Thoughts on the Market (Morgan Stanley) · · 13 min

Europe's AI Moment

With Emmett Kelly, Lee Simpson

ChorusBring the market's conversation to your AI. Learn more

Summary

Morgan Stanley's Emmett Kelly and Lee Simpson discuss Europe's AI build-out, focusing on data centers, power, telecoms and chips. Kelly argues that US data center capacity growth stays strong into 2027 because projects under construction proceed, but that rising build costs (up 12% year on year), politics, financing and permitting could slow the rollout in 2028 and 2029. He expects tech firms to relocate builds to permissive US states, spilling over to Europe and Asia such as South Korea and India, and sees sovereign European data center demand from banking, government, pharma and defense growing with strong pricing power. Kelly also flags power as the binding constraint for five years, with Ireland's data centers already using 25% of national grid capacity, and says telcos face churn from agentic AI and must offer either the best network or the best value. Simpson highlights grid connection delays of five to eight years in the UK, with France and Finland ahead on power offerings, and expects power efficiency gains such as 800-volt rack architectures using silicon carbide and gallium nitride. He also argues agentic AI shifts orchestration workloads to CPUs while GPUs keep the compute layer, and that edge AI will become important within twelve to eighteen months.

Key takeaways

  • Emmett Kelly on AI infrastructure & data centers: Data center build costs are up 12% yearly and are expected to accelerate to 15% next two years.
  • Emmett Kelly on AI infrastructure & data centers: US data center capacity growth stays strong in 2027, but politics could slow the 2028 build.
  • Emmett Kelly on AI infrastructure & data centers: Europe needs sovereign data centers for sensitive data, and corporate capacity demand is insatiable.
  • Lee Simpson on AI chips & semiconductors: Agentic AI shifts orchestration and inference from GPUs to CPUs, making CPUs a new compute resource.
  • Lee Simpson on Power demand & the grid: UK grid connections take 5-8 years; France and Finland lead European data center power offerings.

What was said, by company and theme

10 views from 2 speakers. Every quote is checked against the episode's transcript (how).

AI infrastructure & data centers

Where everyone stands on AI infrastructure & data centers

  • Emmett Kelly

    Data center build costs are up 12% yearly and are expected to accelerate to 15% next two years.

    Morgan Stanley's recent report added price as a new constraint on data center build. The estimated cost of building a data center is up 12% year on year, and inflation is expected to accelerate to 15% next year and the year after. Rising build costs are one of the new constraints that slow the rollout, alongside politics and financing.

    “We estimate that the cost of building a data center is up 12% year on year and we expect this inflation rate to accelerate to 15% next year and the year after.”at 2:13
  • Emmett Kelly

    US data center capacity growth stays strong in 2027, but politics could slow the 2028 build.

    Projects already under construction will continue unimpeded, so US data center capacity growth should be very strong in 2027. Builders must consult the new administration on planning permission, unionized labor, water use and renewables. That consultation could cause a slight slowdown in 2028 that rolls into 2029. Constraints have shifted from power and planning to politics and financing.

    “Where there are shovels in the ground, those projects will continue uninhibited and those data centers will get built. Therefore we think data center capacity growth in The States will be very strong in 2027.”at 2:36
  • Emmett Kelly

    If US builds stall, data centers will relocate to friendlier US states and to Europe and Asia.

    Tech companies are very mobile and will build wherever they can. If a location is blocked, they will move to another location, such as US states with more permissive planning. This could spill over to European data center builders and to Asia, where Morgan Stanley has earmarked South Korea, India and other countries for potentially greater-than-expected build.

    “Our view is that data centers will build wherever they can. So tech companies are very very mobile. If they can't build in location a, they will move to location b.”at 3:19
  • Emmett Kelly

    Europe needs sovereign data centers for sensitive data, and corporate capacity demand is insatiable.

    Sensitive data from banking, government, pharma, aviation and defense must be hosted on sovereign European platforms, sovereign from the software layer down to the physical data center. This follows major geopolitical shifts. Telcos are starting to build data centers despite higher costs, because corporate demand for capacity is growing with strong pricing power.

    “There is a viewpoint that Europe needs its own data centers for its own data. ... the feeling in Europe is the demand from corporates for data center capacity is really quite insatiable.”at 4:54

AI chips & semiconductors

Where everyone stands on AI chips & semiconductors

  • Lee Simpson · Morgan Stanley

    Agentic AI shifts orchestration and inference from GPUs to CPUs, making CPUs a new compute resource.

