Invisible Bottlenecks

Stock markets have gained significantly since the beginning of the year. The leading American index, the S&P 500, reached new record highs several times, driven primarily by large technology corporations around the topic of artificial intelligence.

Anyone who wants to understand this development should not be guided by headlines, but should look at the facts. Because, contrary to what is often assumed, this development is not merely driven by euphoria, but by strongly rising profits. In the first quarter, the profits of S&P 500 companies increased by around 21 percent. A large part of this growth comes from companies related to artificial intelligence. Because profits rose faster than share prices, US stocks are even valued slightly more cheaply than before, and not more expensively, as many assume. In terms of profits, the markets are therefore not overpriced, but it remains to be seen whether profit growth can be maintained at these levels.

It is also fascinating to see how calmly the markets are currently dealing with the challenging macroeconomic environment: inflation, budget deficits, tariffs, and higher interest rates are hardly a burden. There was finally some relief on one of these fronts recently. In mid-June, President Trump announced a tentative agreement with Iran to reopen the Strait of Hormuz, through which about one-fifth of the world's oil flows. The oil price, which had previously risen sharply, subsequently fell significantly.

But while many are looking at oil, a more fundamental development is brewing in another commodity. In one that powers our modern world: electricity.

Electricity is becoming the new oil

For decades, electricity in industrialized countries was a topic for engineers, not investors. More efficient appliances offset the additional demand; in Europe and the US, demand stagnated for over twenty years. This era is over, and the world's largest technology corporations are responsible for it. For artificial intelligence, Microsoft, Amazon, Alphabet, Meta, and Oracle, known in technical jargon as "hyperscalers," are building massive data centers and expanding their capacities at a breathtaking pace.

These capacities are primarily needed for training new AI models, a process that consumes enormous amounts of energy. The AI pioneer Leopold Aschenbrenner describes in his highly acclaimed essay written back in 2024 how these corporations have entered a race for ever-larger computing clusters whose electricity requirements dwarf any previous scale. The cluster on which OpenAI trained the GPT-4 model in 2022 required around 10 MW, as much as about ten thousand households, and cost close to 500 million dollars. Aschenbrenner wrote two years ago that the performance and cost of these facilities decuple approximately every two years. For 2026, this already meant a cluster of around 1 GW, with costs in the double-digit billions.

How close this forecast is to reality is already evident today. In June 2026, Microsoft and Chevron announced a data center in Pecos, Texas, whose electricity demand is expected to reach around 2.67 GW when fully developed. It will be supplied by its own gas power plant with seven large gas turbines, completely independent of the public electricity grid. For context: one gigawatt corresponds approximately to the output of a nuclear power plant. A large gas turbine delivers around 400 MW, while a modern wind turbine only delivers about 6 MW. This data center therefore devours almost as much electricity as three nuclear power plants, and that around the clock.

And according to Aschenbrenner, the development is only just beginning. For 2028, he expects clusters of 10 GW and costs in the hundreds of billions, which in terms of energy technology is as much as a smaller US state consumes. By 2030, he even considers a single cluster of 100 GW possible, with costs of over a thousand billion dollars, which alone would account for more than twenty percent of today's total US electricity production.

The actual bottleneck

Electricity demand is growing massively faster than supply. Since 1985, US electricity production has only risen by about 60 percent. This sluggish supply side is now met by a demand that is advancing into completely new dimensions due to artificial intelligence. According to Aschenbrenner, investments in AI infrastructure are expected to multiply from around 150 billion dollars in 2024 to about 8,000 billion dollars by 2030, which is more than a fifty-fold increase. For 2026, he estimated around 500 billion dollars. In fact, the hyperscalers' announced investment volume for 2026 is now around 725 billion dollars, which is already 45 percent above his estimate. These vast sums end up as revenue for suppliers. Their surging profits are one of the drivers behind the profit growth of the indices mentioned at the beginning.

Where the energy is supposed to come from is therefore not a simple question. Solar and wind are cheap, but do not deliver around the clock. In the short term, natural gas in particular steps in, and nuclear power is also experiencing a comeback. Increasingly, operators are also relying on fuel cells that can be installed directly on site. All solutions share one goal: to bypass the years-long wait for a grid connection by generating the electricity supply directly next to the data center, as in the mentioned project in Texas.

Our Conclusion

Electricity is becoming a strategic resource, much like oil in the 20th century. Already, Gulf states are attracting hyperscalers with cheap, virtually unlimited energy. At the same time, voices like Aschenbrenner warn against leaving this key infrastructure to authoritarian regimes. Because whoever controls the facilities on which the most advanced AI is created possesses a strategic power lever of the highest order. What matters is not only who builds the best models, but who owns the energy, data centers, and control.

As always, we thank you for the trust placed in us!