The AI Bubble: Hype, Reality, and Market Concerns

The air is hissing out of the overinflated AI balloon

Opinion There tend to be three AI camps. 1) AI is the greatest thing since sliced bread and will transform the world. 2) AI is the spawn of the Devil and will destroy civilization as we know it. And 3) "Write an A-Level paper on the themes in Shakespeare's Romeo and Juliet." I propose a fourth: AI is now as good as it's going to get, and that's neither as good nor as bad as its fans and haters think, and you're still not going to get an A on your report. You see, now that people have been using AI for everything and anything, they're beginning to realize that its results, while fast and sometimes useful, tend to be mediocre. Don't believe me? Read MIT's NANDA (Networked Agents and Decentralized AI) report, which revealed that 95 percent of companies that have adopted AI have yet to see any meaningful return on their investment. Any meaningful return. To be precise, the report states: "The GenAI Divide is starkest in deployment rates, only 5 percent of custom enterprise AI tools reach production." It's not that people aren't using AI tools. They are. There's a whole shadow world of people using AI at work. They're just not using them "for" serious work. Instead, outside of IT's purview, they use ChatGPT and the like "for simple work, 70 percent prefer AI for drafting emails, 65 percent for basic analysis. But for anything complex or long-term, humans dominate by 9-to-1 margins." Why? Because a chatbot "forgets context, doesn't learn, and can't evolve." In other words, they're not good enough for mid-grade or higher work. Think of them as a not particularly bright or trustworthy intern. That may be good enough for $20 a month, but - spoiler alert - AI costs will have risen by ten times or more by next year. Will bottom-end AI be worth that to you? Your company? Some businesses that bought into AI wholeheartedly are suffering from buyer's remorse. The Commonwealth Bank of Australia (CBA), for instance, is asking its former call center frontline employees to return to work. CBA found that the call level increased, and managers had to man the phones. The company even, believe it or not, "apologized to the employees concerned." And I bet many of you thought that customer service call centers would be one of the easiest things to switch to AI chatbots. Wrong! Surely, though, AI is getting better. Right? Right!? I mentioned a while back that we're already seeing AI models collapse, so I see no reason to believe that there will be some extraordinary new AI advance. Why should I? Why should you? Remember when ChatGPT-5 was going to be the next big thing? You should; it was only the other week. OpenAI CEO Sam Altman said ChatGPT-5 was like having "access to a PhD-level expert in your pocket." Mind you, it couldn't spell "blueberry," but hey, mistakes happen. The only problem was that mistakes kept happening. ChatGPT-5 has proven to be a dud. Or, as one popular Reddit rant put it in the OpenAI subreddit, normally a hotbed of ChatGPT fanbois, "GPT-5 is awful." I agree. So, what happens if companies decide that, since AI is not delivering any real return on investment, they should stop wasting money on it? Well, Torsten Sløk, chief economist at Apollo, a multibillion-dollar retirement investment company, said in July: "The difference between the IT bubble in the 1990s and the AI bubble today is that the top ten companies in the S&P 500 today are more overvalued than they were in the 1990s." I was around for the dotcom crash, but many of you weren't, so here's a quick history lesson. The NASDAQ saw a 77 to 78 percent collapse. Many companies didn't survive. Many others that you may think of as being too big to fail, such as Cisco, Intel, and Oracle, lost over 80 percent of their market value. Glancing at today's market, I see that all the AI companies have seen severe pullbacks, with Palantir leading the way down with a 17 percent drop in value. Even Nvidia has fallen by 3.9 percent. This isn't a bubble popping, not yet, but you can hear the air hissing out. Even Altman, who should really get an AI cheerleader costume, has admitted that AI is a bubble. His words, not mine. He added: "Are we in a phase where investors as a whole are overexcited about AI? My opinion is yes." But, waving his AI pom-poms, he continued: "Is AI the most important thing to happen in a very long time? My opinion is also yes." Sure, AI is important. In some industries, such as tech and media, according to MIT's researchers, it is changing how business is done. Most companies, though, have found that AI's golden promises are proving to be fool's gold. I suspect that soon, people who've put their financial faith in AI stocks will be feeling foolish, too. ®

