto AI architecture, and is itself increasingly aligned with US infrastructure.
A growth-stage Canadian AI company cannot demonstrate capability-layer integration from Canada because the capability layer is not in Canada. The investors that can lead a growth-stage round are sitting in Silicon Valley, in Boston, and in New York, where the capability layer is being built. The Canadian company that wants to raise growth-stage capital relocates or restructures so the investors can underwrite it inside the capability layer they understand and have access to. The pattern is documented across two decades of Canadian technology company outcomes. Canadian companies that succeed at scale routinely reincorporate in the United States to access the capital that scales them. The reincorporation is not a tax decision and not a regulatory escape. It is the operational expression of where capability sits.
The federal government has been treating capital flight as a question of capital supply. The seven hundred fifty million dollar Budget 2025 commitment is structured as additional capital to plug a gap. The diagnosis underneath the structure is that more capital will fix the problem. The diagnosis is wrong. The problem is not capital. The problem is that Canada has not built the capability layer that Canadian capital, or any capital, would deploy into at scale. The capital is leaving because the capability is not here to receive it.
The scattered subsidies of the AI Compute Access Fund, sixty-six million dollars across forty-four companies, are the procurement-side analog of the same misdiagnosis. The procurement architecture treats AI capability as something that can be subsidized into existence at the demand side, by giving companies money to buy compute from whoever has it. The subsidy goes to the American hyperscalers because they have the compute. The Canadian companies receiving the subsidy use it to deepen their operational dependence on the American hyperscalers. The procurement architecture treats AI capability as something that can be hosted into existence, by building data centres in Canada that run American chips, American networking, American systems software, and American models. The hosting goes to TELUS because TELUS can host. The capability stays American because the capability is in the stack TELUS is hosting.
Capability is built. It is not procured. It is not hosted. It is not subsidized into existence at the demand side. It is concentrated, anchored, and developed at the supply side, by labs and infrastructure operators that have the talent, the capital, the compute, and the institutional commitment to do it. The countries that are building capability now are building it now. The United States is building it through Anthropic, OpenAI, Google DeepMind, Meta, the chip stack at Nvidia and AMD, and the cloud infrastructure at AWS, Azure, and Google Cloud. China is building it through Zhipu, DeepSeek, Alibaba, Moonshot, ByteDance, and the Huawei Ascend chip stack. The European Union is building selective capability through Mistral, Aleph Alpha, and a handful of others under a coordinated EU AI Office framework that has just been granted access to OpenAI’s GPT-5.5-Cyber. The United Kingdom is building selective capability through DeepMind operations, ARIA funding, and the AI Security Institute that is now running the most credible public cyber evaluations in the world.
Canada is hosting. Canada is subsidizing. Canada has not built.
The Middle Has Collapsed
The venture market that Canadian institutional capacity was sized for does not exist anymore. Eighteen months ago it existed. Three years ago it existed. Five years ago it existed in robust form. It is not the market in which capital is being deployed in 2026.
The old market had a middle. A company could raise a seed round at one valuation, build a product, raise a Series A at a higher valuation, build a team, raise a Series B, build a market, and eventually become a company that operated at scale or got acquired by one that did. The capability gap between a starting company and an incumbent was closeable through execution. Talent was assemblable at startup wages. Compute was rentable at reasonable cost. Models were trainable on budgets that venture capital could fund. The middle of the market was where venture capital lived, and the middle was real.
The middle has collapsed. The market has bifurcated into two layers. At the top, the frontier capability layer is occupied by a small number of incumbents operating at scales that require billions of dollars in compute, hundreds of researchers, multi-year coordinated investment, and institutional commitments that are not available to a startup or a Series B company. Anthropic. OpenAI. Google DeepMind. Meta. Zhipu, DeepSeek, and Alibaba in China. The barrier to entry at this layer is not execution. It is scale. A well-funded Canadian AI company with one hundred million dollars in capital, fifty engineers, and a year of runway is not in the same market as these incumbents. It is in a different market entirely.
At the bottom, the application layer is occupied by individual operators and small teams using frontier tools to do work that previously required teams of dozens. A single developer with Claude Code can ship software that would have taken a Series A startup with twelve engineers. A two-person team can build and operate an AI-powered service that would have required a full company in the old market. The application layer is real and growing. It is also not where capability sovereignty lives.
The middle ground, where venture capital traditionally operated and where Canadian institutional capacity was concentrated, is the layer that is disappearing. There is nothing to fund at the old middle because the work that used to require middle-stage companies is either being done by individual operators with frontier tools at the bottom or being absorbed by frontier-scale incumbents at the top. The Canadian institutional answer to the old market structure was sized for the middle. Modest venture funds. SR&ED. Innovation tax incentives. Accelerator programs. Scattered subsidies. All of these instruments were designed to support a middle that produced companies through execution-stage growth. The middle is gone. The instruments persist. The outcomes they produce are smaller than they used to be because the market they were designed for is smaller than it used to be.
