Investors in Firmus and other data centre stocks can’t sweep sustainability under the carpet forever

Imagine now investors in Firmus are feeling right now. Waiting to cash out on their investment? Waiting for even more growth? Concerned about whether or not Firmus can cope with demands for compute capacity as well as anti-data centre NIMBYs? We doubt many would affirm the third of those questions, but they should be.

It true that concerns about if demand for compute capacity is a thing are non-existent. But the harder question is whether a data centre operator can sustain its business model, economically and environmentally, over the medium to long term. Those two dimensions are more intertwined than most promotional documents acknowledge.

The Scale of the Problem investors in Firmus and other data centre stocks are facing

To understand what is at stake, consider the energy context. The International Energy Agency projected that global data centre electricity consumption would reach 650–1,050 TWh by 2026, accounting for roughly 1.5% of total global electricity consumption in 2024. Under its base case, that figure is expected to nearly double to approximately 945 TWh by 2030, representing close to 3% of global electricity use. In Ireland, data centres already consumed 22% of the country’s electricity in 2024, a figure that prompted regulatory intervention and informed moratoriums elsewhere in Europe.

The reason AI-optimised facilities are disproportionately energy-intensive is thermal density. A conventional enterprise data centre was designed for email servers and cloud storage; the racks ran warm but not critically so. An AI factory built to run thousands of Nvidia GPU clusters generates heat at a fundamentally different order of magnitude.

The industry benchmark for operational efficiency is Power Usage Effectiveness, or PUE: the ratio of total facility power to power consumed by IT equipment. A perfect score is 1.0. The global average, according to the Uptime Institute’s 2025 survey, has plateaued at approximately 1.58, meaning that for every watt directed to computing, a further 0.58 watts is consumed by cooling, lighting, and distribution overhead. Hyperscalers such as Google report a fleet-wide PUE of around 1.09, reflecting the efficiency premium of purpose-built, well-capitalised infrastructure.

The gap between 1.09 and 1.58 may seem immaterial, at least when you think of it on a micro-level (i.e. per unit). But on a macro level, the difference adds up. Facilities with high PUE are burning more money per unit of compute delivered, which, in a competitive market for GPU-as-a-service, will ultimately compress margins.

The Cooling Technology Shift

The most significant near-term lever for improving PUE is the transition from air-based cooling to liquid immersion. This is something that Firmus is reportedly doing. In immersion cooling systems, servers are submerged directly in engineered dielectric fluids, typically synthetic fluorocarbon-based liquids, which absorb heat far more efficiently than forced air.

Two-phase immersion systems — where the fluid cycles between liquid and vapour states — can handle heat fluxes exceeding 500 watts per square centimetre, compared with roughly 16 watts per square centimetre for conventional air cooling. Advanced cooling technologies of this type have demonstrated the ability to reduce cooling-related power consumption by 50–60% in operational settings, and can push facility PUE toward 1.03, close to the theoretical minimum.

The global market for data centre immersion cooling was valued at approximately US$1.7bn in 2025 and is projected to grow at a compound annual rate of around 18–20% through to the mid-2030s. Adoption is becoming a prerequisite for AI-density operations at scale, as opposed to just being a ‘nice to have’.

Back to Firmus, the IPO candidate says its proprietary technology claims to reduce energy consumption by 33% and water use by 99% relative to standard cloud data centres. Its flagship Project Southgate development in Launceston, Tasmania (part of an eventual A$73.3bn national build-out) uses closed-loop cooling systems, which the company says eliminates nearly all water consumption on the roughly 355 days per year when Launceston temperatures remain below 26°C.

The Tasmanian site has reportedly achieved a PUE of 1.03, near the theoretical limit and well below the global average. The project will initially house 36,000 of Nvidia’s GB300 Grace Blackwell chips, backed by a US$10bn debt facility led by Blackstone and a reported pre-IPO equity valuation in the range of A$6–8bn.

Why pick Tasmania? Because the state’s grid is overwhelmingly powered by hydroelectric generation, giving Firmus a credible claim to a low-carbon compute footprint that most operators elsewhere in Australia cannot match. However, that same grid characteristic raises a structural tension that investors should not overlook: Project Southgate’s initial Launceston site is reported to account for up to 15% of the state’s power grid capacity. And that’s just one site.