Tech stocks are sending a warning

The first inklings of a cooling of the love affair between investors and big tech may be starting to emerge with a sell off in sector stocks this week. Given how tech occupies a special place in the markets ecosystem, it is a warning that should be taken seriously, not just for bracingly upbeat stock valuations but also the darker corners of the financial system too. Big tech has assumed a dominant position in global public markets because of a winner-takes-all dynamic that has created a handful of almost unfathomably enormous and successful listed corporate giants, particularly in the US. The frenzy around artificial intelligence over the past couple of years has added lashings of fuel to the fire. Biggest of the bunch, chips behemoth Nvidia, is now worth $4.3tn, or one-and-a-half times the UK's entire FTSE 100 index, give or take. This has built a very top-heavy structure in US stock markets, which in turn have swelled to occupy an unusually large chunk of global equities. The 10 biggest companies in the US, which are mostly tech-flavoured, with some finance bolted on at the bottom, now account for some 40 per cent of the S&P 500 and for a third of the revenue growth across the index over the past year. Big tech has done all the heavy lifting for investors in the US this year, hence why the S&P 500 is up 9.5 per cent so far in 2025 while the Russell 2000 index, which tracks smaller stocks, is up a more modest 4.2 per cent. Up to now, what has been good for tech in general and for AI in particular, has been good for global stocks. If, however, something meaningful were to go wrong with tech in general and with AI in particular then, well, it does not take a genius to figure out where I'm going with this. And guess what? Doubts are starting to creep in. To a large extent, this is because the sales patter from the high priests of big tech is really starting to grate, particularly as each new release of supposedly whizz-bang AI models generates diminishing returns. OpenAI discovered this when its GPT-5 AI model failed to dazzle the crowd earlier this month. Just as the excitement starts waning, OpenAI's Sam Altman said at a recent event that the hype has elements of a bubble. "Are investors over excited? My opinion is yes," he said earlier this month. "I do think some investors are likely to lose a lot of money, and I don't want to minimise that, that sucks. There will be periods of irrational exuberance." Few other senior executives in any industry at any other less hubristic time would feel comfortable saying this stuff out loud. Sure, Altman still believes in the "huge" if largely undefined potential benefits to the world stemming from AI. But even he appears to think there's froth here. Adding to the nerves, investor attention has settled on a July report from MIT, which said some 95 per cent of organisations are getting "zero return" on their investment in AI. "Just 5 per cent of integrated AI pilots are extracting millions in value, while the vast majority remain stuck with no measurable P&L impact," the report said. Suddenly, the idea that you can take a humdrum business, mix in some AI magic and spit out a lean, mean profits machine at the other end feels rather more unlikely. This is not just a matter for listed stocks. We should all be more mindful of the strains that are harder to see in more opaque private markets, which are important sources of funds for the baffling array of data centres needed to make AI tick. Huge consumers of AI technology like Amazon and Alphabet will fund only around half of the almost $3tn (yes, trillion) likely to be spent on AI infrastructure over the next three years. The bulk of the rest will come from private equity, private debt and venture capital. Private credit managers, even those with investments in this space, are increasingly wary of the crowding across the industry in to this one theme. "It's absolutely something that we talk about in every executive committee meeting," says Josh Shipley, head of Europe at PGIM Private Capital. "I don't think it's enough as a percentage of the private credit market that we can see a systemic shock" if something were to go wrong here, he said. Still, there is only one way to find out. Analysts at UBS also said in a note this month that private credit has become a "critical engine" behind the growth of AI, with private debt market exposure jumping by $100bn to some $450bn just in the year to early 2025, far in excess of the funding from public credit. The bank expects "large amounts" of funding from this side of markets in the coming months, especially with the prospect of large inflows from retail investors and pension schemes, "sowing the seeds of an upside scenario" on the plus side, but also "increasing overheating risks" on the other. It is not just public equity markets that are running high levels of concentration risk in this scramble for AI exposure. It is all over private markets too. If specialist lenders start falling over and infecting the broader financial system, Altman's "that sucks" will be an understatement.

