Canada’s semiconductor position runs through packaging, photonics, and compound semiconductors. Every gap above that layer has an owner, and the owners are consolidating.
In May, at the Chips North conference in Ottawa, Minister of Artificial Intelligence and Digital Innovation Evan Solomon told BetaKit that Canada will not pursue a standalone national semiconductor strategy. Semiconductor support will live inside the broader AI strategy and the $5 billion Strategic Response Fund, he said, pointing to the fund’s $210 million commitment to semiconductor manufacturing in Bromont as evidence the government has “put a fair bit of money and a fair bit of strategy towards it already.” At the same conference, Liberal MP Jenna Sudds called developing a national semiconductor strategy a stand-out priority.
The minister was answering a specific demand. In December 2025, Canada’s Semiconductor Council, CMC Microsystems, the Information and Communications Technology Council, and ventureLAB jointly published a report warning that Canada is the only G7 country without a national semiconductor strategy, and that the gap threatens its competitiveness, its capacity to scale innovation, and its ability to exercise technological sovereignty over the digital economy. Researchers writing for the Canadian International Council put numbers to the stakes around the same time: Canada exported only $210 million worth of semiconductor devices in 2024, against Malaysia’s $12.3 billion from an economy half Canada’s size, and by 2024 more than half of Canadian companies in computer, electronic, and electrical equipment manufacturing were American-controlled.
Note the symmetry in that first figure. Canada’s entire annual semiconductor device exports equal what Ottawa committed to a single Bromont expansion. Both sides of the strategy debate can read that number as vindication, which is a sign the debate is being argued at the wrong altitude. Whether the policy is called a strategy matters less than what sits underneath it: what Canada can actually fabricate, who owns the capabilities it lacks, and how contested the ground is where it wants to stand. The packaging story established where system performance now gets decided. This piece maps the layer below.
The map starts with an honest inventory. Canada has no leading-edge logic fabrication. There is no facility on Canadian soil that could produce an Nvidia-class AI accelerator, and no credible path to building one this decade. The 2-nanometre and 3-nanometre nodes that power frontier AI live in a handful of fabs in Taiwan, South Korea, and the United States, behind capital requirements that start around US$20 billion per facility and an equipment monopoly, ASML’s extreme ultraviolet lithography, that a single Dutch company controls.
That inventory sounds like disqualification. It is closer to a map. The AI hardware stack has stratified into layers with very different economics, and the layers where Canada holds real capability are the ones gaining strategic weight fastest.
The Stack Stratified
Leading-edge logic fabrication is the most concentrated industrial capability on earth. TSMC, Samsung, and Intel are the only three companies operating at the frontier, and TSMC holds the overwhelming share of AI accelerator production. High-bandwidth memory belongs to three firms: SK Hynix, Samsung, and Micron. In their third-quarter 2025 earnings calls, all three reported HBM supply fully allocated through 2026.
The chokepoint has shifted below the wafer. TSMC’s CoWoS advanced packaging capacity is reported as sold out into 2026 despite growing at roughly 80 percent compounded annually, and Nvidia has reportedly booked more than half of TSMC’s CoWoS capacity for 2026 and 2027. Industry projections for early 2026 put roughly 45 percent of global advanced packaging capacity in Taiwan, 20 percent in South Korea, and 15 percent in China.
The clearest signal of where value has migrated came from Nvidia itself. Its US$5 billion investment in Intel, roughly a 4 percent equity stake in its fiercest data centre rival, was widely read as a packaging access deal. The world’s most valuable chip company paid billions to a competitor because its growth is now constrained by how many chips can be assembled, and not by how many it can design.
Fabrication has become plural. There is leading-edge logic, where three companies compete. There is memory, where three companies allocate. There is specialized and compound semiconductor fabrication, where dozens of niche players operate. And there is packaging, assembly, and test, where the bottleneck now sits. Canada’s position looks weak on the first two axes and genuinely interesting on the last two.
The Canadian Footprint, Facility by Facility
IBM Bromont in Quebec anchors everything. IBM describes the site as the largest outsourced semiconductor assembly and test facility in North America, producing more than 100,000 advanced flip-chip modules per week, with capabilities spanning chiplets, system-in-package, high-power thermal solutions, and co-packaged optics assembly and test. Read that capability list against the technology trends reshaping packaging: chiplet integration, thermal management, and co-packaged optics are precisely where the industry roadmaps point. Bromont is not a legacy facility doing yesterday’s back-end work. It sits on the current frontier of the layer that has become the bottleneck.
