Renewables and power purchase agreements

The largest data center operators are also the largest corporate buyers of renewable energy in the world. Amazon, Microsoft, Meta and Google together have contracted tens of gigawatts of wind, solar and other clean generation, driven by climate commitments, customer expectations and, increasingly, the economics of long-term fixed-price contracts. This lesson explains how data centers buy clean electricity: the accounting rules for claiming it, the instruments available (certificates, utility programs and power purchase agreements), the risks in those contracts, the move from annual to hourly matching, and the policy incentives that shape the market.

Carbon accounting for electricity

Greenhouse gas emissions from purchased electricity are reported as Scope 2 emissions under the Greenhouse Gas Protocol. Two methods are required side by side:

• Location-based: consumption multiplied by the average emission factor of the grid where it is used. This reflects the physical grid and can only be reduced by using less energy, locating in cleaner grids, or making the grid itself cleaner.
• Market-based: consumption multiplied by the emission factor of the electricity the company has contracted for, as shown by energy attribute certificates. A company that buys certificates matching all its consumption can report zero market-based emissions.

An energy attribute certificate (EAC) represents the environmental attributes of 1 MWh of generation. In the US these are renewable energy certificates (RECs), tracked through regional registries such as PJM-GATS, ERCOT's REC program, WREGIS and M-RETS. States also use RECs for their renewable portfolio standards (RPS), creating compliance markets alongside the voluntary market used by companies.

Critics argue that buying cheap unbundled certificates from existing projects far away does little to reduce real emissions. The Greenhouse Gas Protocol has been revising its Scope 2 guidance, with proposals published in 2025 to require hourly matching and that certificates come from the same grid region as consumption, phased in over several years. Leading buyers have moved ahead of the rules, emphasizing additionality (contracts that cause new projects to be built), deliverability and time matching.

Procurement options

From simplest to most involved, data centers can buy clean electricity through:

In the US, virtual PPAs dominate corporate procurement in RTO regions, while green tariffs (such as those offered by Duke Energy, Georgia Power, Dominion and NV Energy) and utility-built generation dedicated to large customers are common in vertically integrated states. Some utilities now offer "clean transition tariffs" letting data centers pay for firm clean resources such as geothermal or nuclear.

Virtual PPAs

A virtual PPA (VPPA) is a contract for differences. The project sells its output into the wholesale market at the local market price. The buyer and project agree a fixed strike price; for each MWh generated:

• If the market price is below the strike, the buyer pays the project the difference.
• If the market price is above the strike, the project pays the buyer the difference.

The project receives a fixed price per MWh overall, which supports financing, and the buyer receives the certificates. The buyer continues buying its own electricity from its utility or retailer at market-related prices, so when market prices rise, its higher bill is partly offset by VPPA payments it receives, providing a hedge.

Risks in PPAs

PPAs typically last 10 to 20 years, and buyers take on several risks:

• Price risk: if market prices fall well below the strike for years, the buyer pays continually. Buyers who signed VPPAs in 2019 at $30/MWh have done well; some who signed at higher prices before prices fell did not.
• Shape risk: solar produces mostly at midday and wind often at night. As more renewables are built, prices fall when they produce (cannibalization), so the project's "capture price" is below the average market price. The buyer's settlement is based on the capture price.
• Basis risk: the project settles at its own node or hub, but the buyer's load pays the price at its own location. Congestion can make these differ greatly, so the hedge is imperfect.
• Volume risk: output varies with weather; fixed-shape contracts transfer this risk to the seller at a higher price.
• Negative prices: in hours of surplus, prices can go below zero; contracts specify whether the buyer pays the strike plus the negative price or whether the project curtails.
• Counterparty, development and change-in-law risks: the project may be delayed or fail, the counterparty may default, or rules (such as tax credits) may change.
• Accounting risk: some contracts must be treated as derivatives, marked to market in the buyer's financial statements.

