Data Centres, AI and Scope 2: How Surging Power Demand Is Changing Corporate Climate Reporting

Two numbers from the same company in the same year tell you everything about the problem.

Google reduced its data centre emissions by 12 per cent in 2024 through clean energy procurement and operational improvements. Over the same period, its absolute data centre electricity consumption grew by 27 per cent.

Both figures are accurate. Both are auditable. They point in opposite directions because they measure different things, and the gap between them is where corporate climate reporting is now under the most strain. For digital-heavy businesses, that gap is widening every quarter, and the accounting rules that permit it are being rewritten around them.

The Demand Picture

Data centres consumed roughly 415 terawatt hours in 2024, around 1.5 per cent of global electricity, having grown at about 12 per cent annually over the previous five years. The International Energy Agency's base case projects that roughly doubling to around 945 terawatt hours by 2030, near 3 per cent of global electricity, and continuing to about 1,200 terawatt hours by 2035.

The rate of change is what matters more than the level. Data centre electricity demand grew 17 per cent in 2025 against global electricity demand growth of 3 per cent, with AI-focused facilities climbing considerably faster. Capital expenditure by five large technology companies exceeded 400 billion dollars in 2025 and is expected to rise by a further 75 per cent in 2026, putting the capex of those five firms above global investment in oil and gas production.

Two counterweights deserve stating, because the debate frequently ignores them.

Efficiency is improving fast. Power consumption per AI task is falling at a rate the IEA describes as unprecedented in energy history. Per-query energy is dropping sharply even as aggregate consumption rises.

In global terms the sector remains modest. The projected 530 terawatt hour rise in data centre demand by 2030 represents about 8 per cent of the overall increase in electricity demand the IEA projects, less than electric vehicles at 838 terawatt hours or air conditioning at 651 terawatt hours.

But concentration is extreme, and concentration is what breaks accounting. Data centres already consumed around 26 per cent of Virginia's electricity, with EPRI projecting 41 to 59 per cent by 2030 and seven further states potentially exceeding 20 per cent. Ireland sits near 21 per cent nationally and Dublin around 79 per cent. Amsterdam, London and Frankfurt ran between 33 and 42 per cent. In advanced economies, roughly a quarter of electricity demand growth to 2030 is expected to come from data centres, with about half of the growth in the United States and Japan.

A company can be a rounding error globally and the dominant load on the grid it actually draws from. Scope 2 accounting is a grid-level exercise, so the local number is the one that counts.

Why This Breaks Scope 2 Accounting

Scope 2 has always been reported two ways, and dual reporting survives the current revision unchanged.

Location-based emissions use the average emissions intensity of the grid you draw from. They reflect physical reality: what was actually generated to serve your load.

Market-based emissions reflect contractual instruments, including renewable energy certificates and power purchase agreements. They reflect what you paid for.

For most of the last decade those two figures moved together closely enough that the difference was a technical footnote. For a company adding gigawatts of round-the-clock load in constrained grids, they now diverge dramatically. Market-based emissions can fall while location-based emissions rise, because procurement is scaling faster than consumption in contractual terms and slower in physical terms.

The Google figures illustrate exactly this. The 12 per cent reduction is a procurement outcome. The 27 per cent consumption growth is a physical one.

Most hyperscalers purchase certificates against operational electricity, and disclosed Scope 2 emissions are nonetheless rising as growth outpaces clean energy procurement. Carbon intensity per unit of compute is generally falling while absolute emissions rise. Reporting both of those honestly is possible. Reporting only the flattering one is where credibility fails.

The Annual Matching Problem

Here is the specific mechanism, and why data centres are more exposed to it than almost any other sector.

Under current rules, a company can buy certificates generated at any point in the year and apply them against consumption at any other point. Solar generated at midday in June can be matched against electricity consumed at 3am in December, and the reported market-based emissions for that electricity are zero.

For a manufacturer with daytime-weighted load, annual matching is a reasonable approximation. For a data centre it is not. Data centre load is flat and continuous, running at high utilisation around the clock. Solar generates for part of the day. Wind is intermittent. A facility matched annually with renewable certificates is, in physical fact, drawing on whatever the grid is running overnight, which in most systems means gas or coal.

The bigger and more continuous the load, the greater the gap between the contractual claim and the physical outcome. Digital infrastructure sits at the extreme end of that distribution.