The CVCA Q1 report is the operational expression of this collapse. One growth-stage deal at one million dollars is not a temporary market weakness. It is a structural signal that the layer of the market the CVCA represents has shrunk to the point of effective non-existence. The CVCA’s recommended response, deploying the Budget 2025 seven hundred fifty million dollars to fund growth-stage rounds, is treating a structural problem as a capital supply problem. It is not. There is no longer a middle market of Canadian growth-stage AI companies large enough to absorb that capital productively at the scales the CVCA is proposing. The companies that would absorb it have already left or never started.
The strategic question for Canada is what to build given that the middle is gone. The old playbook produced the diffuse Canadian AI ecosystem. The new market structure makes the old playbook obsolete. A serious answer requires a different posture.
The Three Signals
Three signals, same direction.
The first signal is geopolitical. Project Glasswing excludes Canada from the defensive coordination architecture being built around frontier model capability. The exclusion is not because Canada is hostile or unreliable. It is because Canada is not building the capability that would make the exclusion costly to the partners. A country that hosts other countries’ capability does not have a seat in the room where the capability’s defensive coordination is negotiated. The seat goes to the participants.
The second signal is technical. The Chinese open-weight parity finding on GLM-5.1 demonstrates that the capability the Project Glasswing window was constructed to defend against is already globally distributed. Canada does not have the option of waiting for capability to mature elsewhere and procuring access. The capability has matured. The proliferation has happened. Canadian critical infrastructure is exposed now, not in three to five months, to capability that exists now and is operational now in adversary hands.
The third signal is financial. The CVCA Q1 2026 report documents the price signal of capability absence. One growth-stage deal worth one million dollars across the country in three months. Sixty percent US-led funding. The capital is voting on whether Canada is building capability. The vote is no. Meanwhile, the cost of frontier model tokens, and building on them, has finally begun to climb.
Three signals. Same direction. The federal government has been responding to each as if it were a separate problem with a separate solution. Glasswing exclusion would be addressed through diplomatic engagement. Chinese capability would be addressed through CSIS threat assessment. VC drop would be addressed through Budget 2025 capital allocation. The three responses do not connect because the federal government has not internalized that the three signals are one signal. Capital follows capability. Canada is not building capability. Everything else is downstream.
The Card Canada Set Down
On February 4, 2026, the United States and Mexico announced a sixty-day Critical Minerals Action Plan establishing a coordinated trade policy framework, including possible price floors and a path to a binding plurilateral agreement. The same day, the State Department’s 2026 Critical Minerals Ministerial brought together fifty-four countries and the European Commission. The United States signed eleven new bilateral critical minerals frameworks or memoranda of understanding with Argentina, the Cook Islands, Ecuador, Guinea, Morocco, Paraguay, Peru, the Philippines, the United Arab Emirates, the United Kingdom, and Uzbekistan, plus a separate $8.5 billion minerals agreement with Australia. The Ministerial launched the Forum on Resource Geostrategic Engagement and a private-sector grouping called Pax Silica with nine US partner countries. Canada attended. Canada did not sign.
Foreign Affairs Minister Anita Anand told The Globe and Mail that Canada would decide whether to join the proposed Agreement on Trade and Critical Minerals only as part of larger CUSMA discussions. Christopher Hernandez-Roy at the Center for Strategic and International Studies described Canada’s February position approvingly. Critical minerals, he said, are one of the aces in Canada’s hand against the United States, and the government had rightly not given them away ahead of CUSMA, as Mexico had already done.
Three months later the framing changed.
On April 29, 2026, US Trade Representative Jamieson Greer told a delegation of Canadian politicians, trade envoys, and businesspeople in Washington that Canada should not attempt to use its energy and natural resources as leverage in CUSMA negotiations. On May 1, 2026, in his first major interview with The Canadian Press since taking office, Prime Minister Mark Carney was asked about the leverage framing. He rejected it. “I reject that characterization of it’s leverage. It’s looking at, is it in our mutual interest to trade more. Where is it? And if it’s not there, we have other options.”
The February position was the realist position. Canada has abundance across the broadest range of critical minerals of any Western country. Canada has the full value chain. Canada has the low-emissions energy. Canada has the Five Eyes status. Canada has the ESG framework. The US AI build-out, the hardware layer it runs on, and the data centre infrastructure it consumes all require what Canada has at scales no other Western partner can supply. That is the definition of leverage. It is also the only asymmetric card Canada holds against the AI-stack dependency this series has been documenting across ten papers.
The Carney framing may be tactically sophisticated. It may preserve the option through the “other options” half of the quote rather than the “I reject leverage” half. The headline that goes into the CUSMA negotiating room on July 1 is the half the US Trade Representative asked for, given six days after he asked for it. The Prime Minister disavowed the framing of leverage publicly. The framing of mutual interest is what Canada will bring to the table.
This matters for the bet. Capability sovereignty in extraction AI is built on the back of capability sovereignty in extraction itself, which depends on Canada having and using the leverage that comes with being the only Western partner that can supply the inputs at scale. The minerals position is the leverage. The leverage is what makes the reciprocity ask at CUSMA Article 19.16 credible. The reciprocity ask is what gets Canada inclusion in the coordination architecture the Glasswing partners are building. Without the leverage framed as leverage, the reciprocity ask is a request. Requests get the answer the asker wanted to give.
The Choice Canada Has Not Made
Three options are available to Canada. Two of them are increasingly unavailable. The third has not been taken.
The first option is competing at the frontier across the full AI capability surface. This requires institutional and capital mobilization at the scale the United States and China are deploying. Anthropic raised over thirteen billion dollars in a single round in 2025. OpenAI’s compute commitments run into the tens of billions. The Chinese frontier labs operate under coordinated state and private capital arrangements that aggregate similarly large amounts of resource. Canada does not have institutional capacity to mobilize at this scale. The federal budget cannot absorb it. The Canadian venture and pension capital structure is not coordinated for it. The talent base is too small. This option is not available to Canada and will not be available to Canada inside any timeline relevant to the capability window.
The second option is accepting dependency at the frontier with mitigation at the policy layer. This is the realist position. Canada concedes capability sovereignty in the strict sense, accepts that frontier model layer and frontier compute will be foreign, and concentrates Canadian effort on the disclosure framework, the procurement architecture, the workforce framework, and the trade-policy position that determine the terms of the dependency. This option is available. It is also a strategic concession of the kind that countries with comparative advantages and existing assets do not have to make. Canada has comparative advantages. Canada has existing assets. The realist position is available but it is not the position Canada is forced into.
The third option is concentrated capability sovereignty in defined verticals where Canada has structural advantages, existing institutional capacity, and assets that can be scaled to globally competitive depth inside the vertical. This is the bet. The bet does not produce sovereignty across the full AI surface. It produces sovereignty in a layer where Canada can plausibly compete and where the capability built inside Canada matters to the global stack. The bet is hard. It requires concentration the diffuse Canadian funding environment has not delivered. It requires institutional and capital structures Canada has not built. It requires a strategic choice Canada has not made. It is the only option that produces capability sovereignty without requiring scale Canada cannot mobilize.
What the Bet Looks Like
The bet has two anchors. Extraction AI and quantum.
Extraction AI is the concentrated capability deployment in Canada’s existing economic strength. The structural case is strong on every dimension. Canada’s economy is resource-heavy. Mining, oil and gas, forestry, and agriculture account for a substantial share of GDP, employment, and export earnings. The companies operating in these sectors are domestic at scale, have decades of operational data, have institutional commitment to continuing operations in Canada, and have direct interest in operational AI that improves yield, safety, environmental compliance, and resource recovery. The Toronto Stock Exchange’s mining and energy listings are among the deepest in the world. The capital pools that fund these sectors are domestic, sophisticated, and structured for long-cycle investment. The institutional capacity to mobilize concentrated capital around extraction-aligned AI exists in Canada in a way that the institutional capacity to mobilize concentrated capital around frontier model labs does not.
The 2026 TSX Venture 50 makes the price signal at the public market layer explicit. Of the 51 top-performing companies on the TSX Venture Exchange in 2025, 48 are mining companies. The mining cohort accounts for a combined market capitalization of $19.9 billion, an average share price increase of 443 percent, and over $1.5 billion in equity raised across the year. The CVCA Q1 2026 report documents where capital is leaving. The TSX Venture 50 documents where capital is going. Both signals point in the same direction. Capital is concentrating where Canadian capability sits. The capital is voting on what Canada is good at. The vote on extraction is decisive. The vote on growth-stage AI is one million dollars.
The global market for industrial AI applied to extraction is large, growing, and not currently dominated by a single sovereign player. Australia, Chile, Brazil, South Africa, and Indonesia have analogous resource economies with analogous AI needs. A Canadian capability anchor in extraction AI has an export market that is real and large.
Canada’s critical minerals position is the strategic layer underneath all of this. Canada has abundant reserves across the broadest range of critical minerals of any Western country. Lithium, graphite, nickel, cobalt, copper, rare earth elements, gallium, germanium, indium, vanadium, uranium, and potash. Canada has the full value chain inside its jurisdiction, from mining to processing to refining to some downstream manufacturing. The International Energy Agency notes that this combination “sets Canada apart from many other countries, which typically have only part of the critical minerals value chain within their jurisdictions.” Canada has abundant low-emissions energy through hydroelectric and nuclear capacity to run processing infrastructure at scale and at acceptable carbon profile. Canada has a stable regulatory framework with established ESG credibility. Canada has free trade agreements with all G7 nations. Canada is the only Five Eyes member in this position.
