Morgan Stanley: AI Data Center Costs Rising 15% and Why Europe Benefits
As seen on the 24/7 Wall St. homepage on November 19, 2026.
Europe's AI Moment
- Data center build costs running 12% higher year over year, with 15% inflation expected the next two years
- Politics and financing join power and planning as constraints; US build may slow in 2028-2029, aiding Europe and Asia
- Sovereign AI demand pushing European telcos to build their own data centers despite rising costs
- Agentic AI shifts inference orchestration to CPUs; 800-volt racks and silicon carbide move into focus
Morgan Stanley's European team is telling clients the AI buildout gets more expensive before it gets cheaper, and that a 2028 US permitting slowdown pushes capacity toward Europe, South Korea and India. For anyone owning data center or power-semiconductor exposure, the cost curve and the geography of the next wave are both moving.
Continue ReadingShow less
Recorded at Morgan Stanley's European TMT Conference in Barcelona, the episode brings together Paul Walsh, head of European Research, Emmett Kelly, head of European Telco and Data Centers, and Lee Simpson, head of European Technology Hardware. Their conversation covers four interlocking themes: the data center cost curve, sovereign AI demand in Europe, the rise of agentic AI, and the accelerating hunt for power efficiency.
On cost, Kelly's figures are direct. Building a data center is already running 12% more expensive year over year, and Morgan Stanley expects that inflation rate to reach 15% for each of the next two years. Beyond raw materials and labor, the team identifies two new constraints sitting alongside the established hurdles of power access and planning approval: politics following the US midterm elections, and tightening access to financing.
The US permitting environment is the geographic pivot point. Projects already under construction will finish, keeping US capacity growth strong through 2027. After that, a consultation period between builders and the new administration covering planning permission, unionized labor, water use, and renewables could slow new US starts in 2028 and into 2029. That gap, in Morgan Stanley's view, creates a window for European data center builders and for markets in Asia, with South Korea and India specifically named.
Sponsored
Sovereign AI is amplifying European demand independently of any US slowdown. Industries including banking, government, pharmaceuticals, and defense are generating data that European operators insist must stay on European platforms, from the physical data center layer up through the software stack. That pressure is pushing telcos, which Kelly describes as a little late to the game, to begin building their own data center infrastructure despite the rising costs.
Agentic AI adds a hardware wrinkle that investors in semiconductor exposure will want to track. Simpson's argument is that inference orchestration shifts from GPUs to CPUs as agents multiply, making the CPU a new focal compute resource. At the rack level, 800-volt architectures and silicon carbide components move into focus as power efficiency becomes a harder constraint, a theme the team expects to remain central at next year's conference.
Power remains the thread running through every other discussion. Ireland is cited as a leading indicator of how power-hungry data centers can become, and Kelly notes that getting a grid connection in the UK can take between five and eight years. Data center owners expressed a preference for nuclear power because of its stable base load, though the practical consensus is for a mix of grid, battery backup, nuclear, wind, and solar, with France and Finland called out as the European markets currently best positioned on power supply.