AI Data Centres & The Grid:
Reality-Checked To 2035

A Tesla/SpaceX-style engineering console for understanding how AI data centre build-out interacts with transmission, firming, and consumer bills in the Australian NEM – using, and correcting, the Clean Energy Finance Corporation / Baringa view.

Perspective: AI Data Centre Positive – Grid Honest
Focus: Who Pays & When – Consumers, DCs, Industry
Horizon: 2025 → 2035 Scenario Engine

This simulator is not anti–data centre. It assumes AI infrastructure is strategically important and then does the hard work Baringa and CEFC did not: aligning load growth with transmission, firming, system strength and cost allocation – so AI build-out can be optimised rather than resisted.

2035 System Snapshot
Live Scenario · Central NEM View
AI DC Load · 2035 24 TWh
+18 % Bill
Approximate change in average residential bill vs. 2025, including wholesale, transmission, distribution and policy charges under current slider settings.
This consolidates what CEFC/Baringa model only at the wholesale level into a bill-level view, explicitly adding the transmission and firming costs usually left off the chart.
AEMO ISP 2024 Baseline
Baringa Central Case
Expert Alternative (You)
Scenario Engine · 2035
Tune the three levers CEFC/Baringa separated: data centre load, renewables/firming built for that load, and how much of the shared grid cost is socialised vs. paid by AI data centres.
Central (Baringa-like)
Grid-Aligned
Stress Test
AI DC Load · 2035
24 TWh · ~4.7 GW average
Matched Renewables
1.0 × DC load  ( 24 TWh)
Firming & Storage Coverage
50% of DC load firmed (4‑hour equivalent)
Who Pays Shared Grid Cost
DCs pay 20% · Consumers pay 80%
DC Attribution to Shared Augmentation
33% of shared grid cost triggered by DC load
What % of Western Sydney/Melbourne transmission augmentation is caused by data centre load vs. broader electrification?
• 0%: Pure stranded capacity (DC-optimistic)
• 33%: Conservative base case (ECIS-7 default)
• 50%: Geographic concentration
• 100%: Full attribution (stress test)
⚠️ Results Robust Across Attribution Range
Even if data centres cause zero shared augmentation (0%), consumers still face +5.3% bill growth from ISP baseline, CER, and firming. Attribution affects magnitude (±15%) but NOT direction. Use slider to test sensitivity.
Note: This models incremental DC-caused costs. Total consumer exposure (including pre-existing ISP/CER/firming) is higher. See ECIS-7 framework for full analysis.
Additional Grid Capex 2025‑2035
$14.0
Billion
≈ $3.0B DC‑funded (connections, BTM storage) · ≈ $11.0B consumer‑funded (shared transmission, firming, system strength).
Grid outlook: constrained but salvageable with disciplined siting.
Average Residential Bill Change
+18
% vs 2025
≈ +$360/year on a $2,000 bill – CEFC/Baringa show only +2‑3% by modelling wholesale prices alone.
Consumer risk: high without explicit DC co‑funding of shared assets.
Non‑Data Centre Industry Impact
+9
% tariffs
Network-heavy industrial users (smelters, chemicals, manufacturing) bear a disproportionate share of socialised RAB growth.
Competes with green industry policy unless DCs co-invest.
Attribution Sensitivity: Results Across 0-100% Range
Attribution Shared Aug. Consumer Bill Industrial
0% (Min) $0.0B +5.3% +3.0%
25% $1.3B +5.8% +4.0%
33% (Base) $1.7B +6.1% +4.5%
50% $2.5B +6.5% +5.5%
75% $3.8B +7.0% +6.5%
100% (Max) $5.0B +7.5% +7.5%
Key Insight: Even at 0% DC attribution, consumers face +5.3% incremental bill growth. Variance across 0-100% range is ±15%. Core finding: attribution affects magnitude, not direction.
Interpretation
Under these assumptions, AI data centres are absolutely compatible with a clean, reliable NEM – but only if we treat firming and shared transmission as first‑class design objects, not externalities. The Clean Energy Finance Corporation Blind Spot is to model DC growth against renewables alone, without integrating ISP transmission and RAB dynamics.
Lens Selector · What Each Study Sees
Switch perspectives. The underlying physics and economics are the same; what changes is which parts of the system each actor chooses to see.
CEFC / Baringa
Mandala
Expert Alternative
Private DC capex
Consumer‑funded RAB
Shared benefit / ambiguous
Load forecasts & renewables matching
Included · high sophistication
Storage (1.9 GW) to firm DC load
Modelled but cost allocation unnamed
$85‑135B \"investment opportunity\"
Undifferentiated private vs. RAB
Innovation: cooling, flexible connections
Correct, but orthogonal to allocation
ISP transmission overruns ($16‑24B)
Absent
System strength (syncons) & firming RAB
Absent
Bill‑level impact (network + wholesale)
Wholesale only
Cost allocation: negotiated vs. prescribed
Never mentioned
The CEFC/Baringa lens is technically strong on load and renewables, but it keeps the Regulated Asset Base off the page. That makes it ideal for press releases, and unsafe as a decision surface for 2035 planning.
Fact vs Model vs Omission clearly separated
Objective: get AI build-out right, not slower
State Deep Dive · TAPR 2025
October 2025 Transmission Annual Planning Reports reveal a bifurcated NEM: NSW/VIC in "deep transition" shock, QLD/SA in "monitoring" phase.
NSW
VIC
QLD
SA
CRITICAL STRESS
The Shockwave
10 GW
DC Enquiries
20×
Demand Jump
$3,500
SSUP/MVA/yr
NSW is the epicenter of DC load growth. Transgrid TAPR 2025 shows 307 GWh (2024) → 6,723 GWh (2035). System Strength Unit Price (SSUP) creates new firmness commodity at $3,500/MVA/yr. 10 synchronous condensers + 5 GW grid-forming BESS required by 2035.
Applicable Causal Chains
Western Sydney Cluster → Voltage Instability → GRID ↑, CONTROL ↑
Capacity Sterilisation → Blocks Renewables → EXECUTION ↓
EnergyConnect → SA Export Enabled → GRID ↑, RELIABILITY ↑
Source: Transgrid, AEMO VAPR, Powerlink, ElectraNet TAPRs (Oct 2025)
Aligned with ECIS-7 v3.8.0 Causal Ontology
Assumptions · 2025 → 2035

