AI Infrastructure Gap: $1.8tn Contracted — $700bn+ CAPEX

Sep 25

3 min read

10 September 2026

Evolution: Oil → Gas → LNG → Power → Power for AI

Key Points

$1.8tn — contracted future business
$700bn+ — planned CAPEX (2026)
17 GW — AI data-centre capacity delivery gap

$1.8 Trillion — Contracted Future Business

$678bn — Microsoft $638bn — Oracle $514bn — Google Cloud [1–3]
$700+ Billion — Planned CAPEX in 2026
$200bn+ — Amazon $200bn — Alphabet $175bn — Microsoft $130–145bn — Meta [4]

85 GW — In Development (458 Facilities)

24 GW — fully secured
68 GW — upper delivery projection
17 GW / 20% 🔴 — delivery gap (2030) [5]

Critical Resources

🔴 Power — generation + grid connection
🔴 Land — powered sites + fibre access
🔴 Cooling — water + heat rejection.
Nearly 5 years — median grid-connection waiting time [6]

Delivery Models

1. Conventional

Land → Power → Cooling → Data Centre Proven. Scalable. Core industry model.

2. Modular

Factory → Modules → Transport → Installation → Connection

🟢 Site Labour ↓ — less on-site construction
🟢 Parallel Execution — factory + site work
🟢 Quality ↑ — controlled fabrication and testing
🟢 Schedule ↓ — faster deployment potential Location — fabrication in lower-cost locations Replication — repeatable designs across multiple projects

3. Floating

Fabrication Yard → Floating Platform → Power Location → Operation

🟢 Land ↓ — minimal land requirement
🟢 Cooling — seawater heat rejection
🟢 Construction — yard fabrication and integration
🟢 Mobility — relocation between locations
Expansion — additional modules / platforms
Decommissioning — removable infrastructure

25-Year Hull vs 5-Year Compute → Replaceable Compute Modules

The Opportunity

85 GW Pipeline → 68 GW Delivery → 17 GW Gap

The Race for Delivery

Who will convert capital into powered computing capacity fastest? Conventional → Modular → Floating

We are exploring modular and floating AI data-centre ideas at an early stage and would welcome dialogue with infrastructure investors and strategic financial partners interested in the next generation of AI infrastructure.

Methodology

Contracted Business — Microsoft/Oracle RPO + Google Cloud backlog; future contracted business, not AI-only.
CAPEX — 2026 company guidance/estimates; approximate and not AI-only.
Pipeline — 458 AI data-centre facilities / 85.4 GW in active development worldwide.
Delivery — ClusterBid projects 62–68 GW operational by 2030. The 17 GW / 20% gap is the difference between the 85.4 GW pipeline and upper 68 GW projection.
Concepts — Modular/floating are early-stage concepts subject to project-specific assessment.

Sources

Microsoft Investor Relations — FY2026 Q4: $678bn commercial RPO.
Oracle Investor Relations — FY2026: $638bn RPO.
Alphabet / Google — Q2 2026: $513.9bn Google Cloud backlog.
Amazon | Alphabet | Microsoft | Meta — 2026 CAPEX guidance/outlooks.
ClusterBid — AI Data Center Construction Pipeline 2026–2028: 458 facilities / 85.4 GW; 28% fully de-risked; 62–68 GW projected operational by 2030.
Lawrence Berkeley National Laboratory — grid-connection queue analysis: nearly five-year median request-to-commercial-operation timeline in 2024.
Meta / Amazon SEC filings — approximately five-year useful life for AI/server equipment.
Microsoft Investor Relations — FY2026 Q4 (29 Jul 2026): useful life of data-centre and office buildings extended from 15 to 25 years, effective FY27. Offshore engineering references — approximately 25-year marine hull design-life basis.