History lessons for our AI future

The massive investment in artificial intelligence infrastructure currently being made probably counts as the biggest and fastest rollout of a general purpose technology in history. This year and next, Google, Amazon, Microsoft and Meta alone will spend a staggering $750bn on data centres to power their AI models, with Morgan Stanley forecasting total global spending in this area will reach $3tn by 2029. But jittery investors are increasingly asking: what returns will this immense capital outlay generate? History suggests they're right to be nervous. There are few better scholars to put AI in a historical perspective than Carlota Perez, author of Technological Revolutions and Financial Capital: The Dynamics of Bubbles and Golden Ages. In her book, Perez identifies five great technological revolutions: the industrial revolution of the late 18th century; the steam, coal and railway revolution of the 1830s; the steel and heavy engineering revolution of the 1870s; the mass production age in the early 20th century; and the information technology revolution beginning in the 1970s. Perez sees AI as an extension of that fifth technological revolution. She also argues these revolutions follow a fairly predictable cycle. An initial installation phase results in lots of creative destruction and social disruption as industries and regions are upended. That is normally accompanied by over-investment, financial mania and stock market bubbles. Nevertheless, those bubbles are often productive, funding the building of vital infrastructure that enables the subsequent mass rollout of technology -- as railways or electricity grids are built -- and their broader economic benefits are realised. As for AI, we remain in that manic installation phase. That point was reinforced by a report from the Massachusetts Institute of Technology, which unsettled stock market investors this week. Researchers found that 95 per cent of the companies they surveyed were getting zero return from their investments in generative AI. Sam Altman, OpenAI's chief executive, was hardly reassuring when asked whether there was an AI bubble. "I do think some investors are likely to lose a lot of money," he replied. A bone-juddering crash -- or several crashes -- therefore seems likely before we reach any golden age of AI. "I have not seen a golden age happening without a crash," Perez breezily tells me. Even more cheerily, she adds that the bursting of the AI bubble could lead to yet bigger upheavals as capital markets are misfiring. They are now focusing more on speculative games, such as crypto, than productive investments, and global debt amounts to more than three times GDP. "This could also be a trigger for gigantic instability," she says. But it is worth investors considering how this technological revolution may differ from previous cycles. It is certainly the first revolution to be driven by software as much as hardware. That changes some financial dynamics as massive network effects come into play. Software companies can scale up quicker and go global overnight. OpenAI's ChatGPT is being used by 700mn people every week, less than three years after launch. But if digital globalisation increases opportunities, it also magnifies risks. Look at how China's cheaper DeepSeek AI model rattled investors in US tech stocks. Perhaps the most intriguing difference, though, is how far today's AI companies will themselves benefit from the financial gains they help unleash. The technology is accelerating advances in many areas: biotech, robotics and material science, for instance. AI companies could well exploit their technological advantage to become significant healthcare, drug discovery or autonomous car companies. To what extent can they morph into general purpose companies and capture the fruits of the golden age? One more important lesson, though, should be drawn from previous revolutions, Perez adds. To enter a golden age, civil society needs to shape the revolution to its own ends. So, for example, earlier politicians established antitrust agencies to tame over-mighty companies and created welfare states to soften labour market disruption. Perez argues that today's dysfunctional financial markets, the concentration of corporate power, the surge of populism and the threat of climate change have brought the world to a new turning point. How we should respond is the subject of her next book. However, as the historian AJP Taylor once wrote about the 1848 revolutions in Europe, countries can sometimes reach turning points -- and fail to turn.

Is the AI bubble about to burst - and send the stock market into freefall? | Phillip Inman