China controls roughly seventy percent of global refining across critical minerals, ninety percent of rare earth processing, ninety-eight percent of gallium, and ninety percent of battery-grade graphite. China has been weaponizing this dominance through escalating export controls since 2023, with new restrictions on rare earths, gallium, germanium, graphite, antimony, and tungsten. China cannot be the Western partner because China is the position the West is diversifying away from. Australia has minerals but lacks downstream processing depth. Indonesia has nickel but lacks regulatory and ESG credibility, and is increasingly aligned with Chinese capital. Chile has lithium and copper but limited diversity across the broader critical mineral surface. Canada is the only country the West can build a critical minerals-to-AI infrastructure pipeline with on Western terms. The G7 Critical Minerals Production Alliance, launched during Canada’s 2025 G7 Presidency, has mobilized eighteen and a half billion dollars in Canadian mineral projects with twelve allied nations. The infrastructure for the bet is partially built. The AI capability layer that would sit on top of it is not.
Quantum is the second anchor. Quantum belongs in the bet because it is the one adjacent frontier where Canada still has institutional depth before the market structure fully hardens.
It’s the riskier, bigger bet, it’s where the puck is going. We have a staked out position here that we can continue to build on in much the same way that we were very early to AI infrastructure but have largely squandered that early competitive advantage. There are obvious pre-AI success stories like Shopify, and the AI-era version is Waabi, the Toronto physical AI and autonomous trucking company founded by Vector Institute co-founder Raquel Urtasun. Waabi raised $750 million USD in January 2026 in the largest fundraise in Canadian tech history, co-led by Khosla Ventures and G2 Venture Partners. But the deployment is in Texas. The capital is American. These are not Canadian capacity, capability, and industry-building stories.
Canada has Xanadu, which went public through reverse takeover in March 2026 and is one of the global leaders in photonic quantum computing. D-Wave in British Columbia has been operating in quantum annealing since 1999. The University of Waterloo’s Institute for Quantum Computing and the Perimeter Institute have produced multiple cohorts of quantum researchers. The federal National Quantum Strategy committed three hundred sixty million dollars in 2023, modest relative to the scale required but more concentrated than the AI investment pattern. Quantum capital requirements are higher than AI application development but lower than frontier AI model training. Development timelines are longer, which favours sustained institutional commitment over short capital cycles.
The two-ended structure of the bet reinforces itself. Extraction AI is concentrated capability deployment in an existing economic strength, with current revenue, current Canadian institutional capacity, and a global export market in adjacent resource economies. Quantum is concentrated capability investment in a frontier where Canada has demonstrated research and early commercial capacity, lower current revenue, longer development timelines, and asymmetric future value. The combination gives Canada exposure to current value through extraction AI and future value through quantum, while concentrating both bets rather than scattering. The two anchors share infrastructure requirements at the talent layer, the compute layer, and the institutional layer, which means the bets reinforce each other rather than competing for resources.
The customer architecture for the bet matters as much as the capability architecture. Canadian-built extraction AI licensed to the global majors (BHP, Rio Tinto, Glencore, Vale, Anglo American) is capability export, not capability sovereignty. The licensing architecture has to be designed to anchor the bet in Canadian operators first. Teck, Cameco, Agnico Eagle, Barrick, Kinross, Nutrien, Suncor, and Cenovus are the primary customer base. The secondary export market is allied resource economies on terms that preserve the capability stack in Canadian hands: Australia, Chile, Brazil, and other jurisdictions inside the G7 Critical Minerals Production Alliance. The licensing architecture excludes non-allied majors and structures any export deployment to keep model development, training data sovereignty, and operational integration domestically held. The frontier labs have not built the domain-specific extraction stack, the operational integration with mineral processing, or the connection to physical resource flows that make extraction AI work. That is the structural opening Canada can occupy. Canadian licensing architecture has to defend it.
The Capability Layer Underneath Both Anchors
Sovereign edge AI sits beneath both anchors as the capability layer Canadian institutional geography forces the country to develop. The frontier labs operate cloud-first because per-token billing requires continuous connectivity. The customers Canadian geography exposes the country to are largely air-gapped, edge-deployed, intermittent-connectivity, or operating under data residency constraints that prohibit cloud AI by default. Remote mines, Arctic operations, naval and military deployments, the fifty-nine Ontario local distribution companies, North American water utilities, port authorities, regional rail operators, sub-transmission operators, pipeline midstream, hospitals operating under provincial data residency rules, and remote First Nations communities are non-customers of frontier cloud AI by structural design rather than procurement choice. Canada has direct exposure across all of these surfaces. The extraction AI anchor deploys to these environments. The Indigenous data sovereignty framework requires this architecture. The defensive coordination layer that Project Glasswing covers for hyperscaler-aligned infrastructure does not cover the long tail. Sovereign edge AI is the connective capability underneath the two anchors and the surface where Canadian institutional geography produces structural advantage the frontier labs cannot economically erode.
What the Bet Is Not
The bet is not “give up on AI and just dig things up.” That framing is wrong and the paper is not making it. Canada is making an AI sovereignty bet. The bet is on the verticals where Canada has structural advantages, existing institutional capacity, and existing assets at scale. The bet produces global AI sovereignty in a layer that matters. The application of AI to extraction is not a small market. It is a global multi-hundred-billion-dollar market over the next decade. Canadian sovereignty in that market is sovereignty in a layer that the rest of the global stack depends on. The hardware that frontier AI runs on requires the minerals Canada extracts. The compute infrastructure that frontier AI consumes requires the materials Canada processes. The bet is not at the periphery of AI capability. It is at the layer underneath it.
