Mark Carney is worried. The former Governor of the Bank of England and current Canadian political heavy hitter recently warned that tightening U.S. restrictions on AI technologies underscore the extreme risks of digital dependence. The prevailing narrative among Ottawa’s policy elite is clear: Canada is at the mercy of American tech hegemony, and Washington’s protectionism will starve foreign markets of computational oxygen.
This panic is entirely misplaced. You might also find this connected article useful: Inside the Military AI Crisis Nobody is Talking About.
The lazy consensus among bureaucrats and legacy executives is that national sovereignty in the 2020s requires owning the entire tech stack, from the silicon up to the foundational model. They look at U.S. export controls and see a threat.
They are looking at the problem completely backward. As reported in detailed coverage by MIT Technology Review, the results are notable.
U.S. protectionism is not a threat to foreign tech ecosystems. It is a massive, subsidized gift to them. By ring-fencing American infrastructure and restricting the flow of specific hardware and raw model weights, Washington is doing something domestic venture capitalists could never achieve. It is forcibly breaking the addiction to commoditized American platforms and forcing local ecosystems to build actual, defensible value.
Distinguished policymakers love to sound the alarm on "dependence" because it justifies massive state-funded consortiums and bloated committees. But the assumption that a country needs sovereign compute infrastructure to compete in the global AI economy is a fundamental misunderstanding of how technology value captures markets.
The Fallacy of the Sovereign Stack
Every tech cycle follows the same script. Capital rushes to the infrastructure layer. The companies building the pipelines and pouring the concrete capture early headlines and dizzying valuations. Then, the infrastructure becomes a commoditized utility, and 99% of the economic value migrates to the application layer.
We saw this with the internet. European telecom giants spent hundreds of billions building out fiber-optic networks and 3G infrastructure. They took all the capital risk. Who captured the value? American software companies built on top of that infrastructure.
By crying foul over U.S. AI restrictions, leaders like Carney are effectively demanding that their nations spend billions of dollars replicating a hyper-expensive, depreciating hardware layer that OpenAI, Microsoft, and Google are already subsidizing to the tune of hundreds of billions.
Let them build the expensive grid. The real money is in selling the power.
When a government tries to build a "national AI champion" from scratch to counter American dominance, it almost always ends in disaster. I have watched mid-sized sovereign wealth funds and national ministries flush hundreds of millions down the toilet trying to build proprietary foundational models. They hire expensive consultants, buy thousands of Nvidia chips at inflated spot prices, and end up with a model that performs worse than an open-source alternative you can download for free.
The unit economics of foundational models are brutal. The cost of compute scales exponentially, while the marginal utility of the model’s intelligence scales linearly. Trying to out-compute Washington or Silicon Valley on pure brute-force infrastructure is an act of economic self-harm.
The Open-Source Loophole Washington Cannot Close
The entire premise of the "dependence panic" ignores the reality of modern software development: open source is eating proprietary AI.
When the U.S. government restricts access to specific API endpoints or high-end enterprise platforms, it creates a regulatory vacuum. Software engineers do not throw up their hands and quit. They turn to open-weight models like Meta’s Llama series, Mistral, or the thousands of fine-tuned variants available on Hugging Face.
Consider the mechanics of a modern enterprise AI deployment. If you are a Canadian bank, a European manufacturer, or an Asian logistics firm, you do not want your proprietary data flowing through a public U.S. cloud API anyway. Regulatory frameworks like GDPR and PIPEDA make that a compliance nightmare.
The strategy that actually works does not involve building a sovereign sovereign model. It looks like this:
- Download an open-weight foundation model that is already 95% as capable as the top proprietary U.S. models.
- Run it locally or on regional cloud infrastructure where you retain absolute custody of the data.
- Fine-tune the model using proprietary, hyper-local domain data that American tech giants cannot access.
The value is not in the weights. The value is in the data and the integration. By restricting access to their proprietary systems, U.S. firms are driving the rest of the world straight into the arms of open-source architectures that can be run anywhere, modified by anyone, and blocked by no one.
Dismantling the People Also Ask Mythos
Look at the standard questions dominating public discourse around this topic. The premises are almost universally flawed.
How can smaller countries achieve AI sovereignty?
The question assumes "sovereignty" means isolation. True digital sovereignty in a connected world is not about isolation; it is about asymmetric leverage. A country like Canada or a bloc like the EU will never win the raw compute race. But they can win on specialized data monopolies. Canada, for instance, has a highly centralized healthcare system with decades of longitudinal patient data. That data is infinitely more valuable for training specialized medical AI than ten thousand generic GPU clusters in Utah. Stop trying to build the engine. Build the specialized fuel.
Will U.S. AI export controls kill foreign tech startups?
No. It will kill unoriginal startups that are nothing more than a thin wrapper around an American API. If your entire business model relies on pinging a server in San Francisco to summarize text, you do not have a tech company; you have a leased digital storefront. U.S. restrictions act as a brutal, necessary evolutionary filter. They force founders to build real intellectual property, leverage local infrastructure, and solve specific, complex industry problems rather than building generic chatbots.
The Dark Side of Local Autarky
To be fair, rejecting the Carney narrative and focusing purely on the application layer has a massive downside. It requires a level of regulatory flexibility and risk tolerance that most legacy economies lack.
If you are not going to build the hardware, you must be incredibly fast at deploying the software. That means cutting through the red tape that paralyzes regional economies. You cannot play the application game if your local privacy regulators take three years to approve a data-sharing agreement, or if your tax structure penalizes founders for taking massive risks.
The contrarian approach demands a ruthless trade-off: you accept dependence on foreign hardware standards in exchange for absolute dominance in vertical execution. If your political system is too slow to allow rapid local deployment, you end up with the worst of both worlds—no infrastructure and no applications.
Stop Funding Infrastructure. Start Compounding Data.
If you are an executive or a policymaker reading the alarmist headlines about U.S. tech dominance, ignore the urge to lobby for national chip foundries or state-backed AI funds. That capital is dead on arrival.
Instead, pivot your strategy to where the asymmetric returns actually live.
First, aggressively audit your organization or region for data monopolies. Identify the datasets that are legally, culturally, or operationally impossible for an American tech giant to scrape. This could be regional supply chain metrics, local legal precedents, or specialized industrial telemetry.
Second, reallocate capital from hardware procurement to talent and integration. The bottleneck is no longer the availability of a model; it is the scarcity of engineers who know how to optimize, fine-tune, and embed these systems into legacy workflows.
The panic over U.S. AI restrictions is an artifact of old-world thinking, a remnant of an era when national power was measured in steel tons and oil barrels. In the AI economy, trying to hoard the infrastructure is a fool's errand. Let Washington overspend on the digital highways. Your job is to own the tollbooths.
