The trillion-dollar bet on data centres

The global data centre market is entering a period of dramatic expansion. 

Some 48 gigwatts of capacity – the amount of computing power data centres can support – will be delivered between 2026 and 2029, a 77% increase in global supply volumes, according to Knight Frank's Data Centre Atlas. AI supply volumes are set to expand to 27GW, from 8GW, making the sector responsible for almost a quarter of operational data-centre capacity.

The massive spending on infrastructure required to achieve all this is making investors nervous. Tech stocks fell heavily this week. Corporate bonds issued by Coreweave, a firm that rents out computing power to AI companies, fell towards distressed levels. Last week, Firmus, an Australian company that builds and operates data centres and rents out computing power, abandoned an IPO over concerns relating to its valuation.

The key concern is whether the firms raising the money to finance the AI infrastructure build out can ever recoup the money and generate adequate returns. AI providers are expected to spend nearly $1 trillion on AI infrastructure this year, and are forecast to invest an additional $6 trillion between 2027 and 2030, according to research by technology consultancy Gartner.

The group reckons AI data centre owners will need to generate a cumulative $8.5 trillion in revenue through 2030 to cover announced capital spending, financing costs and a 12% return on invested capital, yet "total enterprise and consumer AI revenue that can be attributed to AI data centers during this period is forecast to reach no more than $3.6 trillion." Should AI infrastructure investment continue on its current trajectory, more than $4 trillion in asset value could be exposed to write-downs through 2032, the company concludes. 

Post-money

Forecasting through 2032 is a tricky business. Some worry that AI will fail to live up to expectations - in which case these issues crystallise – while researchers from inside the labs worry that AI will cause an extinction event by then. Others are somewhere in between. Elon Musk believes money will be of no value in ten years, in which case any issues recouping losses presumably fall away.

These are issues for investors, regulators and the companies to hash out. The most pressing near-term challenge for developers is delivering anything like the required pipeline within the constraints of power, water and local planning policies. The Knight Frank Data Centre Atlas, for example, describes London’s trajectory as “defined by supply-side constraints”. In core West London availability zones (Slough and Hayes), grid capacity has tightened to the point where securing new connections can take 5 to 10+ years, limiting near-term development. OpenAI paused its Stargate UK project earlier this year due to the unfavourable regulatory environment and high energy costs. 

These problems are replicated to differing degrees elsewhere. Power constraints are common across Europe, Australia and India, for example. There is some irony in the fact that investors are increasingly concerned that the industry is building more capacity than it will ever need, while developers struggle to secure the power and permissions required to build anything like the capacity forecast. Raising more money than the industry can earn back is one risk. Being unable to deploy it is another.

Chilling effects

The spending on AI infrastructure in the US is fuelling inflation, though it pales in comparison to energy prices. The annual rate at which prices are rising climbed to 3.7% in September, up from 3.4% the previous month, according to official figures released this week. 

Core inflation, which strips out energy prices, rose to 2.4%. Economists at Goldman Sachs reckon about 45 basis points of that is due to the AI build out, which will peak at 50 basis points by the end of the year. 

The average 30-year mortgage rate has surged to 7.4%, which has had a chilling effect on the housing market. Existing home sales may fall below 4 million this year for the first time since the 1990s,  Lisa Sturtevant, chief economist at home-listing service Bright MLS told Bloomberg. 

Fully let

The rapid expansion of the AI firms has been a boon for some landlords in London. The Times interviewed British Land's head of campuses Mike Wiseman last week about its success at 1 Triton Square, which is now fully let to the likes of Humanoid, the robot maker, and Anthropic. 

In a trading update published yesterday, the firm said it had signed 78 deals covering 689,000 sq ft across its campuses in the first half of its financial year, with rents 14.8% above estimated rental value. At 2 Finsbury Avenue, following the final agreement with Citadel to take 33% of the building, a further 119,000 sq ft is under offer, bringing the scheme to 52% let or under offer. At Broadgate Tower, a further 6 deals across 49,000 sq ft have completed, taking the scheme to 62% let.

In other news...

Canary Wharf developer George Iacobescu overseeing new London skyscraper (FT), Rejoining EU could boost growth by 6%, says Goldman Sachs (Times), and finally, AI-boosted UK economy beats forecasts, giving Healey pre-budget lift (Reuters).