AI boom: Europe's data centers are not keeping pace with demand growth

The boom in machine learning (ML) and artificial intelligence (AI) is leading to a sharp rise in demand for server capacity and therefore for space in data centers. In the USA, this is triggering enormous growth in data centers. In Europe, however, comparatively, little is happening. "New data centers are becoming increasingly difficult to build", reports real estate specialist CBRE, "providers are finding it difficult to secure power supplies and land." The vacancy rate in the top 5 cities has fallen below ten percent for the first time. Colocation is particularly hard to find in Frankfurt am Main.

This is shown by CBRE surveys. The company monitors the situation in the 15 most important server locations in Europe; the top 5 are Frankfurt, London, Amsterdam, Paris and Dublin (FLAPD, primary market). The second cohort comprises Berlin, Brussels, Madrid, Milan, Munich, Oslo, Stockholm, Warsaw, Vienna and Zurich (secondary market).

FLAPD comprised around 3.3 gigawatts of server capacity at the beginning of the year. In the first half of the year, just 138 megawatts were added, but the increase is likely to total 381 megawatts for the year as a whole, meaning that a total of around 3.6 gigawatts will be available. By way of comparison with the USA: the 381 MW added this year in FLAPD corresponds in terms of size to the growth in twelve months in northern Virginia alone. Europe's largest server center, Frankfurt, represents the largest chunk of additional supply in Europe, with an expected 130 megawatts.

Vacancy rate lower than ever

New buildings and extensions are not keeping pace with demand on both sides of the Atlantic. Tenants on both sides of the Atlantic are primarily large US tech companies that sign rental agreements long before the plants are completed. At mid-year, less than ten percent of capacity was available in FLAPD for the first time, and CBRE expects even less than eight percent by the end of the year. In Frankfurt, where a whole quarter was still vacant in 2017, the vacancy rate should even fall below four percent by the end of the year, despite the 130 megawatts being added. However, this is still more vacancy than in the USA, where just 2.8 percent of the primary market is unlet.

On the European secondary market, CBRE expects 267 megawatts of additional capacity this year, with Madrid, Milan, Warsaw and Zurich contributing the most. CBRE expects the secondary locations to still have almost 16 percent free capacity, although this would be five percentage points less than two years ago. And the ten cities in the European secondary market will be significantly less important. By the end of the year, they will account for a total of around 1.1 gigawatts, less than a third of the capacity of the five large cities (FLAPD).

Another comparison with the USA: 3.9 GW of additional capacity was under construction in the top 8 cities in the first half of the year, plus half a gigawatt on the secondary market in the USA. This current construction activity in the top 16 US cities is therefore roughly equivalent to the total capacity that Adam and Eve will have installed in the top 15 European cities by the end of 2024.

AI makes oil cheaper

Artificial intelligence is very energy-intensive. Each ChatGPT request costs ten times as much energy as a Google search. The lack of power supply is also a major brake on the construction of data centers. The AI-related demand for electricity will cause the price of oil to rise a little over the next ten years, Goldman Sachs Research expects. The analysts estimate the impact on the marginal incentive price at two dollars per barrel. This is the price required to produce the last barrel of oil needed to fully satisfy demand.

Nevertheless, AI should make oil cheaper: This is because the algorithms reduce costs, Goldman Sachs is convinced; the development of new shale oil deposits should become 30 percent cheaper, and existing plants would extract eight to 20 percent more oil thanks to AI support. These factors would reduce the marginal incentive price by five dollars per barrel, resulting in an overall medium-term reduction of three dollars per barrel.

(ds)