AU data centres & energy + AU AI startup landscape (for Eva's desk)

Requested by: Young (2026-09-15): pair Tam with Eva. Tracker task-663d81b0. Split honoured: Eva = market angle/desk feeds; Tam = public evidence. Two notes in one file here (companion lanes).

Summary (3 bullets)

• DC power is real but the pipeline is inflated ~7×: AEMO 2026 ESOO forecasts 15 TWh (~7.8% of NEM) by 2029–30 → 34 TWh (13%) by 2035–36 (central case — a ~7× rise), but the connection queue holds 225 projects / 67 GW of requests vs only ~6 GW expected built ("phantom demand"), with just 11 projects (5.4 GW) truly in transmission connection — 60% NSW, 40% VIC.
• AU AI-startup market: credible, specialised, rising, structurally sub-scale. 2025 AU VC = $5.48B/390 deals (+31% YoY, 3rd-largest ever); direct AI funding $839M–$1.0B; 61% of all AU venture dollars touch AI; Sydney cracked the global top 30 (#26, $4.1B AI-native value); Melbourne AI-native +139%.
• The whole trade is in energised megawatts, not announcements. Scarce asset = firm grid connection/transmission/firming/water — not land, not announcements. Best-fit desk themes: buy boring (transmission, BESS, gas firmers, water infra), rent the hype (DC REITs highest-beta, re-rate only on milestone delivery).

Key findings

Lane 1 — Data centres & energy by city

• National: ~5 TWh DC draw in 2025-26 (~3% of NEM), ~1.5 GW installed IT capacity across 162–250 operational sites (DC Australia census, Mar-2026); $155bn+ local DC pipeline estimated; built capacity heading ~2 GW → 9.5 GW by early 2030s (M3 Property, ~$26bn capex, up to 1.7 GW supply gap by 2030).
• Sydney/NSW (dominant, ~60% of national pipeline): 44 DCs / 11.4 GW pipeline incl. CDC Marsden Park 504 MW, AirTrunk Kemps Creek toward 1 GW ($4.3bn construction financing), NEXTDC S7 612–650 MW (OpenAI's first APAC campus, approvals pending). Industry expects only ~1.2 GW online in Sydney by 2030 — the rest is paperwork. Transgrid enquiries 14 GW (up to 70% speculative). NSW's "power shortage" is partly locational — ~10 GW of capacity already unused in NSW distribution networks (Neara).
• Melbourne/VIC (the cost-optimal state): PGIM's Truganina site (20.7 ha, $165.6m → 200 MW plan) = APAC's priciest per-acre DC land deal; CDC Laverton 400+ MW; cheaper wholesale power (VIC wholesale +23% by 2035 vs NSW +26% absent renewables), Australia's best recycled-water access, cheaper industrial land.
• Brisbane/QLD (the regional wildcard): essentially all paperwork in metro; the real story is Zerra's Western Downs Digital Park — $31.9bn / 2.16 GW peak / 725 ha — Australia's largest-ever proposed DC, DA lodged, air-cooled, CSG/recycled water; plus CopperString-blocked NW QLD interest and Crisafulli's "energy-agnostic" (gas-adjacent) stance.
• Perth/WA (small but off-grid gigafab concepts): CDC Maddington 200 MW sovereign-AI campus (real, under construction); Project Meridien Kimberley (240 MW → 1 GW, $10bn+, FID 2029, 2032 target) and Pilbara off-grid P2 nodes — WA's story is off-grid renewables, not metro DCs.
• Water: Sydney Water forecasts up to 250 ML/day for DCs by 2035 (≈ Canberra's drinking water); national DC water ~5.5 GL/yr (~0.04% of national use); OpenAI's S7 dropped recycled-water cooling (politically salient); ~25.5 ML water/MW/yr rule for evaporative cooling — air/liquid cooling is the dominant AU defense.
• Grid stress: no reliability gaps before 2030 (ESOO, Aug-2026), but under high-DC growth, unserved energy 2035–36 = NSW 5.58% / VIC 3.95%. NEM consumption/peak up >40% over next decade; AEMO ISP calls for 6,000 km of new transmission; federal government (Jul 2026): DCs must "put at least as much energy into the grid as they draw" — renewable-matching, connection-cost causer-pays. ~2.9 GW of approved-unbuilt capacity grandfathered (not retrospective).
• Hyperscaler commitments: AWS A$20bn (2025-29, incl. $7bn Sydney GPU supercluster), Microsoft A$18–25bn (divergent figures), Google undisclosed, OpenAI $7bn Eastern Creek campus (Phase 1 H2 2027), Meta/Apple actively circling via AirTrunk (Sept 2026 per Bloomberg).