What The GHG Protocol Tried, And What Happened

The GHG Protocol proposed to close this gap, and the market rejected the proposal.

Its Scope 2 consultation, which ran to 31 January 2026, proposed requiring hourly matching, so that certificates would need to come from the same hour in which electricity was consumed, and deliverability, so that certificates would need to come from generators connected to the buyer through an electrically linked grid.

The feedback was decisive. Of 909 respondents on hourly matching, 70 per cent gave little or no support, and among companies specifically only 12 per cent were in favour. On deliverability, 59 per cent of 875 respondents were opposed. The consultation drew nearly 1,100 responses from 56 countries.

The objections were substantive. Long-term power purchase agreements, which finance new renewable capacity, become harder to justify if their certificates cannot be matched hourly. Narrower geographic boundaries push investment away from regions with the greatest decarbonisation potential. And granular emission factors are not universally available, particularly outside mature electricity markets.

The GHG Protocol has said it will revise the draft and explore whether offering multiple reporting approaches, reflecting different theories of change, could better respond to the range of views. The Technical Working Group reconvened this month to reconcile the comments. A second consultation is expected during 2026, with final publication anticipated around 2027, and the standard is now being folded into a joint corporate standard with ISO, consultation on which is planned for the second quarter of 2027 with publication estimated for the fourth quarter of 2028.

One technical point widely misunderstood in the panic: the draft never required 100 per cent hourly coverage to comply. A company achieving 70 per cent hourly matched coverage would report 70 per cent. The proposal changed how the number is calculated, not what score you must achieve.

What SBTi V2.0 Already Changed

While the GHG Protocol debate continues, a change that is already final has landed, and digital-heavy businesses should be modelling it now.

The Science Based Targets initiative published Corporate Net-Zero Standard V2.0 on 11 June 2026, taking effect from 31 January 2027 with a transition period running into 2028. It separates Scope 1 and Scope 2 targets, which were previously combined, and it changes the basis of the Scope 2 target.

The Scope 2 target must now be based on the physical, location-based inventory. Market instruments including renewable energy certificates and power purchase agreements remain recognised, but they are handled separately rather than bundled into the target itself, and new quality criteria apply to energy attribute certificates.

Read that against the Google numbers. A company whose market-based emissions are falling while its location-based emissions rise has, under V2.0, a target running against the rising number.

For data centre operators and any business with large, growing electricity consumption, this is the more immediate change of the two. It is final, it has a date, and it removes the mechanism that has allowed procurement to substitute for consumption reduction in target performance.

SBTi has kept hourly matching optional for now, which gives some breathing room while the GHG Protocol position resolves.

PPA And REC Strategy Under Pressure

Three pressures are converging on procurement strategy simultaneously.

Quality criteria are tightening. SBTi V2.0 introduces new criteria for energy attribute certificates. The GHG Protocol revision, whatever form it takes, is directionally moving toward greater granularity. Unbundled certificates purchased far from the point of consumption are the most exposed instrument.

Temporal and geographic granularity is the direction of travel even though the specific proposals were rejected. The rejection was about pace and mandate, not destination. Companies building procurement strategies on the assumption that annual matching survives indefinitely are taking a position on regulatory outcome rather than on physics.

Additionality is under scrutiny. Certificates from existing generation that would have run anyway deliver no incremental decarbonisation. Long-term power purchase agreements financing new capacity do, which is precisely the argument respondents made against the hourly matching proposal.

The strategic response that survives all three pressures is procurement that is physically plausible: generation in the same grid region, contracted long term, ideally with a generation profile that bears some relationship to the consumption profile. That is more expensive than unbundled annual certificates and considerably more defensible.

Grid Constraints Are Now A Reporting Issue

Something that used to be an operations problem has become a disclosure problem.

Interconnection queues, capacity pricing and siting restrictions now directly affect reported emissions, because they determine where load can be placed and what generation serves it. In PJM, capacity prices rose from 28.92 to 329.17 dollars per megawatt-day, with data centres reported as driving 63 per cent of the increase. New York imposed a statewide moratorium on new hyperscale data centres in July 2026, pausing discretionary permits for facilities drawing 50 megawatts or more while it prepares an environmental impact statement, with roughly 12 gigawatts of data centre load sitting in the interconnection queue.

