527 billion US dollars in capital expenditure. That is the sum Goldman Sachs names for 2026 as an estimate for the investment spending of major AI hyperscalers. Added to this are data centers, power lines, fiber optics, cooling, chips, software, security infrastructure and skilled workers. AI has long since ceased to be a simple buzzword that affects only a few companies and providers. It is being built, ordered, financed and consumed almost everywhere. It is a real trend that could still end up costing the stock market dearly.
The railway changed economic regions. The internet reshaped entire industries. Fiber optics, solar power, electric mobility and crypto have triggered further developments of our time. In all these hot phases, many stocks were bought at prices that later did not match the profits of the respective companies. This mistake could repeat itself with AI.
Billions are flowing into concrete, chips and energy
The current AI cycle now affects numerous companies and industries, because the large language models of commercial providers and also the recently much improved open-source models need large data centers, specialized chips, storage, network technology, power contracts, cooling and maintenance. The International Energy Agency estimated global electricity consumption by data centers at around 415 terawatt-hours as early as 2024. Stanford HAI reported private AI investment of 285.9 billion US dollars in the United States alone for 2025. AI consumes real resources that companies are paying for today because they expect productivity, customer loyalty, automation and new revenues from it later.
On the stock market, this development quickly turns into a simple calculation: high investment, a lot of growth and the future always lead to rising prices. But it is not quite that simple. The more capital flows into data centers and chips, the more profit has to come back later. A billion-dollar program is not an end in itself. For 2025 alone, OpenAI, the developers behind ChatGPT, is expected to spend 50 billion US dollars on computing power alone, while the actual product generates only 20 billion US dollars in revenue. One day, the market will always ask whether the spending pays off.
In a gold rush, the suppliers earn first
In many major trends, the early earners are not the later winners, but the suppliers. In the gold rush, shovels, transport routes and supplies were often profitable faster than the actual search for gold. A similar effect can be seen in AI. Semiconductor manufacturers, network technology, cloud providers, data center operators, energy suppliers and cooling are early in the money flow. That explains the strong price movements of many companies in this environment, for example NVIDIA. No models without chips. No computing power without electricity. No scaling without cloud. Demand is enormous and bottlenecks are driving prices and margins.
But even there, the stock market remains merciless. High demand attracts competition. Customers negotiate harder as soon as bottlenecks become smaller. Investment cycles do not run in one direction forever. If everyone is building capacity at the same time today, oversupply can arise tomorrow. Then it is no longer enough that AI is a major topic. Then every company has to show whether it has lasting pricing power.
Good companies can be bad stocks
A company can work excellently and still become a bad stock. That is one of the most important stock market lessons. Anyone who buys too expensively can be disappointed despite a good business model. This danger is particularly great with AI because the stories are so convincing. Software becomes faster. Programmers get better tools. Advertising becomes more precise. Call centers are automated. Industrial processes become more measurable. Data is analyzed more cheaply. All of that may be true and still not be enough if the share price already assumes years of perfect growth.
The higher the valuation, the fewer mistakes the market forgives. A delay in expanding data centers, rising energy costs, export restrictions, new competition, price pressure for models or a slower investment cycle are then enough to hit the share price. Not because the company has become bad, but because expectations were too high. That is the point that is often lost in AI debates. Investors do not buy the importance of a technology. They buy a stake in a company at a concrete price.
AI is already in the world portfolio
Many private investors still treat AI like an add-on topic. An AI ETF here, a semiconductor fund there, perhaps a few individual stocks from the United States. That may look like a targeted future allocation, but the topic is often already long present in the portfolio. Large technology and platform companies have high weights in global indices. Anyone holding a world ETF often already owns a lot of AI risk through Microsoft, NVIDIA, Alphabet, Amazon*, Meta, Apple or semiconductor stocks.
A world ETF plus a Nasdaq ETF plus an AI ETF can ultimately mainly shift the weight toward the same companies, the same valuations and the same expectations. In good phases, this works excellently. In weak phases, however, it becomes clear whether the portfolio has really become broader or whether the same story has simply been told several times. Anyone who buys additional AI exposure therefore needs a clear reason; otherwise another concentration risk emerges.
AI is not a free pass
AI is a real investment cycle. Data centers are being built, energy is needed, chips are becoming scarce, companies are deploying billions. All of this argues against the simple claim that the boom is only hot air. The more plausibly a story can be told, the easier it becomes to accept prices that leave hardly any room for disappointment. Then mere growth is no longer enough.
Conclusion: AI deserves attention, but not a free pass. Good technology does not replace a good price. A megatrend does not become dangerous for investors when it sounds obviously exaggerated. It becomes dangerous when almost all the arguments are right and the market still demands more…
