AI is the marginal buyer: the four largest 2025-26 commitments alone exceed 80% of national operating capacity
The hyperscalers committed more than A$45bn to Australia in ten months, and the four largest 2025-26 commitments alone exceed 80% of national operating capacity. The commitment ledger, contract by contract, with dates.
AI-workload facility demand has risen ~27x over 20 years (M3). The 2025-26 contract wave converted that into named megawatts: CDC's 555 MW binding contract, OpenAI's 550 MW S7 MoU, Firmus's contracted 150 MW Melbourne phase, and Sharon AI's 50 MW NEXTDC expansion with a 1,000-to-20,000 GPU build-to-suit at M3, the first publicly named GPU-tenant build at scale in Australia.
Landmark AI-workload commitments 2025-26, MW
| Item | Value |
|---|---|
| CDC: US hyperscaler contract, May 2026 (555 MW, binding 10-yr min) | 555 |
| OpenAI: NEXTDC S7, Dec 2025 (550 MW, MoU, not yet a lease) | 550 |
| Firmus Southgate MEL ph.1, Mar 2026 (150 MW, contracted) | 150 |
| Sharon AI: NEXTDC expansion (50 MW, signed) | 50 |
Density is the quiet disruptor inside the demand story
Rack density, kW per rack
AI GPU clusters
Traditional enterprise / cloud
AI GPU clusters require 40-100+ kW/rack (NEXTDC S7 disclosure; IMARC cites >80 kW) against a traditional 5-15 kW. Consequences: direct-to-chip and immersion liquid cooling become standard; legacy air-cooled halls are partially obsolete for AI training; and MW-based capacity figures understate the revenue per hall of GPU-dense product.
Sovereign-AI advantage
Australia is one of only four APAC nations exempt from US AI-chip export restrictions (IMARC/openPR), meaning the GPUs that fill these halls can legally land here when they cannot in most of the region.
The hyperscalers have committed A$45bn+ to Australia in ten months: capacity follows cloud regions
| Program | Commitment | Capacity effect | Status |
|---|---|---|---|
| Microsoft Microsoft (23 Apr 2026) | A$25bn (US$18bn), to end-2029 | AU cloud/AI footprint up >140% by end-2029; 29 DC locations; 3 → 4 Azure regions (Melbourne opened Jul 2025, 48 → 150 MW) | Announced 23 Apr 2026 (Nadella + PM); largest-ever AU investment |
| AWS | A$20bn, 2025-29 (from A$13.2bn 2023-27) | Incl. >170 MW offtake from three new solar farms in Victoria and Queensland; ~80 MW Eastern Creek operating (est.) | Announced 14 Jun 2025 |
| OpenAI | A$7bn NEXTDC S7 program | 550 MW GPU supercluster phase; first 'OpenAI for Countries' project in APAC | MoU (Dec 2025); no binding lease |
| Not publicly itemised | Melbourne/Sydney region expansion | Data gap (register item 5) | |
| Anthropic | AU Government MoU (1 Apr 2026) | Reportedly negotiating capacity; CDC 'in the hot seat' | Unconfirmed |
Why hyperscaler self-builds distort every statistic
Microsoft, AWS and Google do not itemise Australian megawatts; site-level figures (e.g. AWS Eastern Creek ~80 MW) are research estimates, with no disclosure behind them. Capacity houses inconsistently include self-builds, one reason national totals range from 1.35 GW to 4.5 GW.
1.35 GW
4.5 GW*
Demand context
Australian public cloud spending crossed A$23bn in 2024 (IMARC citing industry data). EY-Parthenon estimates Microsoft's program alone contributes A$36bn to the local economy and 186,000 FTE-equivalent jobs in FY25.
Read-through for landlords
Microsoft's A$25bn explicitly spans capex and opex across leased and owned facilities. AirTrunk named Google, Apple, Meta, Amazon and Microsoft as Australian prospects (Sep 2026). The leased-campus model coexists with self-build, it does not lose to it.
The commitment ledger
Named program commitments, A$bn
| Item | base | A$bn |
|---|---|---|
| Microsoft | 0 | 25 |
| AWS | 25 | 20 |
| OpenAI-NEXTDC S7 | 45 | 7 |
| Sum (inashanu est.) | 0 | 52 |
What each figure measures
Named program commitments, by sponsor. Source: CNBC (23 Apr 2026); Quartz (23 Apr 2026); ASPI Strategist (3 Jun 2026); M3 via DCD (24 Nov 2025); Certified Strategic (1 & 10 Sep 2026); IMARC (2026); EY-Parthenon via CNBC. As at 21 Sep 2026.
