The Great AI Winter of 2026/2027

Rarely do technologies cross over into the mainstream media. Over the last few years, the Internet, cryptocurrency and AI stand out as several waves of tech that are or have been regularly reported on traditional channels such as the BBC.

Generally, hype is subsumed by reality, as the excessive hubris of Silicon Valley gets replaced by practical applications of technology from businesses that can afford or find ways to implement solutions in a financially positive way.

AI is the next wave of "excessive exuberance" to hit the tech market. As sure as night follows day, the current bubble will be met with a burst that will ripple across the industry.

Of course, that's not to say AI isn't useful technology. Machine learning and AI have been in use for decades, providing insights for the financial industry and offering mundane tasks such as automated quality control on factory production lines.

Today, though, we're talking about generative and agentic AI as some miracle advancement set to give us all much more leisure time and possibly free money. Unfortunately, that version of the future is far from reality. Instead, AI is being used to replace skilled artisans across all kinds of industries (notably programming) in a process widely described as "enshittification".

Dotcom Boom/Bust

Winston Churchill, George Santayana and Edmund Burke are all attributed with the quote (or variations thereof) that says "Those who fail to learn from history are doomed to repeat it". The boom and bust of the first dotcom era is a testament to that sentiment, but, of course, things are not that simple.

I was lucky (or unlucky) enough to be running a start-up in the late 1990s, just as the Internet was taking hold. Our idea for legal customised CDs and digital music downloads was just too early to be widely practical, despite the impact that Napster and LimeWire were having on the entertainment industry.

We succeeded with the technology where we ultimately failed in the business model of persuading the content-owning companies to give us their best material. Ultimately, almost a decade after we started, Spotify evolved the model from downloads (which were inherently risky for the music business) to streaming, where content was more easily controlled.

(Side note: when we started in 1997, broadband and widespread/fast access to the Internet didn't exist, so we were somewhat hamstrung by the infrastructure technology of the time. A decade later, this was less of an issue for Spotify.)

Our experience echoed widely across the IT industry during that time. Many businesses attempted to copy Amazon.com, and only a few of them remain today. Most were crazy ideas that had no hope of making significant revenue, never mind becoming profitable.

The underlying issue wasn't exclusively the technology, but mostly the business model. It was companies selling the wrong things to customers they couldn't either practically fulfil or scale, with overheads that couldn't be made more efficient. However, stick ".com" on your company name and instantly some venture capitalist (or many) would throw money your way. Something's got to stick eventually, right?

AI Boom

What does this have to do with AI? Clearly, the parallels are there. Hundreds of companies are being established and funded to solve niche issues using AI. Most of those start-ups will never make a profit and will disappear without a trace in 18-24 months.

However, arguably, this time around the stakes are different. Masses of capital is being invested (or promised) by well-established companies that currently have the deep pockets to fund massive infrastructure rollouts and LLM model development. The problem is that these companies are reaching the extent of their funding capabilities and looking elsewhere to shore up excessive spending targets.

Again we could ask, "Why is this a problem?". Well, when you visit a casino, betting and losing your own money, the consequences are on you. Visiting a casino to lose other people's money is generally a problem. This is where the AI Boom is headed - lots of businesses betting on other people's money with no obvious guarantees that any of that money can be repaid through revenue.

Meta

Let's look at one example. Meta Platforms, Inc., formerly Facebook, Inc., has bet big on the concept of the metaverse, literally changing the name of the company to reflect its ambitions. Various sources quote the value of Meta's investment in the metaverse to date at around $45 billion. Yet despite spending this eye-watering sum, Meta has barely anything to show for its efforts. In the last quarter (Q3 FY2025), Reality Labs (the VR division) made $470 million in revenue with a $4,432 million loss. Cumulative revenue for FY2025 is $1,252 million, and a $13,171 million loss.

Arguably, Meta can sustain this level of loss-making from revenue generated elsewhere (Reality Labs accounts for only 0.93% of Meta revenue). Zuckerberg has an iron grip on the direction of Meta, so any change of direction will only be influenced by the CEO's office.

However, let's now look at Meta's promised spending for AI. In November 2025, the company announced a three-year investment of $600 billion in AI infrastructure and jobs for a technology that has no obvious revenue. Touching on the investment value first, that's $200 billion a year for a company which made $164 billion in revenue during FY2024, with $62 billion profit. There's a clear gap here between Meta's ambitions and its available funding, even if shareholders see zero dividends for the next three years.

The answer will be to cut into reserves (cash & cash equivalents are $78 billion at the end of FY2024) and increase borrowing (long-term debt is currently $28.83 billion).

Meta's investment could look prescient if there were some prospect of revenue from it. Currently, Meta "open sources" its AI models. This doesn't necessarily imply free, but there is no clear use case for LLAMA today that could recoup $600 billion of spending by the end of the decade.

Champagne Tastes, Beer Money

Looking wider across the industry, a widely shared comment on X indicates JP Morgan believes the AI industry needs to generate $650 billion in annual revenue to deliver a modest 10% return on investment. This is the equivalent of every existing iPhone user spending an additional $35 per month, or every Netflix subscriber spending $180/month. What value could 1.56 billion users see for an additional $35/month spending on AI - even if they could afford it?

Taking an alternative view, again from JP Morgan, the company recently published an Outlook 2026 report that states it sees no bubble in the current AI spending. This report is worth reading as it highlights metrics that could provide indicators of future market stress.

Conclusions

So what are we to think? No boom runs forever (let's not forget UK Chancellor Gordon Brown's famous proclamation from 1999). At some point the AI rollercoaster will come to either a crashing end or gradual stop. The question to ask is whether, with a desire to win, the "Magnificent 7" will eventually temper their spending and accept there will be no single winner, or will continue to bet other people's money on a game of Winner Takes All.

This is the single greatest issue of the AI boom - not whether the technology has value, but the risk that the players will take the house down in their desire to own it. Will we see an AI Winter in 2026/2027? Only time will tell.