All in at the AI casino

Artificial intelligence (AI) is eating the world—its data, its capital, its energy, its attention. Never before has a novel technology been so quickly and so forcibly integrated into every facet of our lives. No time for deliberation, no expense spared.

The biggest boosters of the various technologies and products that fall under the “AI” banner are, unsurprisingly, the big tech companies that have the most to gain from widespread AI adoption. But they are not alone: governments around the world are all in on AI, and Canada is no exception.

In its recent AI strategy, the federal government declared that “to thrive in the era of AI, Canadians need to trust in its promise.” It was a revealing plea. If the benefits of AI were obvious and immediate—and the harms merely speculative—there would be no need to appeal to the technology’s future promise.

Instead, the opposite is true. The proliferation of AI is awash in immediate harms, but the benefits remain elusive. Widespread AI adoption is thus a costly bet—a gamble on future prosperity with our jobs, our planet and very minds at stake.

Welcome to the AI casino.

Ante up

The scale of the AI mania is extraordinary. Global investment in the AI supply chain may top US$2 trillion this year—approaching two per cent of global GDP. Most of that money is pouring into the physical infrastructure that enables the AI economy. In the next three years, an estimated US$3 trillion will be spent on the construction of new data centres alone. Within a decade, AI is projected to be a US$5 trillion industry—comparable to the entire economy of Germany or Japan.

The sums are not only large in absolute terms, but they are also large relative to other sectors of the economy. More than half of all venture capital is now being absorbed by AI companies. In the U.S., AI investment accounted for a third of GDP growth in 2025. Major AI companies, such as OpenAI and Anthropic, are now valued at a trillion dollars apiece. Legacy big tech firms, such as Alphabet (Google) and Microsoft—already among the largest corporations in history—are pivoting aggressively into AI infrastructure and products.

AI industry investment in Canada is smaller by several orders of magnitude. Compared to the US$286 billion invested in U.S. AI companies in 2025, only US$4.3 billion was invested in Canadian firms. Compared to the trillion dollar valuations of leading American AI companies, the largest-AI focused firm in Canada, Cohere, is valued at “only” US$7 billion.

For all its talk of aggressively embracing AI, the federal government is only ponying up $2 billion for new computing infrastructure—a tacit acknowledgment that Canada cannot beat the U.S. at its own game. Nor should we try. But the feds nevertheless remain fervently committed to AI adoption as a matter of principle.

Among other goals, the national AI strategy aims to increase the rate of AI adoption among Canadian businesses from 12 per cent today to 60 per cent within a decade. The strategy hopes to reach a million post-secondary students and thousands of K-12 teachers with AI “literacy” programs. And it aspires to “transform public service delivery” through AI.

And if all of those AI tools happen to be made and owned by American firms? So be it. Digital sovereignty—the principle of domestic (and, ideally, democratic) control over the technology we use—is an aspiration to which the federal government pays lip service. But, when push comes to shove, AI adoption is a higher priority.

After all, the benefits of AI adoption are so incontrovertibly large and widespread that we cannot afford to get bogged down in questions of risk and harm. Right?

Betting the farm on productivity

The AI boom has several premises, including an appeal to the alluring and enduring myth of inevitable technological progress. But the big bucks pouring into the industry today are driven by a simpler promise: productivity growth.

Productivity refers simply to a worker’s output per hour worked. Productivity growth is how much that output per hour worked increases over time. Continued growth in labour productivity is necessary for modern, debt-fuelled economies that would otherwise collapse in on themselves. And, indeed, employers have succeeded for decades in squeezing more and more value (and profit) out of every worker—whether or not that value was shared with workers themselves. Corporate enthusiasm for AI is merely the latest incarnation of that profit-motivated logic.

According to Claude developer Anthropic, AI adoption will double the rate of productivity growth in the next decade. Independent assessments are less rosy, but still tend to converge on potential gains in productivity growth in the range of 10 to 20 per cent. If that pans out, it would indeed add trillions of dollars to the global economy. In Canada, industry-adjacent studies project a cumulative, AI-driven boon of $300 billion to the Canadian economy over the next 10 years.

The problem is that actual productivity gains are elusive. The vast majority of corporate AI pilots are finding no return on investment. A Statistics Canada study concluded that “there is no statistically significant direct association between AI adoption and productivity” among Canadian firms to date. Companies that embraced AI tools and started laying off workers, including Ford and IBM, have started rehiring them. Half of employers who made AI-driven job cuts admit it was a mistake.

There are a few reasons for AI’s economic underperformance. The first is technological. Despite the appearance of intelligence, AI tools—especially the large language models (LLMs) underpinning popular chatbots, such as ChatGPT, Gemini and Copilot—are fundamentally unintelligent. They are prediction machines that do not (and cannot) know whether their own outputs are accurate or reasonable. Even as these tools get more capable and reliable over time, they can never be 100 per cent trustworthy. To make matters worse, they are also inherently backward looking, capable only of inference based on past data, which means they struggle with novel situations and edge cases. Those are big problems in contexts where accuracy, judgment, creativity and liability matter.

The second problem is institutional. Most individuals and organizations do not know how best to deploy these technologies—or whether they should be deployed at all. Handing a nurse a screwdriver and saying, “use this, it’s the next big thing,” does not automatically make that nurse do more or better work. To the surprise and chagrin of many in the tech sector, not every job is like software development.

