Carbon, Clean Power Procurement & 24/7 Carbon-Free Energy

Two facilities on the same grid can draw the same megawatt-hours and both claim "100% renewable." One runs on clean power most hours of the year. The other runs on the marginal gas unit at 2 a.m. and covers it with a renewable energy certificate (REC) from a wind farm a thousand miles away that was generating at noon six months ago. An annual claim is defensible only when its instruments meet the current GHG Protocol quality criteria. Same-hour supply matches the physical timing better, but neither certificate book alone proves how much fossil generation the investment displaced. What separates the two facilities is an accounting choice almost no executive deck makes explicit: the granularity at which consumption is matched to clean generation, annual volumetric matching versus 24/7 hourly carbon-free energy. Procurement strategy, on-site generation, scheduling and disclosure exposure all cascade from it.

Two clocks make the choice urgent in 2026. The load clock: AI is adding gigawatts of new, high-load-factor demand faster than clean supply can be built, so the marginal megawatt-hour a new data center pulls is disproportionately fossil. The standards clock: under GHG Protocol's 29 July 2026 update, the Scope 2 work now feeds a consolidated GHG Protocol–ISO corporate standard targeted for publication in Q4 2028. Hourly matching and deliverability are still proposals; what you must report changes only when the finished standard takes effect. A facility scoped in 2026 should still price a switch to finer matching before it signs a long-tenor REC contract.

Before procurement comes the boundary question: what counts as your emissions at all. The GHG Protocol's three scopes map onto a data center cleanly, and getting the mapping right is what makes a disclosure auditable rather than aspirational.

Scope 1: direct emissions from sources you own or control. For most grid-connected facilities this is small: diesel genset test runs, backup generator hours during outages, refrigerant leakage (a high-GWP line item that is easy to forget; Chapter 5.8 covers the chiller refrigerant rules that set it) and any on-site gas. Once you own or operate a behind-the-meter gas plant to beat the interconnection queue, Scope 1 stops being a rounding error and becomes the dominant term. Organizational consolidation and operational control decide the allocation, not the meter's location. If you buy the output of an on-site plant that an independent party owns and operates, those emissions are your Scope 2. Turbines consolidated under your chosen ownership or control approach are Scope 1, and a campus burning gas for primary power can emit more in Scope 1 than a grid-tied peer emits across all three scopes. That inversion is the carbon cost of speed-to-power and the central tension of the on-site section below. → Chapter 3.2.

Scope 2: purchased electricity, steam, heat and cooling. For a grid-connected AI facility this is the overwhelming majority of operational carbon, and it is where the matching decision lives. It is reported two ways, location-based and market-based (below), and the gap between them is where REC accounting operates.

Scope 3: everything upstream and downstream. For data centers the dominant Scope 3 category is embodied carbon: the concrete and steel of the shell and, increasingly, the manufacturing footprint of the GPUs, networking, and power and cooling plant. On sufficiently low-carbon power and short replacement cycles, embodied carbon can overtake operational emissions. The crossover depends on grid factor, service life, refresh and reuse assumptions, workload and system boundary. The reversal is large enough to warrant its own chapter. → Chapter 15.6.

Annual matching is the incumbent regime. Over a calendar year, you buy enough renewable generation (directly via PPA or as unbundled certificates) to equal your total consumption, then apply the eligible instruments' emission factors under the market-based hierarchy. Annual MWh equality alone does not authorize a zero figure. The method is cheap, liquid and, on its own terms, easy to take to 100%. Its flaw is temporal and geographic blindness. A facility matched at 100% on an annual basis still draws fossil power in every hour its contracted wind and solar are not generating, which on most grids is a large fraction of the year. The certificate says noon in Texas; the load is 2 a.m. in Virginia. Annual matching can be fully satisfied while the physical grid serving the load barely decarbonizes. A high-load-factor AI campus that draws flat power around the clock is the worst-case consumer for annual matching, because so much of its draw lands in hours when intermittent renewables are absent.

24/7 carbon-free energy (CFE) closes the temporal and geographic gap by matching consumption with carbon-free generation every hour, on the same grid. Google's CFE Score formalizes this: match consumption with qualifying carbon-free supply in each hour, cap credited supply at that hour's load, and aggregate using electricity-consumption weights across the year. The difference in difficulty is enormous. Google's fleet sits at roughly 65% CFE globally (2025 data, 2026 reporting), with wide regional spread: Latin America near 90% and Europe at 83%, but Asia-Pacific around 13%, where the grid is coal-heavy and clean PPAs are scarce. The last 10-40 percentage points of hourly CFE, what remains after the ~60-90% the table below shows renewables alone can reach, are the hard part. More solar cannot buy them. They require clean-firm supply that generates at night and in still air.

