Thoughts on the Market

2026-11-20 · Hosted by Mike Wilson · Morgan Stanley

Executive Summary

Recorded live from Morgan Stanley's European TMT conference in Barcelona, this episode focused on four themes shaping European technology, media, and telecom investing: data center buildout constraints, the push for European sovereign AI, the rise of agentic AI, and the ongoing power shortfall. Emmett Kelly, head of European Telco and Data Centers, said data center construction costs are up 12% year-on-year and are expected to accelerate to 15% over the next two years, compounding existing power and planning bottlenecks.

Key Stories & Changes

1. Data Center Cost Inflation and US Political Headwinds

  • Data center construction costs are up 12% year-on-year, expected to accelerate to 15% annually over the next two years

  • Post-midterm political shifts in the US have added a new constraint beyond traditional power and planning bottlenecks: financing access

  • Projects already underway ("shovels in the ground") will proceed, supporting strong 2027 US capacity growth, but a consultation period on planning permission, unionized labor, water use, and renewables could slow 2028-2029 build

2. Geographic Shift in Data Center Build-Out

  • Constrained US locations are expected to see capacity migrate to more planning-friendly states

  • Potential spillover into Europe and Asia, with South Korea and India specifically flagged as markets that could see greater-than-expected data center growth

3. European Sovereign AI Push

  • Data sovereignty remains a core discussion theme for a second consecutive year at the conference

  • Europe's banking, finance, government, pharmaceutical, aviation, and defense sectors are driving demand for fully sovereign platforms — software layer through physical data center

  • European telcos, though late entrants, are beginning to build their own data center infrastructure despite rising costs, citing insatiable corporate demand and strong pricing power

4. Agentic AI as the Next Computing Shift

  • Lee Simpson compared the rise of agentic AI to the app-store moment for smartphones, with agents expected to extend from cloud to edge devices

  • Compute orchestration is shifting: GPUs remain the core compute layer, but inference orchestration is moving toward CPUs as the new driver of compute resource demand

  • Telcos are growing more comfortable engaging with third-party AI agents, though they plan to keep their own agents from directly negotiating with outside agents

  • Telcos with the best networks are seen as positioned to gain market share in an agentic AI world, while operators stuck in the middle (average network, average pricing) risk being squeezed

5. Power as the Central Constraint

  • Ireland now sees 25% of its national grid capacity consumed by data centers — the most advanced data center market in the world on a per-population basis

  • UK grid connections can take five to eight years to secure

  • France and Finland were highlighted as European markets ahead of peers in offering power solutions to data center operators

  • Data center owners reportedly want a "smorgasbord" of power sources: grid, battery backup, nuclear, and renewables (wind, solar)

1. Cost Inflation Reshaping Data Center Economics

Rising construction costs (12% now, accelerating to 15%) are becoming as significant a constraint on data center buildout as power and planning approval, adding a third major bottleneck that could reshape where and how quickly AI infrastructure gets built globally.

2. Politics as a New Build-Out Variable

For the first time, political dynamics — specifically the US midterm election outcome — are cited alongside power and planning as a direct constraint on data center rollout timing, suggesting infrastructure investors must now factor electoral and regulatory cycles into capacity planning, not just engineering and permitting timelines.

3. Compute Architecture Bifurcation

The anticipated shift of AI orchestration workloads from GPU to CPU, alongside new power architectures (800-volt systems, silicon carbide and gallium nitride semiconductors), signals a broader hardware evolution beyond simply adding more GPU capacity — efficiency and architecture innovation are becoming as important as raw compute scale.

4. Sovereign AI as a Durable, Multi-Year Theme

With data sovereignty discussed for a second straight year and telcos only now beginning to build dedicated infrastructure, this remains an early-stage but structurally persistent theme in European technology investment, driven by regulated industries' data residency requirements. ---

Sentiment Analysis

Overall Market Sentiment: Constructively Cautious

Speakers expressed confidence in near-term AI infrastructure demand but flagged structural constraints — cost, power, and politics — that could slow growth from 2028 onward.

Risk Factors Highlighted

Accelerating construction cost inflation: Data center build costs are projected to rise from 12% to 15% annually, compressing returns on new infrastructure.

Political consultation delays: A new US administration-driven consultation period on planning, labor, water, and renewables could slow 2028-2029 capacity growth.

Persistent power bottlenecks: Grid connection delays of five to eight years in the UK illustrate a structural constraint that is not quickly solvable.

Financing access constraints: Access to capital for data center build-out was flagged as a newly emerging blockage alongside power and planning.

Edge security requirements: Running agentic AI securely at the edge will require compute architecture changes that are not yet mature.

This episode was covered in today's [The Market Signal — 2026-11-20](https://marketsignal.beehiiv.com/p/the-market-signal-2026-11-20), a cross-source synthesis of multiple podcast reports.

Keep Reading