The AI boom is driving stock exchanges – but monetization, power shortages and regulation could become problems.
• Hyperscalers are investing hundreds of billions in AI infrastructure
• Analysts see some negative AI returns by 2030
• Electricity grid, regulation and competition from China are increasing the pressure
The AI euphoria has been running at full speed for years and has reached hype status. On the stock markets, industry representatives such as NVIDIA, Microsoft, Alphabet and Co. have pushed the bar for valuations ever higher. But analysts, investors and industry experts are increasingly asking the crucial question: When will the bill come? Four scenarios could herald the end of the boom, or at least trigger a painful downturn.
Ratings with room for improvement
For the Magnificent Seven as a whole, earnings growth of around 18 percent is expected in 2026, according to Bloomberg Intelligence, the slowest since 2022 and barely better than the 13 percent forecast for the rest of the S&P 500 companies. At the same time, the massive increase in capex spending is putting pressure on free cash flow: According to forecasts, the combined free cash flow of Amazon, Microsoft, Alphabet and Meta could fall from around $205 billion in 2025 to around $94 billion in 2026. The seven companies now make up around a third of the market capitalization of the S&P 500, making the index effectively an AI betting index.
Strategy analyst Joachim Klement from Panmure Liberum estimates the current AI investment boom to be around 60 percent larger in nominal terms than the dot-com bubble of the late 1990s, and the technology share of US economic growth has reached a concentration that has not been seen since the end of the Second World War. The key structural difference from back then: Today’s hyperscalers are highly profitable and finance the majority of their investments from their own cash flow, not through debt or IPOs like many dot-com companies. This reduces the risk of a crash, but does not eliminate it. According to Panmure Liberium, cheap valuations do not protect investors because historically there is no statistical connection between current sector valuations and expected price declines.
ROI gap as a systemic problem
The actual risk therefore lies not in the prices of the AI giants on the stock markets, but rather in the question of what ultimately justifies the billion-dollar AI investments. According to an assessment by Deutsche Bank, 2026 will be the most difficult year yet for the industry, marked by a triple test of disillusionment, misalignment and mistrust. A PwC study comes to a similar conclusion: Only 12 percent of the CEOs surveyed stated that AI has brought both cost and sales effects.
The most sensational figures so far are provided by a model calculation by Panmure Liberum, which was published in the Financial Times: Under the deliberately simplifying assumption that operating costs for AI systems are effectively zero, i.e. in the best conceivable scenario, the forecast return on investment for the period 2025 to 2030 remains negative for most hyperscalers. The model shows Microsoft with minus 9 percent, Google with minus 15 percent, Meta with minus 28 percent, Oracle even with minus 35 percent. Only Amazon scores slightly positively.
However, the calculation has a well-known methodological limitation: it assumes that the entire infrastructure must be attributed exclusively to new AI sales. In fact, a significant portion of the investments replace ongoing server infrastructure, secure existing core businesses and are amortized over several years. Against this background, the figure of two to five trillion dollars in additional sales required that Panmure Liberum derives from the model is a theoretical extreme calculation, not a realistic minimum value.
However, the monetization gap remains significant: Direct AI services such as software subscriptions and AI APIs only generated around $25 billion in revenue in 2025, excluding the broader cloud infrastructure revenues that are driven by AI demand. If investment returns consistently fall short of expectations, capex growth is expected to slow significantly in the coming years.
Energy and infrastructure as bottleneck factors
Amazon, Microsoft, Alphabet and Meta together are planning capital expenditures in the three-digit billion range for 2026; the current quarterly guidance of the five largest hyperscalers including Oracle totals over 600 billion US dollars for the current year. The problem: The physical infrastructure is not keeping up. Bloomberg reports that nearly half of all AI data centers in the US planned for 2026 will be delayed or canceled altogether. The central bottleneck is not chips or capital, but rather electrical components: transformers, switchgear and batteries, whose supply chains can no longer meet demand. Since mid-2024, according to research firm Data Center Watch, projects worth over $64 billion have been stopped or massively delayed due to local opposition, with $18 billion canceled completely and $46 billion postponed. Alphabet boss Sundar Pichai has already admitted that cloud sales would have been higher if his company had been able to meet demand.
Added to this is growing local resistance: according to industry observers, it unusually unites environmental activists and budget conservatives, linked by high energy consumption, the low number of permanent jobs and a lack of transparency in planning processes.
Regulation, geopolitics and growing political headwinds
The fourth risk is less quantifiable but hardly less relevant. AI systems are increasingly becoming the focus of legislators in the EU, the USA and Asia – the first political voices are already calling for a tax on AI.
At the same time, social pressure is growing: A survey of 825 executives by the consulting firm Mercer came to the conclusion that 99 percent of the CEOs surveyed are planning AI-related job cuts in the next two years. The proportion of company managers who specifically plan to eliminate entry-level positions jumped from 17 percent in 2025 to 43 percent in the current year.
Waves of layoffs at companies such as Block, PayPal, Intuit and HSBC, but also at Meta and Oracle, have further fueled the debate.
There is also the geopolitical dimension: The DeepSeek shock of January 2025 showed that Chinese providers such as Kimi, Qwen or MiniMax can develop competitive models with prices that are up to 90 percent lower than US market leaders. If these providers gain significant market share, the circular system of AI financing would come under pressure: According to Morgan Stanley, OpenAI and Anthropic account for around half of the demand among hyperscalers. If this demand were to disappear, it would not only affect individual providers, but also the entire investment logic of the sector.
What investors should keep an eye on
The key indicator for all four risks is the same: whether the hyperscalers will adjust their investment plans upwards again for 2027 or row back for the first time. With a view to the figures from Amazon, Microsoft, Alphabet and Meta in the coming months, three metrics deserve particular attention: The ratio of cloud revenue growth to Capex spending, which shows whether investments are starting to be reflected in revenue. The development of free cash flows, which are likely to come under further pressure if investment pressure continues. And the hyperscalers’ guidance on AI-specific revenues, which have so far often been blurred into the broader cloud numbers. Anyone who follows these three data points quarterly will get an idea earlier than the market as to whether the AI bet is working or whether the bill is still outstanding.
Claudia Stephan, editorial team at finanzen.net
This text is for informational purposes only and does not constitute an investment recommendation. finanzen.net GmbH excludes any claims for recourse.
By the way: Alphabet A (ex Google) and other US stocks can even be traded on finanzen.net ZERO until 11 p.m. (without order fees, plus spreads). Open a depot now for free and secure a new customer bonus!
Selected leverage products on Alphabet A (ex Google)
With knock-outs, speculative investors can participate disproportionately in price movements. Simply select the lever you want and we will show you suitable open-end products on Alphabet A (ex Google)
The leverage must be between 2 and 20
Risk warning: Suffer on average 7 out of 10 retail investors Losses when trading with turbo certificates. Turbo certificates are high-risk products and not suitable for long-term investment strategies. Risk warning: Suffer on average 7 out of 10 retail investors Losses when trading with turbo certificates. Turbo certificates are high-risk products and not suitable for long-term investment strategies.
Advertising