For reporting, this matters in three ways. Siting decisions determine your location-based emissions factor, and a facility placed in a coal-heavy grid carries that intensity regardless of procurement. Constrained grids limit the availability of deliverable clean generation, which affects both procurement cost and any future deliverability test. And moratoria and permitting delays are transition risks that belong in climate-related financial risk disclosure, not only in operational planning.

Net-Zero Credibility

The credibility question reduces to whether a company reports the divergence or hides behind one side of it.

Absolute emissions rising while intensity per unit of compute falls is a defensible position, stated plainly. Efficiency gains are real and substantial. Growth outpacing them is also real. A company that discloses both, explains the trajectory and sets out what would have to change for absolute emissions to peak is making a credible claim.

A company that leads with market-based emissions, omits the location-based figure from its headline narrative and describes itself as running on 100 per cent renewable energy while its physical consumption grows sharply is making a claim that the next generation of standards will not support, and that anti-greenwashing regimes in the UK and EU are increasingly capable of testing.

The direction of travel across SBTi V2.0, the GHG Protocol revision and the ISSB standards points the same way: toward the physical number as the anchor and the contractual number as supplementary information.

What Digital-Heavy Businesses Should Do

Report both figures with equal prominence. If your market-based and location-based emissions diverge materially, the divergence is the most decision-useful thing in your disclosure. Explaining it is stronger than featuring one number.

Model your Scope 2 target against the location-based inventory now. Under SBTi V2.0 that is the basis, and for many digital businesses the number is materially worse than what they have been reporting against.

Measure your hourly matched percentage even though it is not required. You cannot manage the exposure without the figure, and if any version of granular matching becomes standard you will need several years of data.

Audit your certificate portfolio by vintage, location and bundling. Unbundled certificates from distant generation are the most exposed to every proposed tightening.

Treat grid region as an emissions decision. Siting is now one of the largest single determinants of your reported Scope 2, and it is decided years before the emissions appear in a report.

Put grid constraints in your climate risk disclosure. Interconnection delays, capacity pricing and moratoria are transition risks with financial consequences, and they belong alongside the emissions numbers.

Track the GHG Protocol second consultation and respond. The first round showed that corporate responses shape outcomes materially, and digital infrastructure has more at stake in the granularity question than any other sector.

Reporting Checklist

1. Calculate and disclose both location-based and market-based Scope 2 emissions with equal prominence, and explain any divergence.

2. Model your Scope 2 position against the physical location-based inventory, which is the basis for targets under SBTi Corporate Net-Zero Standard V2.0.

3. Note that V2.0 separates Scope 1 and Scope 2 targets and handles market instruments outside the target itself.

4. Plan for V2.0 taking effect from 31 January 2027 with a transition period into 2028.

5. Measure your hourly matched percentage now, even though hourly matching remains optional under SBTi and was rejected in the GHG Protocol consultation.

6. Audit your energy attribute certificate portfolio by vintage, generation location, bundling and additionality.

7. Assess exposure to deliverability concepts by checking whether contracted generation sits in the same electrically linked grid as your consumption.

8. Treat data centre siting as an emissions decision, since grid region determines your location-based factor.

9. Disclose grid constraints, interconnection delays, capacity pricing and permitting moratoria as transition risks.

10. Distinguish intensity improvements from absolute emissions in your narrative rather than substituting one for the other.

11. Avoid unqualified renewable energy claims where physical consumption is growing, given anti-greenwashing enforcement in the UK and EU.

12. Track the GHG Protocol second Scope 2 consultation expected during 2026 and the joint standard with ISO consulting in the second quarter of 2027.

Position as of September 2026. The GHG Protocol Scope 2 revision is at draft stage with no proposals adopted, and SBTi Corporate Net-Zero Standard V2.0 is final but not yet in effect. Demand projections are scenario-based and vary considerably between sources. Confirm current requirements against the GHG Protocol, SBTi and your applicable disclosure regime, and take professional advice for your circumstances.

Sources

International Energy Agency, Greenhouse Gas Protocol, Science Based Targets initiative, Electric Power Research Institute, Lawrence Berkeley National Laboratory, PJM Interconnection, Carbon Brief, Brookings Institution, Oeko-Institut, Ember, Data Center Frontier

This article is intended for general professional information and does not constitute legal, financial, or investment advice.