Leon Furze

When the AI bubble bursts – and it will burst – CEOs will be dethroned. Companies will lose billions. The economy – particularly in the US – will take a catastrophic hit. Data centres will suddenly need to downsize their operations and close down entirely. And in education, we’ll need to take yet another long hard look at ourselves, and ask, what’s next?

The current AI bubble, which has mostly been hyped along by products like ChatGPT, image and video gen, and now AI-wearables, is strikingly similar to the dot-com bubble of the late 1990s. I was a young teenager back then, and blissfully unaware of things like “the economy” and “politics”. But with the gift of 20/20 hindsight, it’s easier to see why people are making the comparison.

Monopolistic technology companies, insane amounts of money, and needlessly excessive infrastructure (fiberoptic cables then, data centres now) all line up with the current state of play for GenAI, and even Sam Altman – the king of magical GenAI thinking – thinks we’re probably in a bubble.

The situation is absurd, at least to a layperson like me who has only a vague grasp on the accounting practices of trillion dollar companies. For example, when Microsoft funds OpenAI with billions of dollars that don’t actually exist in order to power a product that doesn’t make any money, and yet both companies still somehow claim a profit. As Cory Doctorow explains it:

Microsoft “invests” in Openai by giving the company free access to its servers. Openai reports this as a ten billion dollar investment, then redeems these “tokens” at Microsoft’s data-centers. Microsoft then books this as ten billion in revenue.

Cory Doctorow – https://pluralistic.net/2025/09/27/econopocalypse/

Meanwhile, in reality (the part of the world where you and I have to pay for things with actual money, and where the exchange of real money is expected to return actual goods and services), the outlook for AI isn’t looking great. People are refusing AI increasing volume. “Workslop” is making workers less productive and interfering with their jobs. Companies using AI are seeing little to no return on investment. And the consultants telling companies and the government to use AI can’t seem to figure out how to use it themselves.

See? Absurd.

If it works, break it

But there’s an interesting problem with all of these technology bubbles. Like the dot-com bubble, and the crypto bubble, and the blockchain bubble, the GenAI bubble is built on a technology that actually works. I use GenAI a lot for tedious administrative stuff like fixing code on this website, converting handwritten notes into .csv files, and tidying up audio transcriptions. I have spoken to many educators and students who use these products daily to make their lives easier.

Image generation, video generation, and audio generation have all reached a point where they create materials broadly comparable to human output. Whatever your personal stance on these AI platforms, it’s hard to deny that the technology behind them actually works.

So if it works, why is the bubble going to burst?

Because for these technology companies, “it works” isn’t enough. To incentivise the billions of dollars of investment being thrown at them, the technology has to go beyond “it works” and into the breathless territory of “it’s a game changer!”, “it’s revolutionary!” and “it’s going to change the world!”. Investors need to be convinced that a huge return on investment is just around the corner, and governments need to be convinced that tedious things like regulation and laws will just get in the way.

And so, the technology is hyped up beyond all reasonable expectations. Persuasive statistics like “ChatGPT has 800 million active weekly users!” are thrown around as a proxy for potential profits, despite the fact that the companies are burning cash just to keep the lights on. Free accounts lure users in, with rate limits and hidden features encouraging people to subscribe, but even the paying customers aren’t anywhere near enough to meet the financial demands.

Even a technology that works, both in theory and in practice, can’t live up to the expectations of the investors and the fierce, zealous optimism of the CEOs.

What gets left behind?

When the bubble bursts, there will be something left behind. In fact, there will be many different somethings; negative and positive residue which impacts economies, technologies, and maybe even the ways in which we interact with our digital worlds. The comparison of the AI bubble to the dot-com bubble is again helpful here.

When the dot-com bubble burst, companies went out of business so fast that startup founders could be seen on the streets of San Francisco selling their furniture and servers as they tried to keep up with the rent on their cool office spaces. But it also created the environment for a flourishing of new ideas, and the seeds of the “web 2.0” era where blogging and personal websites really took off.

With the AI bubble, some are skeptical about what kind of residue it will leave behind. Researcher Alex Hanna, – co-author of The AI Con with Emily M. Bender – brings up industrial era comparisons, believing the bubble-burst will leave destruction in its wake:

“The residue of the bubble will be sticky, coating creative industries with a thick, sooty grime of an industry which grew expansively, without pausing to think about who would be caught in the blast radius.”
Alex Hanna

Doctorow sees the imminent burst as a threat to the economy, but also encourages people to prepare for the productive residue:

“Plan for a future where you can buy GPUs for ten cents on the dollar, where there’s a buyer’s market for hiring skilled applied statisticians, and where there’s a ton of extremely promising open source models that have barely been optimized and have vast potential for improvement.”
Cory Doctorow – https://pluralistic.net/2025/09/27/econopocalypse/

I agree with both sentiments: the residue from this bubble burst will be just as grimy and harmful as the others before it. I’m sure we will see revealed the kind of fraud, criminal behaviour, and corporate in-fighting that followed the crypto crash. We’ll have years of court cases documenting everything that went wrong during the inflation of the AI bubble from 2022-202X. We’ll see news reports of billions lost, companies put out of business, jobs displaced, and the failings of governments to shore up the economy against wishful thinking.

But I also believe that useful technologies will be left behind. Large Language Models might already have reached the peak of their utility, but they will continue to be useful. Subsequent research will continue to make them more efficient to train and run, and open source versions of proprietary models will continue to improve.

Image, video, and audio generation will rock the music, film, and creative arts industries, but people will find ways to co-exist with the technology. There will still be people on both sides of the fence – for and against AI – but without the constant barrage of headlines and advertisements the technologies will recede into the woodwork and become what they always have been: normal, run of the mill, mundane.

Productive residues

A few weeks ago I published a collection of flash fiction, originally shared on LinkedIn (don’t ask). One of the stories looked at the idea of productive residue – the good things left behind when the bubble pops. Sometimes (most of the time) my speculative fiction is… a bit depressing. But this is a rare example of my optimism about these technologies.

I’ll end this article with the full story, and a similar call to Doctorow’s: prepare for a future where GenAI is commonplace, ordinary, and effective technology. Prepare for a time when your students and colleagues use AI, or don’t, because of their own values and not because of tech companies ramming it down their throats. Prepare for a future – hopefully sooner rather than later – where these technologies are genuinely accessible, cheap, sustainable, and helpful. In short, hunker down, wait for the bubble to burst, and prepare yourself for whatever comes next.

Afterglow

When the GenAI bubble burst, it didn’t pop so much as sigh. VC
dashboards dimmed; LinkedIn fell mercifully quiet; chatbots
flickered offline.

Ada barely noticed – until the hospital’s new scanner flagged a
tumour four millimetres smaller than last year’s model. Turns out
the vision-encoder came from a gutted multi-billion dollar
dialogue agent, chatty layers stripped away, image recognition
surprisingly solid.

Across the Yarra, Mal busks with a pocket synth that harmonises
in real time with tram brakes and birdsong, audio engine salvaged
from an abandoned speech model.

Chris spins up a prototype app before handing it over to his
team; expert programmers who decided long ago to dispense with
the anthro parts of the LLM and just keep the good stuff.

Developers call it “harvesting”: break the bot to bits, keep the
stuff that works.

The chatbots have shut their virtual mouths, but they’ve left
behind something greater than the sum of their parts.

Want to learn more about GenAI professional development and advisory services, or just have questions or comments? Get in touch: