Asset managers weigh in on the prospect of a token futures market, chips, and other inputs as AI displaces labour
Artificial intelligence (AI) has promised a new rise in productivity growth, largely untethered to the cost of labour. That promise been realized, to a certain extent, in sectors like software where AI tools have shown promising capacity to write code, improving individual worker output considerably and prompting a sell-off in software stocks on equity markets. More companies are moving to use AI, a study by KPMG released at the end of 2025 found that 93 per cent of Canadian businesses are exploring AI usage, and 31 per cent have implemented AI tools across their core operations. As AI becomes a source of productivity, the question for investors and economists may now lie in how we measure its costs.
If businesses are spending on AI services to increase productivity, then the cost of those services might become as vital of an economic indicator as the price of oil or the cost of labour. Tokens could be one measure for the cost of that AI compute as an economic input. Tokens have become the primary way that AI service providers like Anthropic and OpenAI bill for their services. A token is essentially a unit of account representing compute and all the inputs that go into the compute required to respond to a prompt or execute a task. Tokens in their current state may be an imperfect metric, but China is now establishing the framework for a token futures exchange in Shanghai, according to Reuters.
“They’re absolutely going to become a financial asset and there'll be futures traded on it and options and gamemanship and everything in between. The good, the bad and the ugly of finance,” says Craig Jerusalim, Senior Portfolio Manager at CIBC Global Asset Management in Toronto. “And I think that also spurs innovation as long as it doesn't go crazy or overboard. But I don't think that the financialization of token infrastructure is an important development because what you've seen is that for other technologies the ramp up curve from early adopters to the mass markets can take decades at times.”
Why tokens are an imperfect measure, for now
Elliot Johnson, Chief Investment Officer at Evolve ETFs in Toronto, sees more hurdles to AI tokens as the measure of compute cost. He notes that as it currently stands, tokens should not be seen in the same light as a barrel of oil because they aren’t attached to specific physical scarcity in the same way. Unlike oil consumption, he says, token usage is unpredictable and only calculated after a task has been executed. Users are often surprised, having blown through their daily token allocation by mid-morning when completing certain tasks. Moreover, tokens aren’t fungible. If he has unused tokens left over from his Claude account, he can’t sell them back to Anthropic or use them to run tasks on ChatGPT.
There is a push towards greater tangibility of token usage, that might solve for some of the opacity problems Johnson sees. AI models are adding what’s sometimes called a ‘harness’ or an ‘orchestrator layer’ that communicates with the user and farms out tasks to other AI tools. That layer, Johnson says, can be run on a computer and shouldn’t require additional costs itself, but it may be able to provide something like a metered model that allows users to factor in token usage when they move to execute a task.
The corollary to tokens, therefore, is not a barrel of crude oil, it’s a cubic meter of natural gas or a kilowatt of electricity as provided by a utility company. While prices for those inputs are also tied to market forces, they may not have the same investment case as more raw commodities. If that’s the better analogy, then investors may ask themselves what to look for as the key inputs that will produce AI productivity and how to invest in them.
“If I could answer that question, I’d probably be on a beach somewhere because I think it is the question,” Johnson says. “It’s confounded by so many challenges. How much venture money is being spent to keep the cost of compute at a certain level rather than another level? How much is debt fueled? Where is the business case coming, coming from? But some of these questions are starting to get answered.”
How to invest in AI’s productivity inputs
For retail investors and advisors, Johnson still believes that looking for AI infrastructure bottlenecks will continue to serve as the primary source of exposure to this stage of the AI cycle. He highlights three areas in constant short supply: electricity, data centers, and chips. Within those broad categories, though, there has been a significant degree of widening. Johnson notes how chip scarcity began just with Nvidia GPUs, before moving on to CPUs and memory chips.
As things shift and evolve, Johnson insists on diversification as the route to AI access. He argues that broad exposure to AI companies should allow investors to pick up the winners, especially as some of the biggest private names in AI look set to make IPOs of their own. While the market begins its slow and inexorable process towards pricing AI as an economic input, Johnson argues that advisors and their clients want to remain in the space.
“I think if you’re going to say, ‘I’m going to skip this as a major main allocation of my portfolio,’ I think you have to have strong conviction that this is not going to work out, that AI is going to go away or become less relevant in the future. I certainly don’t feel that way,” Johnson says. “I just feel like every few months I have this other moment of amazement because the AI tools I’m using do something that I didn’t know they could do before. I just feel like that hasn’t run out yet. We’re not yet at the point where it’s mundane.”