Whatever the renewable credentials of that electricity, drawing that share from a single state’s supply introduces meaningful concentration risk, both for the operator (grid instability or price spikes would directly impact costs) and for Tasmanian consumers and industries competing for the same resource. Investors should scrutinise how Firmus manages this exposure, including its battery-based frequency-control ancillary services capability, which the company says returns power to the grid during demand spikes.

And the same goes for investors in future data centre stocks to list. If Firmus is successful, we think it won’t be the last to list on the ASX. But the risk is that if Firmus is successful, retail investors will just buy without considering this pertinent question.

Community Resistance is another Material Risk

The GreenSquareDC episode a few weeks ago offers a sharp illustration of a different kind of sustainability risk: social licence. The Partners Group-backed operator withdrew its development application for a 120MW, AI-ready facility in Hazelmere, Perth on May 14 2026 — just 5 days before Western Australia’s outer metropolitan development assessment panel was scheduled to rule on the project. The withdrawal followed a public comment period that attracted 1,897 submissions, of which 1,836 were objections. Opponents cited noise from diesel backup generators, heat output, electromagnetic energy, and exhaust emissions, as well as the facility’s proximity to the Helena River and a Steiner school. The City of Swan had already recommended the application be refused.

GreenSquareDC had promoted the project as delivering 3,000 construction jobs, 180 permanent positions, and a A$635.5m initial economic contribution to Western Australia. Those arguments did not overcome the community response. The company has not announced plans to resubmit or seek an alternative site. For investors, the lesson is plain: an asset that cannot obtain or retain a planning permit has no value, regardless of the quality of its compute hardware or the strength of its customer pipeline.

What Mandatory Disclosures Now Require

Investors in ASX-listed data centre operators are no longer relying solely on voluntary ESG statements. Australia’s mandatory climate reporting regime, administered under the Corporations Act 2001 by ASIC and the AASB, commenced for Group 1 entities (large listed companies and National Greenhouse and Energy Reporting participants) for financial years beginning on or after 1 January 2025. First reports for 30 June year-ends are due 30 June 2026. Medium-sized entities (250+ employees, A$200m+ revenue, A$500m+ assets) enter the regime from 1 July 2026, and smaller entities from 1 July 2027.

Under AASB S2 Climate-related Disclosures, which aligns closely with the ISSB’s IFRS S2 standard, in-scope entities must now disclose climate-related risks and opportunities, scenario analysis, and Scope 1, 2, and (with a one-year phase-in) Scope 3 emissions. In Europe, the Energy Efficiency Directive already requires data centres above 500 kW to report PUE, Water Usage Effectiveness (WUE), and renewable energy share annually — a template that Australian regulators may reasonably reference as they build out their own sector-specific guidance.

For data centre operators and investors alike, these requirements serve a useful function: they force the disclosure of metrics that were previously treated as commercial-in-confidence, particularly PUE and WUE, which are arguably the most decision-relevant operational sustainability indicators available. Investors should regard any operator that does not voluntarily disclose these metrics ahead of mandatory deadlines with appropriate scepticism. An operator confident in its efficiency profile will typically want the market to know about it.

So What Should Investors Ask

The combination of mandatory disclosures, community approval risk, and rapidly evolving cooling technology means that data centre investment analysis can no longer rely solely on demand-side projections for AI compute. We suggest investors apply at least four operational sustainability tests:

• What is the operator’s stated and verified PUE, and what cooling technology underpins it?
• What share of a local grid does the facility occupy, and is the power renewable?
• Has the project obtained all necessary planning approvals, and is there a history of community engagement?
• And does the company’s forthcoming sustainability report, under AASB S2 or its equivalent, align with operational metrics rather than marketing language?

Infrastructure that can answer those questions credibly will, we believe, prove the more durable investment over the cycle. Infrastructure that cannot will face compounding friction from plenty of parties: communities, regulators, and eventually from the wholesale power markets it depends on.