Your AI Strategy Has an Unpriced Climate Liability
Most boards that approved an AI strategy in the last eighteen months also approved an environmental liability they have not measured, disclosed, or put on the risk register. Very few know it yet.
The Guardian’s recent reporting on Australia’s datacentre boom describes a proposed hyperscale facility on Mamre Road in Sydney’s outer west: 52 hectares, six four-storey buildings, 936 cooling units, 852 diesel backup generators. The last number is the one worth sitting with. Eight hundred and fifty-two diesel generators on a single site, as backup for the kind of compute that runs the AI tools your organisation has spent the year rolling out. That is what an AI strategy looks like at the infrastructure layer, and for directors with ESG or audit committee responsibility the useful question is whether any of it appears in your disclosures. In most organisations it does not, which makes this a governance problem before it is an environmental one.
The Vendor Abstraction Problem
Those of us who sat at board tables during the cloud migration of the 2010s watched this pattern the first time around. Organisations moved workloads to hyperscale providers and, in doing so, laundered the infrastructure risk through a vendor relationship. The servers, the power draw, the cooling, the geographic concentration of data: all of it became invisible because it now sat inside someone else’s contract. The board approved the business case and the complexity disappeared into a service agreement. AI adoption is doing the same thing at a larger scale.
Licensing a large language model, subscribing to a copilot tool, building internal capability on a foundation model API — each of these consumes compute that lives in a physical building somewhere, drawing power and water and producing emissions on your behalf. The vendor runs the building. Procurement runs the contract. Nobody runs the environmental exposure as a first-order business risk, because the abstraction makes it very easy not to.
Not owning the datacentre does not mean not owning the liability, and investors, regulators and customers will only accept that distinction for so long.
Why This Is a Disclosure Problem Now
Australia’s climate disclosure landscape is shifting in ways that make this a near-term governance issue, not a theoretical future one. The Treasury Laws Amendment (Financial Market Infrastructure and Other Measures) Act introduces mandatory climate-related financial disclosures for large entities, aligned with ISSB standards, with a phased start from the 2025–26 financial year. Under IFRS S2, organisations are required to disclose material climate-related risks and opportunities — including Scope 3 emissions, which is where cloud and AI infrastructure consumption sits for most enterprises.
Scope 3 is precisely where AI compute lives. It is emissions that occur in your value chain but outside your direct operations. The ISSB framework does not care that the emissions are generated by a vendor. It requires disclosure of material Scope 3 categories. For organisations with meaningful AI workloads, compute-related emissions are becoming material. The accounting methodology is still maturing, but the disclosure obligation is not waiting for the methodology to catch up.
ASIC has already signalled that greenwashing enforcement extends to omission, not just misstatement. Approving an AI strategy that promises efficiency and innovation, without disclosing the environmental cost of the infrastructure that strategy depends on, is exactly the kind of selective disclosure that regulators have been flagging.
The Risk Register Gap
The practical problem is organisational. The AI programme sits with the Chief Digital Officer or the CTO. The ESG programme sits with the sustainability team or the CFO. The disclosure obligation sits with legal and the company secretary. Those three groups are not talking to each other about this, and none of them has an incentive to start: the AI team is measured on progress, the sustainability team on improvement, and nobody at all is measured on working out how many diesel generators in Western Sydney sit behind the productivity tools approved last quarter.
So the risk is legible from outside the organisation — to regulators, to activist investors, to journalists with a planning portal login — and invisible inside it, which is the only place it could be managed.
Closing the Gap
There is a direct question that surfaces this in about thirty seconds of committee time: what is the estimated carbon and water footprint of our AI workload consumption, and where does that exposure sit in our disclosures? An answer of “we don’t know” or “that’s a vendor question” is the finding.
None of this requires a sustainability audit before anyone is allowed to use AI. It requires honest accounting, and it starts in procurement. Material AI vendor contracts should carry a data request for emissions intensity, water usage effectiveness and the energy mix of the datacentre regions involved. Providers with real sustainability programmes can produce that. Providers who cannot are also telling you something.
The next ESG disclosure cycle then needs an explicit assessment of whether AI compute is a material Scope 3 category for your organisation. If significant AI investment has been approved, it probably is, and a disclosure prepared without that assessment is incomplete by definition.
The third piece is a habit of mind rather than a control. “The vendor handles infrastructure” is a delegation, not a transfer, and delegation leaves the obligation where it was. Where that infrastructure is exposed — to carbon pricing, to water scarcity, to community opposition at the planning stage — your AI capability is exposed with it, because it depends on those buildings continuing to exist and remaining accessible.
The Board’s Real Question
Whether to adopt AI is not the live question; that decision is largely made. The live question is whether the governance around it is proportionate to what it actually carries, financially, operationally and environmentally. For most organisations it isn’t. The strategy was approved with a business case, the environmental liability was not in that business case, and the disclosure framework is now catching up to require it anyway.
Boards that connect AI strategy to ESG disclosure will do it deliberately, or a regulator will do it for them. The alternative is explaining in hindsight why 852 diesel generators never appeared anywhere in your climate risk reporting.