Shares in US tech stocks are falling but it would probably be unwise for fund managers to pull out There are growing fears of an imminent stock market crash - one that will transform from a dip to a dive when euphoric headlines about the wonders of artificial intelligence begin to wane. Shares in US tech stocks have fallen in recent weeks and the prospect is that a flood of negative numbers will become the norm before the month is out. It could be 2000 all over again, and just like the bursting of the dotcom bubble it may be ugly, with investors junking businesses that once looked good on paper but now resemble a huge liability. Jerome Powell, the Federal Reserve chair, is one of the policymakers tasked with keeping the wolf from the door. Speaking on Friday at the annual Jackson Hole gathering of central bank governors in Wyoming, he tried to calm nerves. He said the Fed was concerned about rising inflation, while at the same time willing to help an economy dogged by uncertainty induced by Donald Trump and a global economic slowdown. With stagflation a genuine prospect - as the US economy slows and inflation remains high - Powell gave stock markets a sign that interest rates will fall, easing the pressure on indebted companies. Stock markets are on Powell's radar even more than usual now that so many personal pensions in the US are directly invested in listed companies. And, more particularly, tech stocks making huge AI investments that have yet to make a dollar of profit. A recent Massachusetts Institute of Technology report revealed that 95% of companies investing in generative AI have yet to see any financial returns. This revelation came after Sam Altman, the boss of the ChatGPT owner OpenAI, warned that some company valuations were "insane". Ipek Ozkardeskaya, a senior analyst at the currency trading firm Swissquote, said: "The [Altman] comments may have been a wake-up call for investors, sparking a sharp pullback in high-flying names." At the beginning of the week, the share price of the data mining and spyware firm Palantir, which has billions of dollars worth of US government contracts, plunged almost 10%. The AI chip maker Nvidia fell more than 3%, while other AI-linked stocks such as Arm, Oracle and AMD also lost ground. Most pension funds will be invested in these tech companies, alongside more longstanding names such as Amazon, Microsoft, Alphabet (Google) and Meta (Facebook). Should fund managers pull out? That would probably be unwise. The scale of investment in AI by the likes of Google and Meta is vast and while the technology's potential is the subject of much speculation, white-collar workers are becoming more familiar with the supposed benefits with every passing day. Corporate messages asking them to use AI for presentations, report writing and research are a daily occurrence (accompanied, of course, with hollow assurances that job cuts are not being considered). Microsoft's Copilot and the many other "helpful" AI tools on offer are becoming embedded in office life and beginning to perform large numbers of low-level tasks. If this trend takes off, and in many parts of the economy it already has, there is a soft landing waiting for the tech industry even as some of the flakier, more speculative businesses are weeded out and fail. If anything, a downturn helps the big firms pick up new technological breakthroughs from the wreckage and on the cheap. Palantir's price-to-earnings ratio is north of 500 when most investors would find themselves getting panicky at anything above 50. Nvidia has a price-to-earnings ratio of 56. The Palantir/Nvidia ratios could decline as their share price falls more into line with a reasonable prospect of likely earnings, but the businesses are not going to go bust, even in the most severe equity market storms. Trump is another significant supporter, clearing the path for AI to make even deeper inroads into corporate life. His support for cryptocurrencies, including his own, and deregulated social media platforms, including his own, are indications of where his sympathies lie. AI is probably going to be bad for humanity, given that politicians and regulators are light years behind the tycoons and tech magnates backing AI, many of whom see it as a new way to disempower and dominate workers. However, as an investor, AI isn't going away, crash or no crash.

We may be facing the dotcom bubble 2.0

But it also marked the exact zenith of the TMT (technology, media and telecoms) bubble. Before the year was out, more or less everyone who was at that meeting was nursing crippling losses. Few things better characterised the "irrational exuberance" of the age than the auctions of 3G mobile phone licences that took place around that time. Cleverly structured using game theory to maximise proceeds for the presiding governments, they were almost guaranteed to end badly - none more so than in Britain where, in the scramble for spectrum, newcomers and incumbents alike ended up massively overpaying. It was great for the Treasury, which raised an astonishing £22.5bn from the five licences on offer. What Reeves, with an estimated hole of up to £50bn in the public finances to fill, would give for something similar. Yet it was close to catastrophic for the winners. British Telecom and the like were lucky to survive the following bust. The accompanying dotcom bubble also ended in tears, as all stock market manias invariably do. But this time is different, right? So the high priests of the latest mania - generative artificial intelligence and its progeny, superintelligence - would have you believe. Hundreds of billions of dollars are being poured into their pursuit. If the claims being made about the transformative powers of the new technologies sound familiar, that's because we have indeed heard it all before. Exactly the same was said of the internet. The predictions were not wrong in some respects. The resulting boom in communications infrastructure proved genuinely transformative. But the services it spawned also very rapidly became commoditised and the outright winners were confined to a relatively small number of very large corporations. Along the way, there were many casualties. This is the pattern for virtually all emerging technological revolutions. By way of example, take the birth of the automotive industry. At one stage there were more than 2,000 separate car manufacturers in the US, but only a relatively small number of them lasted the distance. You would have lost money on the vast majority of these ventures - and even if you had invested in all of them, the gains from the winners would not have compensated for the losses from those that fell by the wayside. But even among those with a deep understanding of the industries involved, successfully identifying the eventual winners from rapid industrial change is virtually impossible, so the tendency is to invest in the entire field in the hope that some come good. The result is a self-feeding bubble in which almost any Tom, Dick or Harry can ride the zeitgeist and attract capital. We saw it during the dotcom mania and we are seeing it again today. As an amusing aside, the stock market guru Warren Buffett once observed that it is much easier to spot the losers from any transformative technology than it is the winners. One obvious loser from the automotive revolution was the horse and cart. There were 20 million horses in the US at the beginning of the 20th century. Today, there are fewer than four million and very few of those are kept for pulling wagons. With AI, it is similarly possible to see which industries and professions are likely to be destroyed by it, with anything to do with conventional data analytics being only the most obvious. As for likely winners, it is admittedly a little bit different this time, in that the companies pouring the big bucks into AI are substantially the same as the winners from the preceding IT revolution - Microsoft, Meta, Alphabet, Apple, Amazon and so on. Even so, the potential for extreme misallocation of capital remains much the same. All these companies are driven by a winner-takes-all mentality. Few of them foresee a world in which they can all succeed. The result is a kind of arms race, or a game of poker in which the stakes are constantly raised. All the main players are dramatically ramping up their capital spending with plans for ever bigger, hyperscale data centres. Microsoft has earmarked $80bn (£59bn) for data centres to train AI models and Meta's year-to-date capital spending has already topped $30bn. Going all in, analysts expect Meta's Mark Zuckerberg to sink a further $100bn into the AI gold rush next financial year.

