The Global Capital Context
The world's major hyperscalers are preparing to spend close to US$700 billion in capital expenditure in 2026 alone. A single gigawatt of AI data centre capacity costs US$12 to US$14 billion to build, inclusive of land, GPUs, and renewable power infrastructure. That is a Snowy Hydro 2.0-sized investment, committed every quarter, and growing.
The markets absorbing this capital right now are Malaysia, India, Japan, and the Middle East. Australia captured only a small fraction, not because capital was unavailable, but because the market could not absorb it fast enough. CBRE forecasts Australia's live data centre capacity will grow from approximately 1.4GW in 2025 to around 1.8GW by 2028 — still well short of projected demand, leaving an estimated supply gap of 0.7 to 1.7GW. Across Asia Pacific, CBRE projects a regional shortfall of 15 to 25GW by 2028 due to power constraints and a lack of AI-ready facilities. Once capacity clusters in another market, Khuda warns, it is very difficult to shift.
Lesson 1: A National AI Vision Needs a Shared Fact Base
Khuda points to the UK on AI safety, Singapore on talent, and India on GPU-native model development as examples of countries with genuine strategic clarity. Australia's federal plan is "an important start," but the vision must sharply define the sectors to target, the capabilities to build, the skills to grow, and the energy transition required to underpin all of it.
Competing assumptions about energy requirements, water use, and community impacts create planning friction that delays the capital deployment Australia needs. Unresolved data sovereignty requirements and AI copyright and compensation frameworks are additional constraints influencing where global AI companies choose to build and train their models. A single, trusted national baseline would make planning clearer and materially reduce misinformation in the public debate.
Lesson 2: The Approvals Process Is the Actual Constraint
Khuda's headline point is direct: "Data centres should deliver benefits, not burdens. But we're not achieving this because the approvals process is holding projects back." In most major markets, planning takes six to eighteen months. Victoria has shown it can approve a major project in around four months; NSW's Investment Delivery Authority is targeting the same. Achieving this at scale requires coordinated decision-making across land, water, and energy simultaneously, not sequentially.
A Corrs Chambers Westgarth analysis published in February 2026 confirmed the urgency: AWS alone plans to invest A$13.2 billion in Australian cloud infrastructure by 2027, and that commitment depends on approvals and energy frameworks being aligned. Khuda's prescribed solution is an outcomes-based regulatory model: lift performance standards on emissions, water use, and local impact, but remove the overly prescriptive conditions that slow delivery without producing better real-world outcomes.
Lesson 3: Community Licence Is a Competitive Variable
Communities want data centres to create jobs, invest in renewables, and deploy technologies that strengthen the grid and make water use smarter. Social licence will ultimately be judged on whether operators avoid adding pressure to household bills while minimising traffic, noise, and water use impacts.
This matters practically. AEMO forecasts data centres will consume 12 TWh of electricity annually by 2030, representing 6% of National Electricity Market grid supply, rising to 12% by 2050. At that level of grid impact, community and government scrutiny of data centre development will only intensify. Operators who invest in demonstrable community benefit, including renewable energy offtakes, grid-stabilising technology, and local hiring programs, will move faster through the planning system and face less organised opposition.
Lesson 4: The Skills Gap Is the Long Fuse
The skills exist in Australia, but not at the scale required to support the rapid buildout underway. Khuda calls explicitly on operators, including AirTrunk, to take a structured role in workforce development through education, training partnerships, apprenticeships, and local procurement.
The numbers support the urgency. Mandala Partners research, commissioned by AirTrunk, AWS, CDC, Microsoft and NEXTDC, projects that Australia's data centre workforce needs to grow by 8,300 to reach 17,900 by 2030, with demand concentrated in ICT professional roles and skilled trades. Over A$26 billion in additional data centre investment is forecast in Australia by 2030. Australia has the foundations: abundant renewables, deep capital markets, and a strong investment environment. But Khuda puts an explicit 12 to 18 months on the window.
Three Predictions for the Next 18 Months
Approvals velocity becomes the primary site selection variable by mid-2026. Operators in jurisdictions with streamlined, outcomes-based frameworks will attract a disproportionate share of hyperscale pre-commitments. States without planning reform will see capital route to those that have it.
HCF certification commands a commercial pricing premium by 2027. Gartner warned in January 2026 that 35% of nations will face lock-in to regional AI stacks by 2027 as sovereignty and security overtake scale as the primary AI infrastructure selection criteria. Enterprise buyers, not just government agencies, will require demonstrable data sovereignty assurances. Certified facilities will price accordingly.
The supply gap triggers M&A activity in certified mid-market facilities. CBRE's projected 0.7 to 1.7GW supply shortfall by 2028 means well-located, certified, mid-scale operators become strategic acquisition targets for hyperscale platforms needing last-mile sovereign capacity. Operators who have invested in certification, community relationships, and contracted power procurement will hold the strongest negotiating position.
Why This Is Significant
The clustering risk is real and time-bound. Khuda puts a 12 to 18-month window on Australia's competitive position. Once AI infrastructure anchors in Malaysia, India, or the Middle East, demand follows demand and does not come back.
Outcomes-based regulation directly advantages certified operators. Facilities built for genuine operational performance, not compliance paperwork, are the clear winners as planning frameworks shift toward measurable real-world outcomes.
Data sovereignty is an unresolved commercial differentiator. HCF-certified, sovereign-ready facilities are structurally differentiated for AI workloads that require verifiable data governance. That market is growing, not shrinking.
The AEMO grid forecasts are a structural signal, not a projection. Data centres reaching 6% of NEM grid supply by 2030 and 12% by 2050 means energy planning and data centre strategy can no longer be managed as separate disciplines.
The workforce gap is operator-owned, not government-solved. Khuda's call for operators to lead on skills development reflects a broader reality: the 8,300 additional workers needed by 2030 will not materialise from government programs alone.