*Three-line summary
Gartner forecasts that more than 80% of neocloud providers will cease independent operations by 2030.
Doubts over AI data center profitability are shaking financial markets, and the race to secure power is drawing attention.
Forecasts of surging data center power demand now face fresh questions amid talk of a bubble.
Astronomical sums are pouring into artificial intelligence, but actual revenue will fall far short. Talk of an AI bubble has struck financial markets once again. The argument surfaces in various versions, and this time the main theme is the core AI infrastructure: data centers. The central claim is that, apart from those owned by hyperscalers, a substantial share of data centers will fail to turn a profit and disappear.
This so-called data center shakeout thesis could deliver a major shock not only to financial markets but also to energy, and to electricity in particular. Data center expansion driven by the AI boom is one of the pillars pushing up global power demand. This article examines the ripple effects that data center pessimism could have on energy markets.
A 'Darwinian moment'
The episode began with a short report released on the 14th by global market research firm Gartner. The report said that $3.5 trillion (about 4.7138 quadrillion won) in AI infrastructure spending will be exposed to the risk of asset impairment by 2030. The reason: to meet the capital expenditures hyperscalers have announced so far, their funding costs and a minimum return on invested capital of 12%, they would need to lift revenue to a combined $8.5 trillion by 2030, while actual revenue is expected to reach only $3.6 trillion.
The Gartner report also forecast that neocloud companies, which rely heavily on borrowing unlike hyperscalers with diversified earnings, could face a wave of restructuring. It projected that more than 80% of neocloud providers will cease independent operations by 2030. On the same day, financial risk came to the fore at CoreWeave, the largest neocloud provider, as the burden from AI investment drew attention: its shares fell 15.5% and the yield on its corporate bonds jumped more than 1 percentage point to 15% (coupon of 9.625%, maturing in 2032).
Other cases have emerged as well. Australia's Firmus this month withdrew an initial public offering that had aimed to raise about $5 billion, after struggling to attract investment. Britain's Nscale signed long-term contracts worth more than $100 billion with Microsoft and Anthropic, among others, but is said to have failed to secure sufficient funding to build its data centers. Ludovic Subran, chief investment officer at Germany's Allianz, told the Financial Times that data center bonds have reached a "Darwinian moment," a remark that reads as meaning they stand at a fork between evolution and extinction.
Will expectations of a windfall turn into a bubble?
The power sector, which has treated data centers as a powerful growth engine, is watching this pessimism closely, because electricity demand may not materialize as expected. In particular, the dispute touches a doubt that existed even before it arose: whether data centers will really expand power demand as promised.
U.S. hyperscalers are currently signing electricity contracts wherever they can, in order to secure power to supply their data centers. According to S&P Global Energy, as of February this year the four largest U.S. hyperscalers — Google, Amazon, Meta and Microsoft — had contracted about 135 gigawatts of clean energy capacity worldwide, including renewables and nuclear. That is a 3.6-fold increase over the past three years.
Investment in gas-fired generation is also rising as data center power demand grows. Also according to S&P Global Energy, gas turbine orders totaled 51 gigawatts last year, the highest since 2000.
CoreWeave, at the center of the current dispute, said it has secured power contracts totaling 4.2 gigawatts. Nebius and IREN are each reported to have secured, or set a target of securing, about 5 gigawatts. That means neocloud firms, too, had been moving faster to lock in power.
But what if data center construction plans are cut sharply? On this point, the U.S.-based Institute for Energy Economics and Financial Analysis said last year that investors were being warned of possible overbuilding in the U.S. power industry, noting that utilities are drawing up plans based on assumptions of greater data center demand than the technology industry itself forecasts. James Burke, chief executive of U.S. power producer Vistra Energy, has suggested that electricity demand from grid interconnection applications may be three to five times larger than what will actually materialize.
Will the AI bubble argument shake power demand forecasts too?
A wave of neocloud failures would not necessarily make an equivalent amount of data center power demand evaporate. The International Energy Agency also forecasts that data centers will account for less than 10% of the increase in global electricity demand between 2024 and 2030. In other words, AI-driven growth in power demand is only one part of the picture. The agency says other major factors play the leading role, including rising industrial output, electrification, the spread of electric vehicles and wider use of air conditioning. The likelihood that total power demand will keep rising still appears high.
Even so, the takeaway from this dispute may be that fresh questions have arisen about the forecast that data centers will multiply rapidly on the back of the AI boom, with power demand exploding in step. Whether the AI bubble argument that began in financial markets now spreads into a "data center bubble" argument in power markets is worth watching.
Related 'Petro-Electro' articles
① Are data centers only water- and power-guzzlers?
② Brakes on AI? It is not only AI that stops.
③ The 'AI electricity bill' is coming. Who pays it?
※From petroleum to electricity. Energy is a key keyword for understanding the economy, industry, international affairs and responses to climate change. Subscribe to this reporter, or to the [Cho Yang-joon's Petro-Electro] series, for economic insight through the lens of energy.