Earlier this month, Prime Minister Mark Carney announced the government’s ‘AI for All’ initiative. Coming on the heels of news that Canada is in a recession, Carney’s AI for All plan was designed to signal that the Canadian state was hitching its wagon to the AI economic engine. But broader questions abound about the health and sustainability of AI investment. Carney’s efforts to direct state resources to spur private AI economic activity are unlikely to work and carry immense risks for the Canadian economy.
AI driving increased capital expenditures
While the Canadian economy lurches along, AI related investments in the United States are driving a new wave economic activity. A handful of companies in the tech sector, the so-called “Magnificent Seven” account for a huge proportion of growth in the stock markets. These companies, and a handful of private companies such as Space X, Anthropic and Open AI are attracting massive amounts of investment funds to build AI hyper scaler data centres, new chips and other AI related infrastructure.
Last year businesses in the U.S. invested a whopping $1.5 trillion in IT equipment and software. This level of business investment is only set to grow, as more companies plan to increase these business investments. Hyperscaler capital expenditure from six companies contributed roughly 0.4 percent to U.S. GDP in 2025. In 2026 it’s tracking at 1.2 percent to U.S. GDP. The Bank of America estimated that capital expenditure will top $800 billion this year alone, up 67 percent from 2025. This record level of investments in AI are crowding out investments in all other sectors of the economy — except energy.
After a decade and a half of sagging levels of business investment, 2025 saw a noticeable uptick in the rate of capital expenditure in the U.S. and across the globe. Investments in AI and energy account for the lion’s share of investment growth. But it should be noted that part of the increase is accounted for by the rise in both energy prices and cost of high end chips. The increased costs in AI expenses are intersecting with geopolitical conflict in the Strait of Hormuz, tariff wars, and climate change.
A return of profits and productivity?
The twelve years following the 2008 Great Recession were marked by a slump in the rate of profit growth across the globe. Businesses struggled to attract investments, the rate of productivity growth lagged, and drivers of previous economic growth in the globe such as the Chinese economy cooled. This crisis of profitability was acutely felt in North America and Europe.
Now, in the post-pandemic recovery, average growth rates of profit in the corporate sector in the U.S. and across the globe have ticked up. Since 2021 the U.S. economy from the standpoint of workers has been marred by a cost-of-living crisis — driven by supply chain problems, climate change and geopolitical crisis in Ukraine and the Middle East. While wages have increased, they have not kept pace with inflation. Interestingly, the rate of productivity growth in the same number of years has increased. As the Bureau of Labour Statistics notes, labour productivity in the United States grew at an average rate of 1.5 percent between 2007 and 2019. Since 2020, the growth of the average rate has increased to 2.1 percent. Since 2023, the productivity growth rates have averaged 2.8 percent. This is a significant jump. Since the pandemic, the productivity pay-gap has increased to its highest (or near highest) level on record.
AI and work
While it might be tempting to say that AI is transforming the rate of exploitation in the workplace, this is not the case. AI rollouts in workplaces have only recently accelerated. Most AI use has been consumer facing. In the United States capital investment has surged in the AI race and this has driven up costs. Roughly two-thirds of data centre costs flow to imported components.
While there has been much hype around AI adoption being transformational, when we look under the hood at the broader economy there is scant evidence that AI has dramatically transformed workflows. As a recent JP Morgan brief stated “Survey evidence suggests firms remain much closer to experimentation than scaled AI transformation.” A recent survey shows increasing adoption of AI in medium to large firms, but that firm’s use of AI tools has had minimal impact on productivity or even workflows.
This dovetails with a recent Statistics Canada study on potential changes to productivity brought on by AI, which concluded that “there is no statistically significant direct association between AI adoption and productivity.” Even Sam Altman has walked back claims
In the face of a wide gap between the promise of productivity increases and the reality, even AI boosters such as Anthropic CEO Dario Amodei and Open AI CEO Sam Altman have walked back their claims about AI driven job replacement and productivity increases. Altman admitted “I thought there would have been more impact on entry-level white-collar jobs being eliminated by now than has actually happened.” As the Wall Street Journal’s tech journalist Parmy Olson noted there is “scant evidence of software models replacing workers on any large scale” with layoffs in sectors with wide AI adoption more accurately being the result of previous overhiring than AI adoption.