The bet is not yet Cohere. Cohere is a Canadian AI company. It is making money. It has government equity participation. It has produced Aya, an open-weight multilingual model covering more languages than the US frontier labs explicitly target. Aya is a real artifact and the closest thing in Cohere’s portfolio to a sovereignty-relevant capability play. Its language coverage skews to commercially viable global languages rather than the Indigenous Canadian languages a sovereign multilingual framework would require. Its license is research-restricted (CC-BY-NC) (update May 21: now Apache 2.0 with new release Command A+, which changes the picture) and commercial productization at Cohere routes through enterprise channels. Frontier models offer comparable multilingual capability for the languages Aya covers, which means Aya’s selection over a frontier alternative depends on enterprise procurement preference rather than capability differentiation. Aya is where Cohere is closest to the bet. It is not the bet. Cohere’s primary commercial trajectory is enterprise integration, a valid business and a valuable one. Treating Cohere as Canadian frontier sovereignty conflates a successful enterprise AI company with a national capability layer. Canada needs both. Canada has been treating Cohere as if it were both. The bet does not displace Cohere. It puts Cohere in its actual category and concentrates sovereignty investment where Canada can plausibly produce global capability depth. But it is where the puck is, not yet where it is going.
What Cohere has built, more clearly with each subsequent deal, is sovereign-AI sales and diplomatic assembly capability. The Aleph Alpha merger in April 2026, the Saab GlobalEye MoU in March 2026, and the May 2026 Indra-Cohere MoU under the Canada-Spain sovereign-AI cooperation framework show Cohere operating effectively as a commercial vehicle for Canadian-anchored European AI cooperation. This is doing more work for Canada than the framing of Cohere as a domestic frontier lab suggested it would. It is also not the same work as building sovereign capability at the model layer. While Cohere is not yet the bet, it is becoming clearer that Cohere is a useful adjacent asset whose actual strength has become visible through the partnership architecture being assembled around it.
The bet is not the absence of a public sector AI framework. The capability bet and the policy framework are separate questions. The disclosure minimum that the next paper in this series develops, the procurement reform, the workforce framework, the trade-policy position at CUSMA, all of these still have to be built whether Canada makes the capability bet or not. The capability bet does not solve the procurement scatter or the disclosure refusals or the standing offer architecture. It solves the capability question at the layer where Canada can plausibly solve it, which makes the other problems negotiable from a position of having something to negotiate with.
The quantum bet carries architecture risk. Photonic (Xanadu) and annealing (D-Wave) are not the dominant paradigms in the current fault-tolerant race. Gate-based superconducting (IBM, Google, IQM), trapped ion (Quantinuum, IonQ), and neutral atom (QuEra, Atom Computing) are absorbing the largest share of capital and talent. PsiQuantum, IBM, Google, and the Chinese state are deploying scale Canada cannot match across the full quantum surface. The bet does not require winning the dominant architecture. It requires carving out durable capability in photonic and annealing applications where Canadian research depth and early commercial capacity give the bet a defensible position. The IQC and Perimeter talent base is the asset that survives architecture choice, which is why the bet anchors in the institutional layer as much as in the companies. The bet is concentrated capability in a domain where Canada has demonstrated research and early commercial capacity, where development timelines favour sustained institutional commitment, and where asymmetric future value justifies the investment at lower probability of full architecture dominance.
What the Bet Requires
Element AI is the cautionary example. Canada built frontier-adjacent AI capability in the late 2010s. The company assembled serious research talent, attracted global recognition, and was positioned as Canada’s domestic frontier AI champion. In 2020, Element AI was sold to ServiceNow for a fraction of the value comparable assets were commanding in the United States. The decision to sell was rational at the company level. The capital that would have allowed Element AI to scale into a Canadian frontier company was not available in Canada at the scale the company required. The capital that was available was the acquisition offer from a US incumbent. The pattern is not unique to Element AI. It is the pattern across two decades of Canadian technology outcomes. Canada builds. Canada loses what it builds because the capital and institutional structure to hold it does not exist.
The bet on extraction AI and quantum requires the capital and institutional structure to hold what Canada builds. This is the part of the bet that has not been built. Three components are required.
A concentrated capital structure. The bet cannot be funded through the diffuse instruments Canadian innovation policy has historically used. SR&ED will not produce it. The Canada Growth Fund’s general mandate will not produce it. The Strategic Innovation Fund’s project-by-project structure will not produce it. The bet requires sovereign-scale capital concentration of the kind the Norwegian sovereign wealth fund, the Singapore Temasek model, or coordinated public-private commitments at the scale of major infrastructure programs deliver.