Lane 2 — AU AI-startup market

• 2025 totals: $5.48B/390 deals (+31% YoY, 3rd-largest year on record); Q4'25 alone >$2B, strongest since 2021; direct AI $839M–$1.0B (Tracxn); 61% of all AU venture dollars touch AI.
• Largest rounds: Baseten $1.5B Series F @ $13B (June 2026, Blackbird's biggest-ever ~$200M cheque — Australian co-founders, US company); Firmus $500M + $330M + $505M (Nvidia/Coatue, $5.5B pre-IPO, $10B Blackstone debt, ASX IPO targeted mid-2026); Airwallex $498M Series G @ $12B then ~$460M @ $16B; Synchron $303M; Harrison.ai $112M Series C; AdvanCell $112M; Diraq ~$100M (incl. $20M NRF); Heidi Health $65M (Point72, ~A$500M); Relevance AI $37M; Ivo $55M (NZ-founded, Blackbird-led); Lorikeet $54M; Syenta $36M; Dataro $14.3M; Brainfish $10M; Metacognition AI record $10M pre-seed.
• Sector split: AU specialises in vertical AI — health/medtech is the flagship ($829M biotech+medtech; Harrison.ai, Heidi, Synchron, AdvanCell); AI agents/enterprise SaaS (Lorikeet, Relevance AI, Brainfish); sovereign AI infrastructure (Firmus, NEXTDC×OpenAI) is the second pillar; deep tech/quantum (Diraq, Syenta). Underweight vs global: defence AI (no AU Anduril), fintech AI (one company — Airwallex).
• Geography: Victoria overtook NSW for the first time in 2025 — VIC $2.2B/134 deals vs NSW $1.7B/160 deals; QLD $504M/61 deals; WA single-digit. Sydney #26 globally / #1 Oceania ($59.7B ecosystem value, $4.1B AI-native); Melbourne #30, AI-native +139%. University-spinouts are the standout regional story — 200+ VIC spinouts valued $6.2B, UNSW Founders 60+ startups, Diraq/Syenta via GO8 pipeline.
• Headwinds: exit/IPO drought (the single biggest constraint per Cut Through); top-20 deals captured 58% of all capital; Series A valley of death; AI engineer chronic under-supply; 2026–27 R&D tax incentive tightening.
• Tailwinds: 66% of deals had international investors (up from 57%); National AI Plan (Dec 2025); sovereign-AI infrastructure positioning (Firmus $10B Blackstone debt, NEXTDC×OpenAI); $4T superannuation pool increasingly drawn toward VC; strong founder alumni flywheel (Atlassian/Canva).

Verdict (desk / Pulse)

• Desk framing: buy boring (transmission/BESS/gas firming/water infra + copper/transformers chains), rent the hype (DC REITs — HMC/DigiCo most direct AU-listed DC exposure, NEXTDC highest-beta but milestone-sensitive; AirTrunk private under Blackstone/CPP). Firming is the play: DC 90% load factor + minimal demand-flex directly benefits AGL/Origin/CLY-type peaking/firming portfolios.
• What actually matters: track energised MW, not announced GW — the connection queue dilution is the whole trade diagnostic (11 real projects out of 225; only 27% of allocated network capacity currently used).
• Watch-list: AEMO ISP transmission plan timing, AEMC DC connection rules H2-2026, NSW water politics (drought_years), Firmus's ASX IPO as the market's AI-infrastructure sentiment test, QLD CopperString date.

Sources

All claims URL-cited inline to primary sources (AEMO ESOO 2026, Reuters, City-of-Sydney/Ausgrid, The Urban Developer, ABC, Renew Economy, Guardian, SmartCompany, TechCrunch, Blackbird blog, Startup Genome GSER 2026, Tracxn AI report, Cut Through Venture 2025 report). Full note lives in lane files /tmp/au_datacentre_energy.md + /tmp/au_ai_startups.md, mirrored at the sync path.

Collaboration status

Eva: pinged with the market verdict (local loop, :9901, delivered ~21:13-21:20 AEST 2026-09-15); her desk feeds + listed-plays list pending merge-credit. This note auto-syncs to vault/docs via mirror cron.