Part of the reason that AI’s impact on productivity and workflow has been much lower than expected is that firms actually do not know what to do with the technology itself. This is the problem with Large Language Models — they are probabilistic text predictors, not reasoning engines. They are inherently prone to mistakes or hallucinations, with some studies putting the rate of hallucination at 6 percent to 10 percent. This makes them unreliable, often producing text that needs to be rechecked and verified. AI might save time on basic functions such as note taking and standard emails, but it also creates work when being deployed for more complex tasks.
The work of babysitting or botsitting AI has largely offset many of the potential gains in the workplace. As a recent Financial Times article put it, “We find a lot of the [time] gains are being absorbed by what we call ‘botsitting’, the hidden work of making AI usable, feeding it context, checking outputs, rerunning prompts and cleaning up mistakes. Workers report spending 6.4 hours a week on botsitting.”
In firms that have adopted AI into their workplace, there is now a wide recognition that workers are using AI as a form of workplace theatre, to show they are using the technology rather than actually engaging in productive tasks. Considering how expensive AI is, this is extremely unsustainable.
Over the last number of months there is a growing concern by firms who have adopted AI about the exorbitant cost of using AI tokens. AI developers are struggling to find models that can see a return on investment. Major AI developers have been adept at attracting billions of dollars in investment. The problem is they are all losing money hand over foot. They have yet to find a model that could conceivably break even — let alone turn a profit. Chat GPT is on track to lose $14 billion this year, while Anthropic is set to lose close to $5 billion this year.
Initial flat fee user subscription models to AI to usage-based pricing. The cost to companies using the service is enormous and there has been a move to cut usage of tokens. Companies such as Uber, Microsoft, and Amazon are all instituting AI token caps for internal use. Companies are realizing the price of deploying AI is in some instances exceeding labour costs. Companies are less likely to adopt and stick with AI on a mass scale unless it shows major productivity gains, allowing them to cull their workforce.
Blowing into the bubble
The growing use of LLMs by companies and individuals over the last five years has raised all sorts of social and economic issues. Individuals and firms are trying to separate the wheat from the chaff in terms of how best to utilize this technology. The flawed nature of LLMs is creating real barriers about how this tech can be deployed at a firm and individual level. How useful is AI? And when you factor in the costs of developing AI, is it worth it?
The UK Investment Bank Panmure Librium notes that the five largest U.S. hyperscalers (Amazon, Microsoft, Alphabet, Meta and Oracle) need wrangle up between $2 trillion and $5 trillion in additional annual revenue to meet their planned data centre investments, which are projected to reach $658 billion in 2026 alone. Even under the most favourable circumstances assuming zero costs, four of the five companies are on track to lose billions of dollars over the next five years on their AI investments.
As the economist Michael Roberts notes, the economy is now just one big trade on AI. The AI engine is powering renewed profits and economic growth, but its underlying valuations seem to defy economic reality. For now, AI seems to be working as an economic investment. The global economy is growing, the rich are getting richer on the promise of AI — Elon Musk became the first trillionaire on the back of SpaceX’s Initial Public Offering. This all rests on assuming AI will be a profitable investment, but if it fails to deliver, it will be a world of economic pain.
Carney’s budgetary commitments to invest 1.5 billion in AI and his recent ‘AI for All’ sound smart, responding to a trend in the market, but it could be a disaster. The initiative is less about a sound investment in the Canadian economy and more about furthering the interests of capital. Money for child care and hospitals will deliver far more positive economic and social impacts than handing out cash to private companies for yet more data centres. But for now, Carney is committing to blowing public money into the private AI bubble regardless of the potential long-term impact on workers and the economy. The question is who will be left holding the bag if it pops.
Author
Did you like this article? Help us produce more like it by donating $1, $2, or $5. Donate