The Canada Strong Fund, announced by Prime Minister Carney on April 27, 2026, is the structural instrument that comes closest to what the bet requires. The fund is seeded with twenty-five billion dollars over three years, structured as an arms-length Crown corporation with an independent board, designed as an equity co-investment vehicle distinct from the Canada Infrastructure Bank’s loan focus, and mandated to invest alongside private capital in projects of national interest. The targeted sectors as announced are infrastructure, energy, critical minerals, and advanced manufacturing. The fund’s existence validates the diagnostic this paper is built on. Canada needs sovereign-scale capital concentration. The government has accepted that needs and stood up an instrument. The instrument is real. The mandate is not yet what the bet requires.
Three gaps separate the Canada Strong Fund as announced from the capital structure the bet requires. The mandate as described targets projects and assets, not capability. Infrastructure, energy, and advanced manufacturing are project categories. AI capability is not a project category. Critical minerals is in scope, which is the closest the announced mandate comes to the bet, but extraction AI as the capability layer on top of the minerals stack is not specified. John Ruffolo, who has publicly advocated for Canada to create a sovereign wealth fund, observed on announcement day that the mandate as articulated does not mention risk-based investments, venture, growth, or innovation. The capability bet sits exactly in that absent category.
The fund is debt-funded rather than surplus-funded. Norway’s Government Pension Fund Global, the global benchmark, is funded by oil and gas surplus revenues. The Singapore Temasek model is funded by accumulated state capital. The Canada Strong Fund is funded by adding to the federal deficit. This is a sustainability question the fund’s design has to resolve. The capability bet operating inside a sovereign vehicle that is itself debt-funded compounds the political risk of the bet across the ten-to-twenty-year timeline the bet requires. The fund’s transition office, governance structure, and asset recycling mechanisms are the instruments through which the debt-funded character can be converted into a durable capital base. Whether that conversion happens determines whether the fund survives a political cycle change.
The governance design is still open. The CEO has not been named. The board has not been constituted. The investment mandate has not been finalized. The retail investment product has not been designed. The pension funds the fund will need to attract as co-investors were not consulted in advance and are awaiting detail before committing. Indigenous equity participation is identified in legal commentary as a design question the fund will have to resolve, which intersects directly with the Indigenous data sovereignty framework the bet on extraction AI requires. The design window is now. The bet can be designed into the fund’s mandate during the transition office consultation period, or the fund can be designed around the bet and the capability gap persists inside a sovereign vehicle that should have been the answer to it.
The political commitment to sovereign capital concentration has been made. The vehicle has been announced. The mandate that determines whether the vehicle funds the capability bet has not been written. This is the inflection point. The bet on extraction AI, quantum, and sovereign edge AI is the capability layer the Canada Strong Fund can fund if the mandate is designed for it. Without that mandate the fund will deploy into infrastructure and energy projects that produce hosting sovereignty without capability sovereignty, replicating the TELUS pattern at sovereign-fund scale. With the mandate the fund is the capital structure the bet requires. The window to design the mandate is the months between now and the fund’s operational launch.
An institutional commitment that survives political cycles. The bet operates on a ten to twenty year timeline. The federal political cycle operates on four. Provincial cycles are similar. The bet requires institutional structures that are insulated from electoral turnover. Crown corporation. Sovereign investment vehicle. Multi-party legislated mandate. Federal-provincial coordination instrument. The specific form is negotiable. The requirement that some form exists is not.
A governance framework that includes Indigenous participation, environmental commitments, and benefit-sharing structures designed into the bet rather than negotiated around it. Resource extraction in Canada operates under treaty obligations, constitutional duties, and contested jurisdictional arrangements. The bet on extraction AI cannot be made around those constraints. It has to be made inside them.
The framing of Indigenous participation as constraint is incomplete. Indigenous data sovereignty is itself a Canadian sovereign asset the US frontier labs structurally cannot replicate. The OCAP principles (Ownership, Control, Access, Possession) developed by First Nations communities, the First Nations Information Governance Centre, and the broader Indigenous Data Sovereignty governance stack constitute an alternative data-governance regime built on consent, community ownership, and jurisdictional authority. The US labs have chosen the opposite architecture: centralized training on aggregated data, opaque consent, and global rather than community-specific governance. Operating inside an OCAP-compliant framework would require the labs to become something they have explicitly chosen not to be. Canada has an existing operational framework the labs cannot occupy. Extraction AI deployed across Indigenous territories, cultural heritage applications, community-controlled health and language infrastructure, and environmental monitoring on traditional lands are capability layers the frontier labs are structurally locked out of. The bet that succeeds treats this as leverage and designs the capability around it. The bet that fails treats it as compliance overhead and reproduces two decades of failed extractive consultation.
Two structural questions remain that the bet has to resolve in its design. Whether the bet sits at the federal layer, the provincial layer, the institutional layer at Vector, Mila, Amii, IQC, and Perimeter, or some combination. What the capital structure looks like in operational form. Neither question is resolved in this paper. Both have to be resolved before the bet can be made. A subsequent paper in this series will develop the capital structure and governance design in detail.
The Policy Framework Around the Bet
The capability bet does not replace the policy work the series has been arguing for. It anchors it.