The next 'AI winter' is coming

Artificial Intelligence turns 70 years old next year, but the birthday celebrations may be muted. The stock market is braced for a crash: a rerun of the dramatic collapse in confidence that ended the dotcom bubble overnight. The market hasn't fallen so far or so fast yet, but the three assumptions that have underpinned the current AI boom now look very dubious. The first is that generative AI will yield productivity and efficiency gains for businesses which will be reflected across the economy. The second, that a handful of clever AI companies will benefit greatly from this. And the third is that any setbacks are temporary, and AI will just keep getting better and better. Firstly, we can now see that the business benefits are not evident. MIT reported this week that most AI investments - some 95 per cent - yield no net return. This is consistent with other recent studies into "agentic AI". IBM found only one in four AI projects show a net RoI, and just 16 per cent merit full deployment across the enterprise. Reliability is a mortal weakness. The chatty parrot of generative AI is no more than a linguistic trickster that is too flaky to be used to automate serious business functions. Lloyds has even begun offering insurance for "unforeseen performance issues" arising from "AI-driven products and operations". While a fear of missing out, or "FOMO" - as IBM suggests - is the main reason for the AI boom, more people can now see that the Emperor has no clothes. As I wrote in my column two weeks ago, businesses know prices will fall, and can choose to wait. Only fools will rush into AI now. The second reason why a correction is overdue is the fundamental assumption that huge capital investments are necessary to maintain progress. Billions are being spent on data centres and expensive graphics chips. But in reality, there is no competitive advantage to this, or in the jargon, there's no "moat" for a business to defend. When Victorian railway companies like the LNER built lines and locomotives at great expense, they could control the trade conducted on them. Later, patents and trade secrets allowed pharmaceutical and software companies to justify huge R&D expenditures, although both had to grapple with piracy. But today every generative AI feature or technique is copied by competitors instantly, within days. Open source models are built specifically to be copied by other people. So a billion dollars of capital expenditure training an AI model suddenly looks like a terrible and stupid waste of capital. Meanwhile the Chinese are focusing on performance, and producing AI models that run well at a hundredth of the expense. We have yet to see this reality reflected in the equity market. Here in the West, we may have backed the wrong kind of AI. And OpenAI's launch of GPT-5 put paid to the notion that if you spend lots more money, you get much better results. Then there are externalities that generative AI creates, the vast social and economic costs that others must pay: from the devaluation of education to the destruction of markets for creative work and news, to drowning us in low quality 'slop'. Deploying AI also opens up huge security holes that enterprises are frantic to fix. And worst of all, unlike in the dotcom crash, there may be no latent unmet demand to save the day once the smoke has cleared. After the 2000 crash, it was still the case that very few people had broadband at home, and the latent demand for consumer services from movies to social media, and online banking for example, was still untapped. There has been a huge expansion in digital markets since. It's far from clear that anything similar will happen with AI. The underlying machine learning technology should continue to serve niches well, in fields such as data analysis, prototyping and language services. The AI sceptic Gary Marcus envisages a sector generating $50-100bn from these. That's not trivial, but for perspective, that's one tenth of the size of the global advertising market today. The FTSE100 took fifteen years to recover and savers saw real pain in the shape of diminished savings and pensions pots. For most of AI's seventy year history, the field has been shunned and ignored, in what AI historians call "AI winters". The coming one may be the coldest yet.