    Agentic AI is like the app moment on the smartphone, bringing agents to cloud AI and eventually the edge. Compute is separating into layers. The compute layer stays with GPUs, but orchestration, particularly inference, moves to CPUs. This makes the CPU the new compute resource, driven by greater agentic AI use.

    “There's a compute layer that will carry on being done by the GPU, but the orchestration, particularly inference, is going to move to a CPU, and that CPU is going to be the new compute resource driven by this greater use of [agentic] AI.”at 6:48
  • Lee Simpson · Morgan Stanley

    Agentic AI moving to edge devices is 12-18 months away and will need efficient, secure chips.

    At some point much of cloud AI will move back out to the edge. Edge devices will need to be much more power efficient and secure to run agentic AI. Compute architectures may have to change to allow for that security. The theme is not front and centre yet, but it is lurking roughly twelve to eighteen months away.

    “At some point we will be transferring much of this back out to the edge. What we've done in the cloud will come to the edge, and here it will have to be a lot more power efficient, it will have to be secure.”at 8:46
  • Lee Simpson · Morgan Stanley

    Power efficiency is becoming a key AI currency, driving 800V racks and new compound semiconductors.

    Efficiency matters in power supply to the data center, in efficient use across the facility, and in the server rack. New server-rack power architectures are coming, including an 800-volt architecture, which will require compound semiconductors such as silicon carbide and gallium nitride. Device architectures will also move closer to the processor to deliver power more efficiently. This theme will roll for many years.

    “It's a huge currency in AI, be it in the supply of power to the data center, be it in the efficient use across the data center, particularly in the server rack, or even at the edge device.”at 10:47

Power demand & the grid

Where everyone stands on Power demand & the grid

  • Emmett Kelly

    Power is data centers' biggest constraint; Ireland already uses 25% of grid capacity for them.

    Power has been the greatest constraint in data centers for five years. Ireland, the most advanced data center market per capita, now sees 25% of its national grid capacity used by data centers. Owners would ideally like nuclear for its baseload, behind-the-fence pricing, water and security. In practice they want a mix of grid, batteries, nuclear and renewables.

    “Power has been the greatest constraint in the data center world for the last five years... Ireland now sees 25% of its national grid capacity used by data centers.”at 9:44
  • Lee Simpson · Morgan Stanley

    UK grid connections take 5-8 years; France and Finland lead European data center power offerings.

    Getting a grid connection in the UK can take five to eight years, which constrains data center siting. Some countries are at the forefront of offering power solutions to data center operators. France and Finland stand out as markets where the power offering is ahead of other European countries.

    “To get a grid connection in The UK can take between five and eight years. Some countries are very much at the forefront of offering power solutions to data center operators and I would highlight France and Finland as markets where the power offering is ahead of other European countries.”at 10:47

Telecom & connectivity

Where everyone stands on Telecom & connectivity

  • Emmett Kelly

    Agentic AI could trigger telco churn; operators with the best network or best value win share.

    Some large telcos will deploy their own agents to handle third-party agents, but will not negotiate with them. If agentic AI triggers churn as customers move between telcos, operators with the best network can win share, since poor network quality is a common complaint. Operators with average networks and average prices risk being squeezed, so they must be either best network or best value.

    “If [agentic] AI does trigger churn and customers moving between telcos, the operators with the best network feel that they can win market share because one of the most common customer complaints in telcos is my network isn't very good.”at 8:11

Topics

  • CPU demand from agentic AI
  • European grid connection for data centers
  • US data center build outlook
  • agentic AI and telco churn
  • data center build costs
  • data center siting
  • edge AI compute
  • power efficiency in AI compute
  • power for data centers
  • sovereign data center capacity in Europe

More from Thoughts on the Market (Morgan Stanley)

All Thoughts on the Market (Morgan Stanley) episodes

Ask your AI about this episode

Chorus gives Claude, ChatGPT and other assistants every view from this episode and every other show Chorus follows. Ask who agrees, who changed their mind, and what else was said about the same companies.

A weekly email with the calls that mattered from the most important voices in investing. Unsubscribe any time. Privacy Policy

Chorus reports what speakers said. It isn't investment advice. How Chorus works · About Chorus