Ottawa has noticed. In November 2025, the federal government announced up to $210 million through the Strategic Response Fund toward a $662 million project to expand packaging and commercialization capabilities at IBM Bromont and the neighbouring MiQro Innovation Collaborative Centre, C2MI. The announcement language was explicitly sovereigntist: Canada must be its own best customer and cannot rely on importing technologies developed and produced elsewhere. The commitment builds on roughly $60 million in federal support announced in 2024, alongside $40 million from Quebec, and sits within IBM’s five-year, billion-dollar spending plan for the site. C2MI, a nonprofit R&D centre with more than $175 million in equipment, adds a commercialization bridge and has stated it will operate the world’s first open foundry for quantum superconducting chips.
Actual wafer fabrication in Canada is specialized rather than leading-edge, and the specializations matter. The National Research Council’s Canadian Photonics Fabrication Centre in Ottawa describes itself as Canada’s only end-to-end pure-play compound semiconductor wafer manufacturing facility, working in indium phosphide, gallium arsenide, and gallium nitride for fibre networks, defence, aerospace, medical imaging, data centres, and AI infrastructure. Compound semiconductors are the materials basis for the optical interconnect layer, the same co-packaged optics transition that packaging roadmaps treat as the next major integration challenge. Teledyne’s Canadian operations offer MEMS and microfabrication from 150-millimetre prototyping through 200-millimetre volume production, and received their own Strategic Innovation Fund support for a Bromont facility.
Around these facilities sits a wider ecosystem the federal government counts at more than 500 companies, over 100 design firms, 30 applied research labs, and five commercial facilities in compound semiconductors, MEMS, and advanced packaging.
The materials story is where the inventory gets complicated. Canada produces indium and germanium as by-products of zinc smelting and holds antimony reserves, a genuine asset given that China controls the large majority of global gallium and germanium refining and has restricted exports of both since 2023, including an outright ban on shipments to the United States announced in December 2024. But Ottawa’s own assessment concedes that Canada currently produces only a small amount of the critical mineral inputs required for compound semiconductors, and that few domestic entities can process these minerals to the ultra-high purity levels fabrication requires. Canada has the rocks. The refining and purification capacity that converts rocks into semiconductor-grade material largely sits elsewhere, and much of it sits in the one jurisdiction actively weaponizing that position.
Every Gap Has an Owner
Mapping the gaps against their current owners makes the sovereignty picture concrete.
Leading-edge logic belongs to TSMC in Taiwan, Samsung in South Korea, and Intel in the United States. High-bandwidth memory belongs to SK Hynix and Samsung in South Korea and Micron in the United States. EUV lithography belongs to ASML in the Netherlands, a literal monopoly. Advanced packaging at AI scale belongs to TSMC and the large outsourced assembly and test firms, ASE in Taiwan and Amkor in the United States, with the geographic weight in Taiwan and South Korea. High-end package substrates are dominated by Japanese suppliers. High-purity gallium and germanium processing is dominated by China.
Then there is the gap that receives the least policy attention and may matter most: ownership of Canadian capability itself.
The recent record reads like a controlled demolition. GaN Systems of Ottawa, arguably the country’s most promising compound semiconductor company, sold to Germany’s Infineon for US$830 million in 2023. Alphawave, founded in Toronto in 2017 and a leader in exactly the die-to-die connectivity IP that chiplet architectures depend on, completed its US$2.4 billion sale to Qualcomm in December 2025. Untether AI of Toronto, whose inference chips benchmarked at six times the energy efficiency of competitors in MLPerf testing, failed to raise capital in early 2025, saw its engineering team absorbed by AMD in a June acquihire that excluded the company’s technology, and filed for bankruptcy in October with liabilities of $128.6 million. Its statement of affairs valued the intellectual property that Canadian pension capital and venture funds had spent roughly US$150 million developing at one dollar. Nvidia acquired Toronto’s CentML. Tenstorrent, with deep Toronto roots, redomiciled to the United States in pursuit of funding.