A published Canadian national security framework for AI deployment in federal departments and agencies, including the public sector AI disclosure minimum the next paper develops in detail. The framework has to be published before the CUSMA review opens on July 1. Without it Canada cannot invoke CUSMA Article 32.2 to defend any of the procurement decisions it has made. The disclosure minimum is the floor below which federal and provincial AI deployment cannot operate.
A negotiating ask for capability-sharing reciprocity at CUSMA. The United States has built a defensive coordination architecture with strategic allies. The European Union has been granted access on a separate framework. Canada has to negotiate inclusion on terms that do not require Canadian critical infrastructure to be downstream of US-controlled coordination without independent audit. The legal anchor for the ask is the Article 19.16 cybersecurity cooperation obligation. The negotiating leverage for the ask is the critical minerals position. Canada is the only Western partner that can supply the inputs to the hardware layer at the scale the US AI build-out requires. The reciprocity ask is not a request. It is an exchange.
A procurement architecture that distinguishes capability-building procurement from operational tool procurement from infrastructure hosting from equity investment, with separate governance frameworks for each. The scatter pattern of the past two weeks, the Dayforce category collapse, the AI Compute Access Fund subsidy structure, and the standing offer architecture are all symptoms of the missing taxonomy. The procurement reform follows from the capability bet because the bet creates the categories the taxonomy has to recognize.
A procedural framework that allows federal policy to operate at the speed of the capability environment. The federal AI strategy has been delayed for nine months while the capability environment has compressed by an order of magnitude. The procedural problem is itself a sovereignty problem. A country that cannot move at the speed of capability cannot defend its sovereignty in the capability environment. The procedural reform required to operate at the necessary tempo is the precondition for everything else.
The Argument
This series has argued across ten papers that Canadian AI sovereignty is a question of architecture. The architecture has been described at the procurement layer, the disclosure layer, the workforce layer, the constitutional layer, and the trade-policy layer. This paper adds the capability layer underneath them. The capability layer is the binding constraint.
The procurement-strategy gap, the disclosure refusals, the standing offer architecture, the workforce framework absence, and the trade-policy improvisation are all symptoms of the capability gap. The capability gap is the disease. Canada has not built sovereign AI capability. The procurement decisions, the spending announcements, and the regulatory frameworks that have been built or delayed in the past year have been built or delayed inside an assumption that capability would arrive through procurement, hosting, or subsidy. None of those produce capability. The capability has not arrived. The window in which the defensive posture has to be set is now three to five months. The window in which the capability bet has to be chosen is also now, because the capability it would produce belongs to the next two decades.
The bet is the only option that produces capability sovereignty inside the constraints Canada actually faces. Extraction AI anchored in Canada’s critical minerals position, full value chain integration, low-emissions energy, and the strategic position of being the only viable Western partner for resource-to-AI infrastructure. Quantum anchored in Canada’s existing research talent and the commercial capacity at Xanadu and D-Wave. Concentrated capital structure. Long-cycle institutional commitment. Indigenous participation, environmental performance, and benefit-sharing designed into the bet rather than negotiated around it. The policy framework around the bet that the series has been describing across ten papers.
The bet is the choice Canada has not made. The diffuse pattern of the last ten years has produced the diffuse outcomes documented in this paper and the ten before it. The market structure that the diffuse pattern was designed for no longer exists. The capability window in which a different choice could produce different outcomes is now three to five months. The choice is available. The cost of not making it is documented in the CVCA Q1 report, in the Glasswing partner list, in the Mythos timeline collapse, and in the $800 million figure that does not add up to a strategy.
Capital follows capability. The capital is leaving. The capability is not here. The bet that produces capability is available. Canada has not yet made it. Other countries have done so successfully. Is the choice as binary as it appears, between selling off infrastructure and making risky bets? It has been elsewhere. The window in which the bet can be made and produce the capability the next two decades require is now.
Jennifer Evans is the founder of Pattern Pulse AI, publisher of B2BNN, and co-founder of Tech Reset Canada. The Canadian AI Sovereignty series is published via Zenodo and ResearchGate at PatternPulse.ai.
References
The Three-to-Five-Month Window
Klarich, L. (2026, May 13). A New Era of Security: Frontier AI Defense. Palo Alto Networks. https://www.paloaltonetworks.com/blog/2026/05/frontier-ai-defense/
CNBC. (2026, May 13). Palo Alto Networks CTO on AI cyberattacks, Mythos, and GPT capability window.
UK AI Security Institute. (2026, May 1). Expert-tier cyber evaluation results: GPT-5.5, Claude Mythos Preview, Claude Opus 4.7.
Adversary Capability Already Operational
Anthropic. (2025, November 14). Disrupting the first reported AI-orchestrated cyber espionage campaign. https://www.anthropic.com/news/disrupting-AI-espionage
Anthropic. (2025, November). Disrupting the first reported AI-orchestrated cyber espionage campaign [Full report]. https://assets.anthropic.com/m/ec212e6566a0d47/original/Disrupting-the-first-reported-AI-orchestrated-cyber-espionage-campaign.pdf
Chinese Open-Weight Parity
Zhipu AI. (2026, April). GLM-5.1 release announcement and technical documentation. MIT License.
