Why do chips and computing power decide who leads in AI?

A language model learns by doing an enormous number of small multiplications, over and over, on the text it is trained on. Ordinary processors can do this, but far too slowly. The work runs instead on graphics processing units (GPUs) and similar chips designed to do thousands of calculations at once, packed by the thousands into data centres. The Government of Canada's own definition is a good one: AI compute is the computing capacity needed to process data, run algorithms and train machine-learning models.

More compute lets a lab train larger models on more data, test more ideas in a month, and answer millions of users without long waits. That is why talent and clever research, while essential, are not enough on their own. A country or company that cannot get the chips, or the electricity to run them, ends up renting capacity from someone who can.

The AI value chain, layer by layer

Canada's national AI strategy, AI for All, launched in June 2026, describes AI as a chain of layers that each depend on the one below. It is a useful map for anyone trying to understand who holds power in AI:

Read from top to bottom, the table explains the problem. A country can be excellent at research and applications and still depend on others for the layers that everything else needs. If those layers sit in another country, the strategy notes, its laws and decisions can reach Canadian data and services.

Why the chip layer is so hard to catch up on

Making advanced chips requires specialized fabrication plants, and Canada, like most countries, relies on global fabrication capacity for its GPUs. What it can do is design chips and build the facilities that use them. Data centres are faster to build, but they need land, cooling and a reliable supply of power. This is why AI announcements so often mention megawatts and gigawatts rather than software features: power capacity is now a direct measure of how much AI a country can run. The strategy itself cites estimates that Canadian businesses will need about 5.5 gigawatts of AI compute by 2030.

For a learner, the practical lesson is that the price, speed and availability of the AI tools you use all trace back to this physical layer. When compute is scarce, prices rise and limits tighten; when it grows, cheaper and faster models follow.

What is "sovereign AI", and what is Canada doing about it?

Sovereign AI means having enough AI infrastructure, data and talent under your own laws that critical services do not depend on decisions made elsewhere. It does not mean building everything at home. Canada describes its approach as "build, partner, buy": build the key capabilities domestically when possible, partner with trusted allies, and buy existing products when that makes more sense.

The money comes in two waves. First, the Canadian Sovereign AI Compute Strategy, funded by $2 billion over five years announced in Budget 2024, has three parts: up to $700 million through the AI Compute Challenge to attract private investment in Canadian AI data centres; up to $1 billion for public supercomputing, including a new Canadian-owned AI supercomputer, a smaller secure facility run by Shared Services Canada and the National Research Council, and up to $200 million to expand existing public systems in the short term; and up to $300 million for the AI Compute Access Fund, which helps businesses pay for compute. On April 15, 2026, the government opened applications to build and operate that supercomputer under the AI Sovereign Compute Infrastructure Program.

Second, AI for All adds targets: a world-leading public supercomputer by 2031, an extra $700 million so small and medium-sized businesses can buy affordable Canadian compute, and partnerships, still being finalized, that propose 850 megawatts of data-centre capacity by 2030 and could scale to 2.3 gigawatts. It also points to Canada's natural advantages, such as a cold climate that lowers cooling costs and clean electricity, and promises more work on chip design and fabrication.

Whether these plans close the gap will depend on delivery, power supply and how fast other countries move. If you run a Canadian business or research project that needs serious computing, the Compute Access Fund is the part of this strategy you can use directly.