AI’s $3T infrastructure binge continues despite lack of clear profits
Summary
Moody’s 2026 Outlook warns that global investment in data centre capacity to support AI, cloud and internet services could total roughly $3 trillion through to the end of the decade, with spending expected to peak in 2029. The surge is driven by hyperscalers’ heavy capital expenditure, but Moody’s flags multiple risks: power‑grid limits, construction and labour bottlenecks, rising costs, local opposition over electricity and water use, and doubts about whether the massive build‑out will generate clear returns.
Key Points
- Moody’s estimates about $3 trillion is required globally to expand data centre capacity for AI and cloud demand through 2029.
- Six major US hyperscalers (Microsoft, Amazon, Alphabet, Oracle, Meta and CoreWeave) spent nearly $400bn in 2025; projections show $500bn in 2026 and $600bn in 2027, with global investment peaking in 2029.
- Critical constraints include power availability, skilled labour shortages, materials and equipment bottlenecks, and rising construction costs.
- Financial risks stem from circular deals and large long‑term contracts (examples include OpenAI’s arrangements with big cloud players) that may spook investors if revenue doesn’t materialise; OpenAI reported very large net losses in 2025.
- Some tenants are now accepting delivery risks (eg exempts for power/utilities) and regions with supportive planning (such as the UK’s AI Growth Zones) are likelier to attract new builds.
Content summary
Moody’s outlook, seen by The Register, predicts continued strong demand for server capacity to run AI and cloud services, and calculates an approximate $3tn investment requirement covering buildings, IT kit and power. The report cautions about physical and financial bottlenecks: utilities struggling to meet sudden electricity demand, local pushback around environmental impacts, and stretched construction supply chains that are driving up timelines and costs.
The piece highlights hyperscalers’ rapid capital expenditure growth and charts a peak in global investment by 2029 followed by a decline in 2030. It also spotlights growing investor concern over financing structures — particularly ‘circular’ deals tying developers, cloud providers and AI firms together — and notes that significant losses at major AI firms intensify credit risks for the wider ecosystem.
Finally, the article notes behavioural shifts: tenants willing to shoulder risks to speed delivery, and governments or regions revising regulation to attract data centre investment, while sustainability and demonstrable revenue generation become ever more important to justify the build‑out.
Context and relevance
This coverage matters to investors, CIOs, planners, utilities and policy makers. The scale of projected spend affects capital markets and supply chains, while the power and resource constraints it exposes intersect with national energy planning and local planning policy. For businesses buying or selling cloud capacity, and for regions courting data centre investment, the piece flags where the biggest operational and financial hazards lie.
It also ties into broader industry trends: the shift to large generative models that demand huge inference/train capacity; growing scrutiny over AI economics after studies showing limited near‑term returns for many enterprise AI projects; and regulatory moves to steer where and how such infrastructure is permitted.
Why should I read this
Short version: billions are being poured into data centres, but the money might not buy returns — and someone’s going to be stuck with the electricity bill. If you care about where AI actually creates value (or who pays for its infrastructure), this gives you the quick lowdown on the risks and why planners, investors and utilities are getting twitchy.
Author style
Punchy: the article lays out stark numbers and clear red flags. If you’re involved in infra, investment or policy, the warnings here are actionable — not just noise. Read the detail if you make decisions tied to capacity, finance or energy planning.