Annual matching remains a valid choice when eligible instruments meet the reporting method and your customer contracts. 24/7 CFE buys a tighter temporal match and costs you hourly tracking, storage and supply in difficult hours. Part of the case for the harder path is a hedge against the standard, but the risk is smaller than it first looks. GHG Protocol's consultation draft would apply hourly matching and deliverability only to large organizations; those under a size threshold would be exempt. The July 2026 summary of feedback on that Scope 2 consultation names no threshold, so a large data-center buyer should assume it is covered. The draft also proposes a legacy clause that keeps existing long-term contracts eligible. Respondents gave the hourly-matching and deliverability proposals, as drafted, little support, and GHG Protocol's Independent Standards Board now wants further work on several market-based reporting approaches. The final rule may therefore arrive phased in, softened or alongside alternatives. What is exposed is mainly what you buy after any effective date and the claims you publish. Contracts already signed are not. A buyer that assembles an annual-REC book today may still need hourly-matched supply for later load additions, bought in a tighter market. → disclosure mechanics in Chapter 15.7.

You cannot reach high hourly CFE on intermittent supply alone without absurd overbuild. An LUT University techno-economic study (journal Energy, 2026) models Finnish wind-and-solar supply with renewable-fuel backup. It finds a data center needs roughly seven times its baseload requirement in intermittent nameplate capacity, at a modeled €103–121/MWh for full baseload. That is feasible on its own terms and materially cheaper where the load is flexible. Read it as one modeled geography and technology set, not a universal solar multiplier: the overbuild factor and the price move with the resource, the storage, the backup fuel and how much demand flexibility you can offer. The other lever is clean-firm power, carbon-free generation that runs regardless of weather or time of day. The procurement question is which clean-firm lever you can actually contract on your timeline, at what premium against the overbuild-plus-storage-plus-backup alternative. Each carries a different bet on technology maturity, cost and schedule.

Nuclear: restarts and uprates first, SMRs later. The fastest clean-firm megawatts in 2026 come from existing reactors, in two forms. Restarts and uprates add new carbon-free output. Microsoft's 20-year, 835 MW offtake from Three Mile Island Unit 1, which Constellation is restarting as the Crane Clean Energy Center, is the example: it targets a 2027 restart as of 6 August 2026, accelerated from the original 2028 schedule. Restarts are few, and each restart date that follows is a target, not a commitment. NextEra targets the 615 MW Duane Arnold restart by the first quarter of 2029, with Google buying most of the plant's output. Palisades has no commercial-operation date at all. Fuel loading there is paused, Holtec said in September 2026, because one assembly tilted from upright after it was seated in the reactor vessel. Meta's Clinton deal keeps an operating plant running and adds a 30 MW uprate (each schedule is tracked in Chapter 3.5). Contracts for an operating plant's output, such as Amazon's 1.92 GW front-of-meter PPA with Talen at Susquehanna, secure firm clean supply but add none to the grid. Both forms are large, firm, long-tenor and tied to a known reactor.

Small modular reactors (SMRs) are the next wave but a slower one. Google's Kairos master agreement (Oct 2024; ~500 MW) anchored the first corporate deal for an SMR fleet, and its first implementing PPA (Kairos–TVA, Hermes 2, 50 MW, Google offtake) followed in Aug 2025. The IEA counts about 45 GW of conditional offtake agreements between data-center operators and SMR projects, up from 25 GW at the end of 2024 (April 2026). The economics are still modeled. DOE's Liftoff update (Sept 2024) puts the median overnight cost of a 300 MW boiling-water reference design at ~$13,000/kW, and EPRI's August 2023 analysis used $8,460/kWe first-of-a-kind and $5,134/kWe nth-of-a-kind for light-water SMRs, in 2022 dollars. Delivery dates are project targets: Google and Kairos aim for a first deployment in 2030 and the full fleet by 2035. SMRs are a 2030s clean-firm bet you contract now and wait on. → Chapter 3.2.

Enhanced geothermal: the dark-horse 24/7 baseload. Enhanced geothermal systems (EGS) reached contractual commercial operation in 2026. Fervo declared commercial operation on 30 September 2026 for its first GeoBlock at Cape Station, Utah, rated 33 MW net, and calls it the first greenfield EGS project to get there. Phase I totals about 100 MW across three blocks. Fervo targets the other two by 1 January 2027 and a 400 MW second phase, now under construction, for 2028. Google holds two separate Fervo contracts. Its 115 MW Nevada agreement, approved under NV Energy's Clean Transition Tariff, will add around-the-clock EGS capacity. In September 2026 Google signed a second deal: 396 MW from Cape Station for a potential Utah data center, expected to deliver in 2028, plus an option on about 600 MW more. That deal depends on the campus's approvals and on Utah regulatory approval. Geothermal is firm and carbon-free without new nuclear's licensing tail, but only favorable resource basins can host it. When you contract, ask for measured net output from a production block rather than nameplate.

Gas plus CCS: clean-firm in name, conditional in practice. Natural gas with carbon capture and storage is marketed as clean-firm, and it can be sited where nuclear and geothermal cannot. The catch is capture rate and lifecycle accounting. A 90%-capture plant still emits, upstream methane leakage is not captured at all, and the economics depend on storage geology and incentives that are not universal. Treat gas plus CCS as a partial-decarbonization firm option, not a zero-carbon one, and disclose the capture rate, not just the nameplate.