CyberGym Benchmark. (2026). GLM-5.1 evaluation results across 1,507 real cyber tasks.
Project Glasswing
Anthropic. (2026, April 7). Claude Mythos Preview announcement and Project Glasswing partner list.
European Commission. (2026). EU AI Office access framework for OpenAI GPT-5.5-Cyber.
Federal AI Announcements, May 11-13, 2026
Government of Canada and TELUS Communications. (2026, May 11). $2 billion sovereign AI infrastructure partnership announcement, Vancouver.
Office of the Auditor General of Ontario. (2026, May 12). Value-for-money audit: Artificial Intelligence in the Ontario Public Service. Auditor General Shelley Spence.
Innovation, Science and Economic Development Canada. (2026, May 13). AI Compute Access Fund announcement, $66 million across 44 companies.
Federal AI Procurement Disclosure
CityNews Toronto. (2026, May 13). Federal government spent more than $800 million on AI contracts and licensing over three years. Response to order paper question from MP Jagsharan Singh Mahal.
Canadian Venture Capital Q1 2026
Hemmadi, M., Balakrishnan, A., & Smith, J. (2026, May 14). Venture investing sinks to 9-year low as growth-stage deals dry up. The Logic. https://thelogic.co/news/cvca-q1-growth-stage-deals-decline/
Globe and Mail. (2026, May 14). Venture capital investment in Canadian growth-stage firms fell to near zero in latest quarter, report says. https://www.theglobeandmail.com/business/article-venture-capital-investment-in-canadian-growth-stage-firms-fell-to-near/
BetaKit. (2026, May 14). Canadian VC sees lowest quarterly deal count in nearly a decade. https://betakit.com/canadian-vc-sees-lowest-quarterly-deal-count-in-nearly-a-decade/
Canadian Venture Capital and Private Equity Association. (2026). Q1 2026 VC and PE Canadian Market Overview.
TSX Venture Exchange. (2026, February 18). Mining Sector Leads Unprecedented Market Cap Surge on the 2026 TSX Venture 50. https://investors.tmx.com/English/News–Events/news/news-details/2026/Mining-Sector-Leads-Unprecedented-Market-Cap-Surge-on-the-2026-TSX-Venture-50TM/default.aspx
Canadian Press / BNN Bloomberg. (2026, February 18). Mining stocks dominate TSX Venture 50 list amid rotation into resource sector. https://www.bnnbloomberg.ca/markets/tsx/2026/02/18/mining-stocks-dominate-tsx-venture-50-list-amid-rotation-into-resource-sector/
Critical Minerals and Resource AI
International Energy Agency. Canada is set to play a leading role in supplying the world with responsibly produced critical minerals. https://www.iea.org/commentaries/canada-is-set-to-play-a-leading-role-in-supplying-the-world-with-responsibly-produced-critical-minerals
Canadian Mining & Energy. (2026, March 11). Canada’s Critical Minerals: The Resource Superpower. https://www.miningandenergy.ca/read/canadas-critical-minerals-the-resource-superpower
Corporate Knights. (2026, May). Canada plays its hand on critical minerals with a high-stakes investment drive. https://corporateknights.com/mining/canada-plays-its-hand-on-critical-minerals-with-a-high-stakes-investment-drive/
Royal Bank of Canada. (2025, May 27). The New Great Game: How the race for critical minerals is shaping tech supremacy. https://www.rbc.com/en/thought-leadership/the-trade-hub/the-new-great-game-how-the-race-for-critical-minerals-is-shaping-tech-supremacy/
Canadian Mining Journal. (2026, March 12). The AI boom beneath our feet: How data centres are rewriting mineral demand. https://www.canadianminingjournal.com/featured-article/the-ai-boom-beneath-our-feet-how-data-centres-are-rewriting-mineral-demand/
Natural Resources Canada. The Canadian Critical Minerals Strategy. https://www.canada.ca/en/campaign/critical-minerals-in-canada/canadian-critical-minerals-strategy.html
Government of Canada. (2025-2026). G7 Critical Minerals Production Alliance, $18.5 billion in mobilized projects with 12 allied nations.
Element AI
BetaKit. (2021, January 15). ServiceNow closes $230 million USD acquisition of Montreal’s Element AI. https://betakit.com/servicenow-closes-230-million-usd-acquisition-of-montreals-element-ai/
Globe and Mail. (2020, December 21). Element AI sold for $230-million as founders saw value mostly wiped out, document reveals. https://www.theglobeandmail.com/business/article-element-ai-sold-for-230-million-as-founders-saw-value-wiped-out/
Quantum
Xanadu Quantum Technologies. (2026, March). Initial public offering announcement, Nasdaq and Toronto Stock Exchange.
Government of Canada. (2023). National Quantum Strategy, $360 million commitment.
Previous Papers in the Series
Evans, J. (2025-2026). Canadian AI Sovereignty Papers 1 through 10. Pattern Pulse AI / B2BNN / Zenodo / ResearchGate.