The big question for 2026: has AI caused a stock market bubble?

Many comparisons have been made between the artificial intelligence (AI)-driven rally and the dot-com bust of the early 2000s. Like the internet, AI will be transformative across every part of our lives. But there are important distinctions that make AI less frothy than the dot-com bubble. Moreover, there are deeper long-term challenges – most notably the energy required for the build out - that could hold back its potential for 10 to 20 years. 

• Our Services: SIPP Account | Stocks & Shares ISA | See all Investment Accounts

The top five AI hyperscalers - Microsoft Corp (NASDAQ:MSFT), Amazon.com Inc (NASDAQ:AMZN), Alphabet Inc Class A (NASDAQ:GOOGL), Meta Platforms Inc Class A (NASDAQ:META) and Oracle Corp (NYSE:ORCL) - are profitable, established companies that control the compute, data and distribution rails on which AI depends, outside China.

In sharp contrast, many of the late 1990s internet firms had crazy business models such as Kozmo.com, which promised free delivery of videos and food within the hour, and went bust in a matter of months.

Valuations and debt levels are not extreme

Valuations in large tech are also relatively modest. Today’s mega-tech stocks are on an average price/earnings (P/E) ratio of 26x, compared with the Nasdaq at 60x-70x in March 2000. At the turn of the millennium, individual stocks such as AOL and Cisco Systems Inc (NASDAQ:CSCO) hit P/Es of around 200x and 135x respectively. Of course, Tesla Inc (NASDAQ:TSLA) is a wild card up there in the hundreds, and NVIDIA Corp (NASDAQ:NVDA) is on 44x, but the market as a whole is not in the giddy territory of the world’s great stock market bubbles.

“Today’s tech giants stand on solid financial ground, supported by diversified revenue streams, mature business models, and fundamentals-driven valuations,” says David Aujla, multi-asset fund manager at Invesco. “For example, their combined net income margins (a company’s profit as a percentage of its revenue) average around 30%, nearly twice that of dot-com firms in the early 2000s.” 

The risk of underdevelopment in this environment is severe and existential, however. Consequently, the major hyperscalers ploughed over $200 billion (£149 billion) into data centers and associated hardware in the first half of 2025 alone, together with forging innumerable circular partnerships to build infrastructure, which is also causing investor concern. Accounting for inflation, this figure is 10 times more investment than in the dot-com era but four of the five main hyperscalers can fund their forecast capital expenditure entirely out of cash, with only Oracle the major exception.

• Investment outlook: expert opinion, analysis and ideas
• Funds beating global markets in good times and bad
• Experts name fund and trust opportunities at start of 2026

“The absence of broad-based debt and the robust balance sheets of leading players suggest the risks of a dramatic burst are currently lower than in past cycles,” says Oliver Jones, head of asset allocation at Rathbones. “That said, the sector’s reliance on the investment decisions of a few key firms introduces its own vulnerabilities. We need to monitor what’s happening to the hyperscalers’ cash flows, the guidance they provide on their investment plans, and how they plan to fund their investment.”

Concentration a key concern

The concentration of the market into a few huge firms is indeed a concern. Jamie Mills O’Brien, a fund manager at Aberdeen, says that OpenAI’s CEO Sam Altman’s recent assertion that it has about $1.4 trillion in data centre commitments over the next eight years “is highly unrealistic, (and has) led to greater focus on the durability of a demand cycle led by a three-year old start-up, and a repricing of AI-linked risk.”

He continues: “That uncertainty has led to underperformance of certain pockets with questionable business models, debt issues and companies with perceived structural threats to their top of the funnel demand, for example Duolingo Inc (NASDAQ:DUOL), as well as companies where rising operating expenses seems to have come alongside open-ended duration of that spend.”

However, the productivity gains across many industries are already astonishing. For example, Amazon’s AI-powered search algorithm is much more effective at driving sales than competitors’ search functions with 42% of customers likely to click through to a purchase, compared with an industry standard in the teens. Take-up is also rapid – most obviously after three years, OpenAI’s Chat GPT boasts around 800 million weekly users, a milestone that took the internet 13 years to achieve.

• Why this risk will become a bigger headache for markets in 2026
• Did it pay off to reduce US concentration risk in 2025?
• Racy specialist funds: more room to run?

“Most of the focus and debate is on spend (AI capex) – whereas the focus should actually be on the capability enhancements generative AI tools are having at companies,” says Ben James, investment specialist at Baillie Gifford. “Companies are doing things now that they couldn’t do before. If people are looking for the return on investment (ROI) on AI they should look to companies like Shopify Inc Registered Shs -A- Subord Vtg (NASDAQ:SHOP), which continues to grow rapidly while not increasing headcount, or Lemonade Inc Ordinary Shares (NYSE:LMND), which is growing premiums without increasing headcount.”