- A$25bn
- Microsoft, to end-2029: explicitly spans capex and opex across leased and owned facilities; lifts AU cloud/AI capacity by >140%.
Named program commitments, by sponsor. Source: CNBC (23 Apr 2026); Quartz (23 Apr 2026); ASPI Strategist (3 Jun 2026); M3 via DCD (24 Nov 2025); Certified Strategic (1 & 10 Sep 2026); IMARC (2026); EY-Parthenon via CNBC. As at 21 Sep 2026.
- A$20bn
- AWS, 2025-29 (from A$13.2bn for 2023-27), including >170 MW of offtake from three new solar farms in Victoria and Queensland.
Named program commitments, by sponsor. Source: CNBC (23 Apr 2026); Quartz (23 Apr 2026); ASPI Strategist (3 Jun 2026); M3 via DCD (24 Nov 2025); Certified Strategic (1 & 10 Sep 2026); IMARC (2026); EY-Parthenon via CNBC. As at 21 Sep 2026.
- A$7bn
- OpenAI-NEXTDC S7 program, a non-binding MoU (Dec 2025).
Program scale in context. Source: CNBC (23 Apr 2026); Quartz (23 Apr 2026); ASPI Strategist (3 Jun 2026); M3 via DCD (24 Nov 2025); Certified Strategic (1 & 10 Sep 2026); IMARC (2026); EY-Parthenon via CNBC. As at 21 Sep 2026.
- A$23bn
- Australian public cloud spending crossed A$23bn in 2024 (IMARC citing industry data), a market-wide spend measure.
Program scale in context. Source: CNBC (23 Apr 2026); Quartz (23 Apr 2026); ASPI Strategist (3 Jun 2026); M3 via DCD (24 Nov 2025); Certified Strategic (1 & 10 Sep 2026); IMARC (2026); EY-Parthenon via CNBC. As at 21 Sep 2026.
- A$36bn
- EY-Parthenon's estimate of the Microsoft program's FY25 contribution to the local economy, alongside 186,000 FTE-equivalent jobs.
How to read it
Read the ledger as three different measures: multi-year program commitments (A$25bn, A$20bn, A$7bn), annual market spend (A$23bn) and an economic-contribution estimate (A$36bn). They are not additive.
Sources
- Bloomberg via Energy Connects, as at 6 May 2026
- Certified Strategic, as at 1 Sep 2026
- W.Media, as at 2026
- DCD
- Kalkine, as at Jul 2026
- M3 Property, as at Nov 2025
- Troview, as at Jun 2026
- IMARC, as at 2026
- IMARC/openPR, as at Apr 2026
- CNBC, as at 23 Apr 2026
- Quartz, as at 23 Apr 2026
- ASPI Strategist, as at 3 Jun 2026
- M3 via DCD, as at 24 Nov 2025
- Certified Strategic, as at 1 & 10 Sep 2026
- EcoGPT, as at Apr 2026
- RNZ, as at 6 Jul 2026
- EY-Parthenon via CNBC
- Amazon, as at 14 Jun 2025
- Microsoft, as at 23 Apr 2026
- EY-Parthenon via CNBC. Program sum: inashanu estimate
Related
- How much of Australia's announced data centre pipeline will actually be built?
- What is in Australia's data centre development pipeline?
- What are Australia's data sovereignty rules for data centres?
- How much data centre capacity will Australia have in 2030?
The full primer is published as an 85-page PDF (download) with the underlying market model as a workbook (download).
This page is one of 30 in Australian data centres: the research, indexed by the question each one answers, with the four pages where the published sources disagree set first.
The research is published by inashanu, a studio that produces fixed-scope market research for Australian private equity, priced and dated before work starts. The sample engagement is this primer in full.
This standard, on your question. Six fixed-price deliverables, A$14,000 to A$45,000 excluding GST, opening at the A$14,000 market map in 10 business days. State the decision and the date it has to be made; the written assessment of whether the evidence base will carry it follows within two business days and carries no charge. Request a brief assessment · the catalogue and its prices.