The pattern is consistent. Canada develops semiconductor capability to the point of global relevance, and then the capability transfers to a foreign balance sheet, usually American, at the moment it becomes strategically valuable. Even the flagship domestic asset follows the pattern in softer form: IBM Bromont is a strategically located facility owned in Armonk, New York, and the $210 million Ottawa committed in November flows to an American multinational’s capital program. That can still be sound policy. Trusted-node status inside an allied supply chain has real value, and the alternative was watching the investment land in New York State instead. But it is industrial anchoring rather than ownership, and the distinction determines who captures the returns and who decides what happens to the facility in a crisis.
The Battleground Is Crowded and Heavily Armed
How competitive is the ground Canada wants to hold? Brutally.
The United States committed US$52.7 billion under the CHIPS Act, including a dedicated National Advanced Packaging Manufacturing Program funded at roughly US$3 billion, an amount for packaging alone that exceeds Canada’s entire announced semiconductor commitment by an order of magnitude. The European Union mobilized its own €43 billion Chips Act. Japan is funding Rapidus toward leading-edge logic and has drawn TSMC fabs to Kumamoto. India, Malaysia, Vietnam, and Singapore are all bidding aggressively for exactly the assembly, test, and packaging tier where Canada’s advantage sits, in several cases with lower labour costs and larger committed subsidies.
Trade policy compounds the exposure. Washington has attacked the CHIPS Act’s successor arrangements, threatened tariffs on semiconductor imports, and already demonstrated with autos its willingness to tax cross-border assembly flows. Bromont’s business model, in which chips fabricated elsewhere enter Canada for packaging and test before shipping onward, is structurally exposed to precisely that instrument. IBM has continued expanding anyway, which says something about the facility’s technical irreplaceability in the near term. It says nothing about where the next increment of capacity gets built if the tariff wall goes up.
And the honest counterargument to the whole thesis deserves stating plainly: specialization in the layers around the chip is a leverage position rather than an autonomy position. If the leading-edge logic, the memory, and the lithography are all owned elsewhere, Canada’s packaging and photonics capabilities give it a seat at the table and some resilience against disruption. They do not give it the ability to build an AI accelerator if the table disappears. Full-stack semiconductor sovereignty is unavailable to every country on earth except, arguably, none of them. Even the United States cannot currently produce a frontier AI system without Taiwanese fabrication, Korean memory, Dutch lithography, and Japanese materials.
Capital Follows Capacity, Again
Which returns this series to its recurring law. Capital flows to demonstrated capacity, and capacity compounds where capital lands. The November Bromont announcement exists because the facility spent five decades building capability worth anchoring. The Untether bankruptcy exists because capital did not arrive at the moment capability needed it, and the capacity dispersed to Santa Clara within weeks.
Canada’s realistic fabrication-layer position is now visible in outline. It is an allied, trusted node specializing in advanced packaging, co-packaged optics assembly and test, compound semiconductor and photonics fabrication, MEMS, and, if the processing gap gets closed, high-purity critical mineral inputs. That position is genuinely valuable and getting more valuable as packaging becomes the industry’s binding constraint.
Holding it requires three things the current policy mix does not yet guarantee. First, closing the purity gap between Canadian mineral production and semiconductor-grade processing, which is the single largest unclaimed opportunity in the domestic stack and one where China’s export weaponization has created an urgent allied market. Second, a retention answer for domestic semiconductor firms, because a policy that develops capability for foreign acquirers is a talent subsidy dressed up as industrial strategy, and the Canadian International Council’s proposed remedies, IP-domicile covenants and equity conditions on public funding, deserve a serious hearing. Third, sustained procurement and anchor demand, because Ottawa’s own Bromont announcement language, that Canada must be its own best customer, is a test the government has set for itself.
Solomon may even be right that the label matters less than the money. A strategy document without capital is a press release, and $210 million of committed capital without a strategy document is still $210 million. But the three conditions above are exactly the kind of cross-cutting commitments that embedded, program-by-program funding tends to miss, because no single fund owns the purity gap, the retention problem, or the procurement signal. The minister has declared the existing mix sufficient. The ownership ledger in this piece is how that claim gets audited.
The fabrication layer will never be where Canada wins the AI hardware contest outright. It is where Canada decides whether it holds leverage in that contest or watches it entirely from the stands. The footholds exist. The question is ownership.