These are deliberately simple, conservative and traceable so a system dynamics modeller can port them straight into a Monte Carlo engine or a regulatory submission.

AI DC Load: Central 24 TWh by 2035 (≈4.7 GW average), range 8‑32 TWh consistent with CEFC/Baringa and AEMO IASR sensitivities.
Renewables for DC: \"Matched\" factor is applied to DC load only, not total NEM load. Baringa’s 3.2 GW renewables is interpreted as ≈1.0× DC share of load in their central case.
Firming & Storage: Slider maps to 4‑hour storage equivalent. 50% coverage with 4.7 GW DC average corresponds to ≈1.9 GW BESS in CEFC/Baringa modelling.
Capex Benchmarks (Order‑of‑Magnitude): transmission augmentation ≈ $2‑3M/MW for heavily constrained urban corridors; 4‑hour BESS ≈ $300‑400M/GWh; system strength remediation (syncons) ≈ $70‑150M each.
Bill Translation: 2025 reference retail bill $2,000/year. Wholesale ≈40%, network ≈55%, policy/other ≈5%. Wholesale deltas from CEFC/Baringa are combined with network deltas implied by RAB growth and DC allocation sliders.
Cost Allocation: DC connection assets treated as negotiated services (DC‑funded); shared augmentation and system strength as prescribed services (consumer‑funded) unless DC share slider explicitly increases their contribution.
Attribution · What This App Is Built On
Primary Studies Referenced (Facts):
  • CEFC / Baringa “Getting the balance right: Data centre growth and the energy transition” (Dec 2025)
  • Mandala “Data Centres as Enabling Infrastructure” (Nov 2025)
  • AER AusNet Cost Allocation Methodology decision (Nov 2019)
  • AEMO ISP 2024 & 2025 IASR / ESOO data centre sensitivities
Expert Alternative Layer (This App):
  • Distinguishes connection vs. shared transmission, negotiated vs. prescribed services
  • Adds firming & system strength to DC impact analysis explicitly
  • Translates capex streams into RAB growth, then into bill‑level effects for consumers and industrials
  • Labels CEFC/Baringa omissions as Clean Energy Finance Corporation Blind Spots to keep system dynamics honest while remaining AI‑positive.
Intended Use: internal scenario work, regulator briefings, board decks, and as a scaffold for a deeper Monte Carlo engine. Not a forecast; a transparent, hackable ontology for thinking clearly about AI data centre integration into the NEM.