In fact, AI adoption can often have the opposite of the intended effect. For example, the amount of time workers must now waste reviewing and parsing what researchers call AI “workslop” often offsets any time saved in the first place. The assumption that indiscriminate AI adoption will make all workers and/or organizations more productive has not been the case in practice, and it is bound to be wrong in some (perhaps most) cases moving forward.

The big question then—and the big bet underpinning the trillions of dollars in AI investment—is whether those promised productivity benefits will actually materialize at scale. If they don’t, AI mania will prove to be the largest financial bubble in history, and its collapse would have massive and far-reaching consequences. Index and pension funds, including Canadian funds like the CPP, are now deeply implicated in the AI industry, so the damage would not be localized.

But even if the technological and institutional hurdles are overcome and the AI industry delivers on its promised productivity panacea, an even more important question remains: is it worth it? 

Skin in the game

A true accounting of the costs of the AI boom is a daunting task. There are the environmental costs of data centres, especially the burning of fossil fuels to meet ballooning electricity demand. There are the democratic costs of AI-fuelled mis- and disinformation. There are the human rights costs of AI-enabled corporate and state surveillance. There are the cognitive costs of offloading more and more of our thinking to machines. There are the security costs of entrenching dependence on foreign, especially American, AI infrastructure. There are the cultural costs of AI systems imbibing the totality of human intellectual and artistic expression without consent, credit or compensation. The list goes on.

Each of these costs represents an existential threat in its own way—to the habitability of the planet, to the viability of democracy, and so on. AI is proving to be a force multiplier for so many of the challenges we are already struggling with.

Any claims of AI’s economic benefit to Canada must be weighed against the profound risks it poses across all of these other domains. But even in the domain of the economy and of worker productivity, it is not clear that workers in the aggregate will be made any better off by the federal government’s enthusiasm for AI adoption. We are rolling the dice both in terms of job quantity and of job quality.

On job quantity, there is intense academic disagreement about the risk of AI-related automation of the workforce. While most workers, especially knowledge workers, are theoretically exposed to AI for at least some tasks, that does not necessarily equate to replaceability. As noted above, AI-driven layoffs have proven to be premature—and have been reversed—in many contexts where the technology was simply less capable than the humans who lost their jobs. To date, AI adoption has not had a significant effect on overall employment levels.

Yet it is already evident that AI is having structural effects on the workforce. Stable overall employment levels hide worrying trends. In the tech sector, for example, hiring of senior level employees has remained stable or even increased as hiring of junior level employees has fallen precipitously. Banking, law and other sectors are following suit. In Canada, entry-level job vacancies are at their lowest levels in a decade despite a growing population, and youth unemployment is near record levels.

Young workers are thus the canaries in the coal mine. As AI systems improve and employers get smarter about deployment, the universe of at-risk positions (and whole professions) will inevitably grow. Canada’s national AI strategy aspires to create 250,000 new jobs through AI by 2031. It does not mention how many jobs will be lost along the way.

On job quality, AI is reorganizing work itself in ways that undermine the autonomy and dignity of workers—even where they are not outright replaced. Workers who were previously responsible for producing creative or intellectual work are increasingly relegated to reviewing the outputs of AI systems. Workers must also contend with algorithmic management, a term that refers to managerial tasks, including hiring and disciplinary decisions, being offloaded to AI systems. All of this is happening in contexts where AI-powered workplace surveillance is increasingly ubiquitous and brazen.

The upshot is an intensification and dehumanization of labour that would make Marx blush.

The house always wins

The cruel irony here is that, to the extent AI is already or will potentially improve workplace productivity, many of those benefits may prove to be short-lived. Gutting entry-level employment and deskilling the professional workforce creates enormous vulnerabilities for individual institutions and the economy writ large. Organizations that depend on senior professionals overseeing AI systems are going to be in a lot of trouble when those leaders move on and there are no junior workers in the ranks.

Geopolitical risk also looms over Canada’s AI enthusiasm. Canadian firms were recently cut off from leading American AI models by decree of the U.S. government. Microsoft, which is a pillar of the federal government’s AI infrastructure, has acknowledged it cannot guarantee the security of sensitive public data. Outsourcing public and private sector capacity to U.S.-controlled services is a noose we tie for ourselves.

When it comes to gambling, the only winning move is not to play. Eventually, the house—which, in this case, is the nexus of U.S. big tech companies and state imperialism—always wins.

For Canada, that has two major implications.

First, we need to slow down. Rushing to indiscriminately adopt AI across every sector and domain of Canadian life, as the federal government aspires to, is a dangerous and unnecessary risk, especially where those systems are not domestically controlled. The harms are too real and too immediate to justify the speculative, long-term benefits.

Second, to the extent that the various technologies underpinning the AI boom have the potential to be genuine public goods—as they do in narrow, sector-specific applications such as energy efficiency and scientific research—we need Canadian-controlled options. That will require investment and experimental adoption, but it is a far cry from AI chatbots being forced into every workplace and school in the country.

There’s no denying that AI is here to stay. And there is little we can do about rapacious U.S. tech companies. But we don’t have to go to their casino. It’s time for Canada to play our own game.

About the author

Hadrian Mertins-Kirkwood

Hadrian Mertins-Kirkwood (he/him) is a senior researcher and political economist at the Canadian Centre for Policy Alternatives. His work focuses on federal economic, social and environmental policy, especially in the areas of climate change, artificial intelligence and industrial strategy. Hadrian edits the monthly Shift Storm newsletter on climate and labour.