Unsurprisingly, companies with strong fundamentals performed better in 2025, most obviously in the memory semiconductor space, where AI-linked demand alongside supply shortages has boosted earnings estimates.

Which stocks will be the winners and losers?

Companies associated with future data-centre demand appear to be the most overheated segment in valuation terms. Over 30 of the 100 largest re‑ratings in the index since May 2023 have been companies directly tied to the AI data-centre build‑out, such as power, storage, cooling and construction.

James notes: “The price-to-sales ratios of those companies have increased by about 130% on average, compared to about 18% for the rest of the index. Companies like Western Digital Corp (NASDAQ:WDC) (price-to-sales +432%) and Seagate Technology Holdings (NASDAQ:STX) (P/S +230%) in data storage, as well as Vistra Corp (NYSE:VST) (P/S +331%) and Constellation Energy Corp (NASDAQ:CEG) (P/S +261%) in electricity generation, are among the most stretched.”

He thinks it is helpful to break down the elements of the AI value chain. Baillie Gifford US Growth Ord (LSE:USA) trust has exposure to Figma Inc Class A (NYSE:FIG), Tempus AI Inc Class A common stock (NASDAQ:TEM), Duolingo, AppLovin Corp Ordinary Shares - Class A (NASDAQ:APP), Shopify, and Meta in the application layer; Tesla and Samsara Inc Ordinary Shares - Class A (NYSE:IOT) in the Embodied Intelligence category; Cloudflare Inc (NYSE:NET), Amazon and Circle Internet Group Inc Ordinary Shares - Class A (NYSE:CRCL) in Agentic Infrastructure; Datadog Inc Class A (NASDAQ:DDOG) and Snowflake Inc Ordinary Shares (NYSE:SNOW) in Data; and Nvidia and AWS in Computing.

• Ian Cowie: two of my most exciting investment trust holdings
• Two UK funds I'd own in 2026
• The shares attracting fund managers at start of 2026

Investors will need patience, however, and ultimately a great deal of good luck with their stock picking. The original established players in tech revolutions such as the railway and automobile booms of the 1920s do not always turn out to be the winners, and may fail or be acquired for a pittance. Although today’s AI giants are much more sophisticated, it is likely that some relatively unknown companies will find a disruptive competitive advantage.

Invesco’s Aujla points out that historically the technology sector has seen new entrants challenge incumbents and eventually dominate, especially in the US. He adds: “For perspective, only about 10% of Fortune 500 companies have remained on the list since 1955, and just one company from the 2000 list was still in the top 10 by 2020. This offers a clear reminder of why diversification matters. Markets move in cycles, and what wins today may falter tomorrow.”

As well as watching out for any increase in debt funding, a string of overhyped IPOs like the ones in previous speculative peaks can indicate a market that has got ahead of itself. The most anticipated tech IPOs are Databricks (cloud-based data analytics); Stripe (payments infrastructure); Chime (online banking); Revolut (digital banking),; Discord (communications platform) SpaceX (aerospace/defense) and, of course, OpenAI, which could target a valuation of $1 trillion as early as late 2026, although Altman remains cautious about public market restrictions. 

Even if you are not impressed by the underlying company’s story, there will inevitably be short-term trading opportunities in any new issues that are oversubscribed. 

• IPO market outlook for UK in 2026
• Sign up to our free newsletter for investment ideas, latest news and award-winning analysis
• The investments trusts that beat Nvidia in 2025

“Episodes of exuberance will inevitably lead to pockets of excess - funding of weak start-ups, over-investment in unprofitable projects, and disruption across mature industries,” says Kate Lakin, portfolio manager at Putnam Investments. “As with prior transformative cycles, this wave will create tremendous investment opportunities…alongside significant pitfalls, as many early entrants are likely to fall short of lofty goals and many entrenched companies will face disruption.”

“The AI build-out is generating massive opportunity, but it is also introducing earnings volatility across the technology sector. Depending on where a company sits in the supply chain, AI demand can either amplify upside or expose structural downside.

“We continue to own leading large-cap infrastructure beneficiaries that have captured much of the early value creation…and are equally focused on the broadening of the opportunity set - both within the physical layers of AI - memory, storage, and energy - and across the deployment layer - software, services, and gaming - where we expect the next leg of value creation to emerge.”

However, the AI revolution will be markedly different in terms of capacity, which in most booms tends to be built too quickly. This time, firms are using their entire infrastructure as soon as it gets up and running. Then, as they deploy more advanced chips, the pressure on energy supply is intensifying.

According to S&P Global Energy, an estimated 44GW of additional energy capacity will be required by the top five US players to run their data centres by 2028, but only about 25GW is forecast to come online, creating a shortfall of over 40% of the power needed.