From annual to hourly matching

Most companies have claimed "100% renewable" by buying certificates equal to their annual consumption, regardless of when the renewables produced. A data center runs around the clock, so a solar-heavy portfolio may cover midday hours many times over while nights are supplied by fossil generation. Google introduced the goal of 24/7 carbon-free energy (CFE): matching consumption with carbon-free generation on the same grid every hour. Its CFE score measures the fraction of hourly consumption matched:

where \(L_h\) is load and \(G_h\) is contracted clean generation (plus the grid's clean share) in hour \(h\). Achieving a high CFE score requires a mix of resources that produce at different times (solar, wind, storage) and, beyond about 80 to 90%, firm clean resources such as hydro, geothermal, nuclear or gas with carbon capture.

import numpy as np

# hourly profile for one day: flat load, solar bell curve, steady wind
h = np.arange(24)
load = np.full(24, 100.0)
solar = np.where((h >= 7) & (h < 17), 300 * np.sin(np.pi * (h - 7 + 0.5) / 10), 0)
solar *= 1800 / solar.sum()          # scale to 1800 MWh per day
wind = np.full(24, 40.0)

def cfe(load, gen):
    return np.minimum(load, gen).sum() / load.sum()

print("volumetric match:", round((solar + wind).sum() / load.sum(), 2))
print("CFE solar+wind:  ", round(cfe(load, solar + wind), 3))

# add a 100 MW / 400 MWh battery, 90% round-trip, charging from surplus
def with_storage(load, gen, p=100, e=400, eff=0.9):
    soc = 0.0
    for day in range(2):          # day 2 starts with charge left from day 1
        matched = 0.0
        for t in range(24):
            if gen[t] >= load[t]:
                soc += min(p, (e - soc) / eff, gen[t] - load[t]) * eff
                matched += load[t]
            else:
                dis = min(p, soc, load[t] - gen[t])
                soc -= dis
                matched += gen[t] + dis
    return matched / load.sum()

print("CFE with storage:", round(with_storage(load, solar + wind), 3))

volumetric match: 1.15
CFE solar+wind:   0.637
CFE with storage: 0.803

With a realistic solar curve, generation in the early morning and late afternoon falls short of load, so the score is 64% rather than the 65% of the hand estimate. A 100 MW, 400 MWh battery charged from the midday surplus raises it to 80%; the remaining gap is in the night hours, which need more wind, longer storage or a firm resource.

Hourly matching also changes procurement: buyers value resources by when they produce, not just how much, and pay premiums for evening and night generation. Analyses suggest that pursuing 24/7 CFE does more to reduce grid emissions per dollar spent at high match levels than buying surplus annual certificates, because it drives investment in storage and firm clean resources.

Incentives and policy

Federal tax credits have been the main US support for clean electricity, and changed substantially in 2025:

• The Inflation Reduction Act (2022) created technology-neutral production (45Y) and investment (48E) tax credits for zero-emission generation from 2025, worth up to about 30% of capital or a per-MWh credit with bonuses for domestic content and energy communities.
• The One Big Beautiful Bill Act (July 2025) ends these credits early for wind and solar: projects must have begun construction by July 4, 2026, or be placed in service by the end of 2027.
• Storage, geothermal, hydro, nuclear and fuel cells keep the credits longer, phasing down after 2033, but restrictions on components and ownership from "foreign entities of concern" apply from 2026.

The early expiry of wind and solar credits raised PPA prices and pushed developers to start construction before the deadline, while increasing the relative attractiveness of storage, geothermal and nuclear for data center buyers. State renewable portfolio standards, clean energy standards and state-level incentives continue to support clean generation in many states.

Building a procurement strategy

A data center operator's electricity strategy typically combines:

• Efficiency first: lower PUE and better utilization reduce the amount of energy to buy (DCE-320).
• Location: siting in clean, well-supplied grids lowers location-based emissions (DCE-410).
• A portfolio of contracts: PPAs with different technologies, locations and terms to spread shape, basis and price risk, plus utility programs where markets are not open.
• Storage and flexibility: shifting load or using batteries to improve hourly matching and reduce peak charges.
• Firm clean resources: long-term contracts for nuclear, geothermal, hydro or other firm sources (Lesson 10).
• Transparency: reporting both location- and market-based emissions and hourly matching where possible.

Lesson 6 develops the storage and firm clean resources needed for high hourly matching, and Lesson 9 the PPA terms that determine how much a contract really costs.