It has been a bumper quarter for big tech. Microsoft and Meta each added half a trillion dollars in market value to their stocks after reporting strong earnings, and are now worth twice as much as the entire FTSE 100 index combined. The previous week, Google’s parent company, Alphabet, reported a 14% increase in revenue.
These results are something of a Rorschach test. For some, they suggest that big tech’s huge investments in AI technologies may be starting to reap rewards. Meta has said its capital spending this year will be between $66bn (£48bn) and $72bn, Alphabet is raising its 2025 capex forecast to $85bn and Microsoft’s spending is above analysts’ expectations. “The bumper week has clearly done a lot to reassure those who were nervous that capital has been allocated in a very free fashion,” said Russ Mould, an analyst at investment platform AJ Bell.
Butanalysts look at these vast capital expenditures and hear echoes of the dotcom bubble – a moment defined by heavy spending and speculation. “The difference between the 90s IT bubble and the AI bubble today is the top 10 companies in the S&P 500 are more overvalued than they were in the 90s,” Apollo’s chief economist warned in a recent note.
Other gauges of market giddiness are also rising. Goldman Sachs’ speculation indicator – which measures trading activity in penny stocks, unprofitable companies and very highly valued stocks – has reached historic extremes.
Data analysis by The Observer shows that forward revenue multiples (FRM) – the ratio of a company’s valuation to its revenue – for the world’s major generative AI firms, are much larger than for other industries. OpenAI – valued at $300bn with $12bn of revenue – has an FRM of 25, while Anthropic’s is 42.5. The Canadian startup Cohere is in talks to raise up to $500m at a $6.3bn valuation, which implies an FRM of 60. These figures, which are based on self-reported results, are markedly higher than average FRM in other industries, which tend to range in value from single digits to high teens. According to data from KPMG, industries with low estimated FRM ratios include banking, passenger airlines, automobile manufacturing and oil and gas. On the other end of the spectrum are real estate property investment and media – though their FRM ratios are still only roughly half that of the major AI companies.
Revenues in generative AI businesses are starting to increase, but progress is slow. According to a report in The Information, OpenAI roughly doubled its revenue in the first seven months of the year to $12bn. But it has burned through $8bn in cash since 2025, up $1bn from projections earlier this year began.
Since the launch of ChatGPT in November 2022, the AI industry has been defined by hype. Industry leaders, including at OpenAI and Anthropic, have lofty goals of achieving so-called artificial general intelligence, a term that vaguely refers to AI that can perform at human level. At Meta, Mark Zuckerberg is now hailing a new era of superintelligence, writing in a note this week that it “has the potential to begin a new era of personal empowerment where people will have greater agency to improve the world in the directions they choose”.
“ Given that capital investments are still going up and that companies are being rewarded for the more they spend, at some stage, they will need to prove they can turn that into revenues, into profits and into cash,” Mould said. “This is true no matter how ambitious their goals.”
Most big tech firms don’t break down how much money they make from artificial intelligence specifically. But it’s clear that not much of this quarter’s earnings can be attributed directly to AI – particularly not generative AI, which has accounted for the bulk of tech investments and includes large language models like ChatGPT.
Meta, for example, attributes its strong growth this quarter to AI recommendation and ranking systems – not to chatbots. Microsoft’s revenue increases were driven by its cloud services – infrastructure for AI, but not the models themselves – which increased by 23% year on year.
“There is a long way to go and valuations offer little downside protection should something unexpected go wrong or the timing not quite pan out as it has hoped,” Mould said.
Photograph by Bloomberg/Getty