Everyone seems to have an opinion about when the AI bubble will burst.
Some believe the enormous spending on models, data centres and infrastructure cannot continue. Others think AI will eventually justify these investments by transforming how businesses operate.
And then there are people waiting for the bubble to burst so they can go back to doing things the way they did before AI.
I think we're mixing up two very different questions.
What happens if the AI investment bubble bursts? And what happens if AI becomes so effective that it starts weakening the market it depends on?
The two outcomes aren't mutually exclusive. And the second could become a problem even if the first never happens.
A financial correction doesn't reverse technological adoption
Let's start with the bubble everyone is talking about.
AI companies are spending heavily on infrastructure, computing capacity, research and products. Whether the returns eventually justify that spending remains an open question.
Some companies may fail. Valuations may fall. Investors may become unwilling to fund expensive experiments.
But none of that automatically means businesses will stop using AI.
An investment bubble bursting doesn't make the underlying technology disappear.
Companies that have already found cheaper or more effective ways to perform certain tasks aren't necessarily going to return to their old processes because an AI company lost its valuation.
Some implementations will prove too expensive or unreliable and be abandoned. Others will become permanent parts of how businesses operate.
And that's where the second problem begins.
Cost-cutting doesn't begin and end with layoffs
Whenever we discuss AI replacing human work, the conversation almost immediately turns to jobs.
Understandably. Layoffs are visible, personal and consequential.
Canva is restructuring parts of its workforce. Microsoft and Google have offered voluntary exit programmes, and now Meta has reportedly introduced voluntary exits too.
Different companies, different decisions, but all part of a broader conversation about how businesses are rethinking their workforce and operating costs.
Not every workforce reduction is caused by AI, and voluntary exit programmes aren't the same as layoffs.
But employees aren't the only costs businesses are trying to reduce.
Think about everything a company pays for to operate.
Marketing agencies, design partners, consultants, software subscriptions, outsourced support, research services, development vendors, operational partners and a long list of B2B relationships.
Each expense exists because the business needs something done and has decided that paying someone else is the most sensible way to get it done.
Now introduce technology that makes some of those activities possible internally, faster or at a lower cost.
The calculation changes.
Why renew a contract at the same price if a smaller team can accomplish enough of the work? Why maintain five subscriptions if one product covers most of their functions? Why outsource a process if it can be automated reliably?
Of course, cheaper doesn't always mean better. Some businesses will retain specialist partners because quality, reliability and expertise matter.
But not every customer needs the best possible solution. Many need something good enough for the job at a price they can justify.
Any expense for which a sufficiently good, cheaper alternative emerges becomes vulnerable to replacement.
This isn't limited to AI either. Businesses have always looked for ways to reduce costs. AI simply expands the range of work for which alternatives are becoming available.
And when a business eliminates an expense, someone else can lose revenue.
One company's saving is another company's revenue
Consider a company that reduces spending with an external agency.
For the client, that's a cost saving.
For the agency, it's lost revenue.
If enough clients make similar decisions, the agency may reduce hiring, cut its own subscriptions, renegotiate vendor contracts or downsize.
Those vendors then face the same decisions.
A software company losing enterprise accounts might reduce its spending on contractors. A consulting firm losing retainers may stop purchasing certain services. An outsourced operations provider may need fewer employees.
The effect doesn't have to begin with layoffs. It can begin with a contract that doesn't get renewed.
And it doesn't have to affect only businesses selling AI-replaceable work. Reduced revenues can eventually affect spending across other sectors too.
This is the domino effect I find concerning.
The customer one business can afford to lose may be the customer another business cannot survive without.
At the individual company level, every decision might make financial sense.
But across thousands of companies making similar decisions, the consequences become more complicated.
The employee is also a customer. So is the vendor.
Businesses don't operate in isolation.
The employee whose salary appears as a cost on one company's books is a customer somewhere else.
The agency whose contract gets discontinued buys software, pays employees, rents offices and purchases services from other companies.
The consultant losing projects is also a consumer.
And the software company losing subscribers has its own suppliers and employees.
If enough people and businesses lose income without finding comparable replacement opportunities, they have less to spend.
Lower spending can weaken demand. Weaker demand can encourage businesses to cut costs further.
Lower costs for one company can mean lower revenues for another. Lower revenues can trigger more cost-cutting. And the cycle can reinforce itself.
That doesn't mean every cost reduction is harmful to the economy.
The money saved doesn't vanish.
Businesses can reinvest it, lower prices, pay higher wages, develop new products or distribute profits. Cheaper products can help consumers afford more. New industries and employment opportunities can emerge.
But where those productivity gains go matters.
If the gains are distributed widely through lower prices, new employment and rising incomes, they can support demand.
If a substantial share accumulates among a relatively small group of businesses and investors while displaced workers and suppliers struggle to replace lost income, the effect can be different.
The economy may become more productive without enough purchasing power reaching the people and businesses expected to buy what it produces.
And there's another issue: timing.
Businesses can cancel contracts, freeze hiring and automate workflows much faster than affected people and companies can necessarily develop replacement sources of income.
AI can also affect several kinds of work simultaneously, from design and writing to software development, analysis and administrative operations. That creates the possibility of disruption across multiple industries before workers and businesses have fully adjusted to the changes.
Even if new opportunities eventually emerge, the transition itself can be painful.
The companies selling AI still need paying customers
This is where the argument comes back to the AI industry.
AI businesses are not exempt from the economics affecting their customers.
They need companies willing to pay for enterprise tools, developers willing to pay for APIs, and individuals who can justify subscriptions.
They also need those payments to support the costs of delivering their products.
What happens if businesses increasingly use AI to reduce spending across their operations, but the cumulative effect leaves more of their potential customers with constrained budgets?
A company may be producing more work with fewer resources and still face customers who are unwilling or unable to spend as much.
An AI company might have an excellent product, growing usage and impressive technological capabilities, yet struggle to turn that usage into sufficiently profitable revenue.
This is not inevitable. AI could create entirely new markets, lower the cost of producing goods and services, and enable businesses that would never otherwise have existed.
Those possibilities matter.
But we cannot simply assume that every income lost will be replaced, or that every saving made by one company will eventually return as spending somewhere else.
The question is not only whether AI creates enough value. It's whether enough of that value reaches the customers the industry needs.
The bubble I'm more concerned about
I use AI extensively.
It has made my life meaningfully better. I've been able to build and experiment with products that would have been much harder to attempt independently a few years ago.
I'm probably busier than I was before, but it's a better kind of busy.
I also understand what subscriptions, API calls and the ongoing costs of experimentation feel like when you don't have regular cash flow coming in.
The ability to do something with AI and the ability to keep paying to do it are two different things.
That experience is part of why I keep returning to the customer question.
I'm not arguing that AI will eliminate every job, destroy every B2B service business or inevitably cause a collapse in consumer demand.
And I don't know when, or even whether, the current investment boom will experience a major correction.
What interests me is the possibility that the technology could succeed at reducing operational costs while the wider economy struggles to adjust to where the resulting income and spending go.
The irony is that the industry could face two very different kinds of financial trouble.
One because investors stop believing AI will generate sufficient returns.
The other because AI becomes effective enough to change the economics of its customers, suppliers and workers, potentially weakening the demand on which future returns depend.
And bursting the first bubble would not necessarily solve the second.
If businesses become increasingly capable of operating without paying as many people and other businesses, who will have enough money to keep buying what those businesses produce?
That's the question I think deserves more attention than predicting the date of the next market correction.