The scale of digital infrastructure is undergoing an unprecedented expansion. The era of AI, cloud-native services and hyperscale data centres is upon us, and with it comes a challenge: infrastructure that once might have been a relatively modest energy user is now becoming a major and growing source of emissions. For organisations in the built-environment, real estate, sustainability and software, this matters: because the same forces of scale, demand and infrastructure lock-in that apply to buildings apply to data centres.

Electricity, cooling, materials and embodied carbon

“Data centres” (i.e., IT equipment, servers, cooling, power supply, buildings, supporting infrastructure) consume electricity, require cooling (water and/or air), and have embodied emissions (building shell, server manufacture, infrastructure). According to International Energy Agency and analyst firm data, data-centre electricity demand has grown ~12 % per year since 2017.

In one forecast, data-centre electricity consumption could reach ~1,000 TWh by 2026/2030, roughly equivalent to the annual electricity demand of Japan. What’s more, the emissions embedded in those operations will depend on the grid mix: globally, many data centres still rely on fossil generation, so growth in demand risks increasing absolute emissions.

Material, embodied & lifecycle aspects

Beyond operations, the build-out of new data centre campuses uses large quantities of concrete, steel, servers, cooling equipment and has water-use implications. For example, the embodied carbon of server manufacturing, cooling systems, racks and structural shells is non-trivial. While many analyses focus on operational emissions (Scopes 1-3) from electricity, the growth of infrastructure means lock-in of high-carbon materials.

Why this matters for net-zero

From a net-zero goal perspective, the problem is two-fold:

• First: absolute emissions must decline (or be offset) if a company or system is to achieve net-zero (or net-negative) by a certain date. But if demand is rapidly increasing, even high efficiency gains may be outpaced.

• Second: growth in high-emission infrastructure creates “carbon baggage”. The built asset has embedded emissions that must be amortised over its life, but if the operational emissions remain high (or increase), it undermines the decarbonisation trajectory.

In short: increasing data-centre deployment + increasing AI loads = structural challenge to net-zero commitments.

AI workloads increase demand

AI and machine-learning workloads (training large models, inference at scale, edge AI, high-performance computing) are extremely energy-intensive. The training of advanced models consumes significant electricity, and the data-centre facilities required are large, secure, highly-powered, cooled aggressively. For example, some analyses indicate that AI-optimised data centres could quadruple electricity demand relative to traditional data centres by 2030.

Data-centre demand and grid stress

As the IEA and other analysts warn, data-centre growth may challenge national/regional grids and infrastructure, especially if demand growth competes with electrification (EVs, heating) or is served by fossil backup.

Efficiency gains vs load growth

There is good news: many operators are increasing efficiency (servers per kWh, cooling metrics like PUE, re-use systems). Google reported a six-fold increase in computing power per unit of electricity compared to five years ago. However, as mentioned, even large efficiency gains may be offset by massive scale-up of infrastructure and cloud demand.

Emissions remain a major concern

Fossil fuels still supply ~60% of data-centre electricity globally in some studies (check out Carbon Brief’s research on this topic). This means that unless data-centre electricity is supplied by low-carbon sources and growth is managed, doubling of demand risks meaningful increase in global emissions.

1 Alphabet Inc. / Google

• Google publicly pledged in 2021 to reach net-zero emissions across its operations and value chain by 2030, with a base year of 2019.

• However, its emissions from 2019 to 2024 rose by ~51% according to the Guardian; one independent report argues the true rise is ~65%.

• Furthermore, in 2025 the net-zero pledge was quietly de-emphasised on its website: the net-zero goal moved from headline to appendix, raising concerns of soft reversal.

• The company acknowledges the challenge: in its sustainability site it says “While we’re still committed… it’s become clear that achieving them is now more complex and challenging across every level”.

• Key driver: AI and data centre expansion. For example, Google’s data-centre energy use increased ~27% year-on-year and its electricity‐purchase emissions increased ~121% in recent years (according to an advocacy group).

Implication: Google’s ambition remains, but the visibility and credibility of its net-zero target are under question. In effect, Google appears to be recalibrating realistic path-to-net-zero in light of structural demand growth.

2 Microsoft Corporation

• Microsoft set a target to become carbon negative by 2030, removing more carbon than it emits, and to remove its historical emissions by 2050.

• Yet, as of its 2025 disclosures, overall emissions increased ~23.4% since 2020, driven by AI/cloud expansions and new data-centre builds.

• Microsoft acknowledges the challenge and is pivoting strategy: moving away from “non-additive” offsets, investing more in high-impact carbon removal, and placing larger emphasis on clean electricity procurement near its data-centre sites.

Implication: While Microsoft has not abandoned its target, the path to achieving it is now admitted to be more difficult; the company is shifting strategy rather than lowering ambition, but risk remains of missing its milestones.

3 Amazon.com, Inc. (including AWS)

• Amazon co-founded The Climate Pledge and committed to reach net-zero carbon emissions across its global operations by 2040.

• But in 2024, Amazon’s absolute emissions rose 6% from the prior year (to ~68.25 million tCO₂e) – the first such rise after years of decline.

• The rise is attributed in part to its data-centre build-out (via AWS) and the associated energy/materials demand.

Implication: Though the target remains publicly stated, the trajectory is moving away from it; practitioners and analysts perceive a widening gap between ambition and reality.

Why are these companies struggling?

• Structural demand growth from AI, cloud services and data-centre build-out is eroding expected reductions.
• Clean energy procurement constraints: even if companies contract renewables, grid availability, timing, geography and site-specific constraints hamper full decarbonisation.
• Material/embodied emissions and supply-chain (Scope 3) burdens: many emissions lie outside direct operations (e.g., servers, building materials, supply-chain).
• Offset/credit reliance: Some companies are shifting away from generic offsets; the focus is increasingly on real reductions rather than purely offsetting. Microsoft, for example, is dropping “non-additive” certificates.
• Messaging and credibility risk: when the pledge is de-emphasised (e.g., Google removing it from its main webpage) trust and stakeholder scrutiny increase.

Data centres, once peripheral to the built-environment discourse, are now central to the climate-carbon equation. The “AI boom” and associated growth in cloud-computing demand have shifted the calculus: even companies with ambitious net-zero goals now face structural headwinds from demand growth, materials, and energy supply.

As we’ve seen, major tech players such as Google, Microsoft and Amazon are under pressure: either their emissions are rising despite targets, or the visibility and credibility of their commitments are being questioned. For the built-environment and sustainability professionals, this means that carbon-footprint modelling, asset-due-diligence and corporate ESG advice must evolve to include these infrastructure dynamics.

The key takeaway: net-zero commitments are not static tick-boxes. They must be backed by credible, absolute-emissions reduction trajectories, transparent supply-chain and infrastructure assessment, and realistic growth assumptions. The era of “we’ll buy offsets and grow fast” is hitting the wall of energy and material reality.