Field Notes from the Deal Room

Field Notes

From the Deal Room

Buyers have professionalised. CUDO's primary research is a Censuswide survey of 701 infrastructure decision-makers across the UK, US and Europe (fielded February 12–25, 2026), complemented by a small qualitative interview programme. Survey findings are reported below with their base and question; interview and external material are labelled separately.

The winning pitch leads with token-level economics, power-secured timelines, goodput-based SLAs and sovereign-compliant architecture — not GPU counts.

Survey · Censuswide · base n=701

What's actually constraining scale

Base: all 701 respondents. Constraint question is multi-select; cost-driver split is share of cost-driver selections. Figures are CUDO survey findings.

Top scaling constraints

Multi-select · % of 701 respondents selecting each barrier

Cost-driver split

71% of cost-driver selections are infrastructure

Hyperscaler capex supercycle

$B · $250B (2024) to $600B (2026)

The Confidence Paradox

Capacity-planning confidence by region

The market committing the most capital (US) reports the weakest grasp — 90.7% confident overall, from 98% in the UK to 82.6% in the US. Base: all 701 respondents.

Confidence in capacity planning

Single-select · % of 701 respondents

Site Intelligence

Credible sites vs. risky locations

✓ Credible signals

Pacific NW hydro

Lowest documented power cost (~$50,800/mo per 1,000-GPU cluster)

Nordic hydropower

Stargate Norway (Narvik) — abundant low-cost hydro, cool climate

Brownfield / industrial reuse

Jay, Maine; Hamilton “Steelport” — mitigates grid/environmental concerns

EU secondary markets

Berlin, Milan, Madrid, Warsaw, Nordics — decentralized away from scarcity

⚠ Risky signals

FLAP-D primary hubs

Grid-constrained; Ireland already at 22% of national electricity

Moratorium-exposed

Maine, New York, Hamilton (Ontario) — 75 projects (~$130B) disrupted

High power-cost markets

Japan / Germany >$300,000/mo per 1,000-GPU cluster

Single foreign-provider dependency

CLOUD Act reach & “sovereign lock-in” exposure

Deal Room

What buyers are really asking

Procurement has shifted to a governance function — “Orchestration Before Intelligence.” Buyers scrutinise liability exclusions for AI harms (especially agentic AI), vendor lock-in and switching costs, data-privacy/IP risk, and demand evidence-based compliance (“AI evidence”). Contracts now align to ISO 42001 / ISO 23894 with shorter cycles.

Turn these signals into a placement decision

Discuss power, location, cluster architecture and TCO with a CUDO AI infrastructure specialist.