A large new load in a primary market now waits more than four years for a grid connection (JLL, 2026; → Chapter 3.2). The most-announced way around that wait is on-site generation behind the meter, largely gas turbines and engines. Here the operational win and the carbon loss point in opposite directions, and you cannot have both. Cleanview's mid-2026 tracker counts about 90 GW of behind-the-meter generation announced for 59 US data centers, about 82 GW of it since January 2025, but only about 2 GW was operating. Behind-the-meter generation delivers primary power on its own contracted schedule (equipment delivery, fuel, permits and commissioning), which can beat the host utility's service date. Where the operator owns or controls a gas plant, every megawatt-hour it burns is Scope 1 combustion that a grid-tied facility would have reported (largely cleaner) as Scope 2. Where an independent developer owns and operates it and sells you the output, the same combustion lands in your Scope 2. Either way the physical emissions, the air permit and the community fight sit at your site.

On-site options span a spectrum of carbon profiles, and the choice turns on three questions. Is this bridge power (run gas now, move to grid or clean-firm when the interconnection clears) or permanent primary generation? What is the fuel and capture posture? What do the community and the air permit allow? The consequences compound. A permanent gas campus is a 20-year emissions liability and an air-quality fight. A temporary-turbine bridge can use EPA's KKKKa temporary-turbine NOx subcategory (turbines of 850 MMBtu/h or less, up to 24 consecutive months at one location), which changes neither its CO2 nor the need for an air permit.

Supply (what power you procure) has a demand-side complement: when and where you run the work. Carbon-aware scheduling shifts flexible compute toward hours and regions where the grid's marginal emissions are lowest, in two forms. Temporal shifting moves interruptible work (batch inference, eval sweeps, embeddings generation, non-urgent fine-tuning) into clean hours such as midday solar or windy nights, and throttles or pauses it during dirty peaks. Spatial shifting routes flexible work to the cleanest available region. The mechanism is the one that makes a workload curtailable for the grid, so the carbon lever and the flexibility lever are two readings of the same capability.

Scheduling for carbon trades against time-bounded goodput, the metric that pays for the building. Temporal deferral trades deadline slack against accelerator utilization. Spatial shifting consumes destination capacity and inter-site bandwidth, plus checkpoint and migration overhead for work already in flight. The math works for interruptible, deadline-flexible work, which excludes the revenue workload for most operators. Online inference has a hard latency SLO and cannot be time-shifted. Synchronous pre-training runs continuously and is geographically pinned to its cluster. The deferrable surface is the batch and offline tier. Identify what fraction of the fleet is carbon-flexible and cap the goodput you will trade for it, rather than declaring the whole facility carbon-aware and quietly missing SLAs. The same scheduling primitive, pointed at price and grid-stress signals instead of carbon, becomes a grid-services revenue lever. → flexibility economics in Chapter 15.8.

Carbon offsets, which pay for reductions elsewhere to compensate for your own emissions, sit at the bottom of the hierarchy for a reason. The mitigation order is fixed: avoid (efficiency, reduce the draw), then match with clean supply (PPAs, 24/7 CFE, clean-firm), then, only for the irreducible residual, offset. The failure mode the industry keeps repeating is using cheap offsets in place of harder, more expensive hourly-matched procurement, because an offset retired today costs less than a clean-firm PPA contracted for a decade. The GHG Protocol already keeps offsets out of the inventory: credits are reported separately and never netted inside Scope 1, 2 or 3. Substituting offsets for procurement therefore does not reduce the number you report. It only decorates it.

Offsets belong in a separate compensation statement for residual emissions you have already minimized through avoidance and matching. They must be high-integrity (additional, permanent, verified; durable carbon removal is preferred over avoidance credits), and they must be disclosed separately, never netted silently against gross emissions to manufacture a "net zero" headline. An offset-heavy claim on a facility that never seriously pursued 24/7 matching is the disclosure equivalent of the annual-REC gap: technically reportable, increasingly indefensible. → reporting frameworks and audit posture in Chapter 15.7.

Settle the procurement boundary before you value carbon. The sustainability lead assigns each generator under the organization's chosen ownership or control approach, and the energy buyer records who owns and retires each PPA or certificate attribute. Selling an on-site renewable attribute removes it from the campus's market-based claim even while the same electrons serve its racks. Reconcile hourly consumption, local carbon-free supply, imports, storage-charging attributes and residual supply. Annual matching is total eligible attributes divided by annual consumption. Hourly matching is the sum of eligible supply, capped at each hour's load, divided by total consumption. Neither is a marginal-emissions calculation. GHG Protocol's Scope 2 Guidance sets the inventory rules. Dispatch only the jobs whose deadlines, residency and customer permissions allow the shift, and price that permission in Chapter 15.8.