Hour of Need:
Every hyperscaler claims to run on renewables. One of them publishes the hour-by-hour truth, and it has not improved since 2019.
EXECUTIVE SUMMARY
- Google has matched 100 per cent of its electricity consumption with renewable purchases every year since 2017. On an hourly, grid-by-grid basis it has been stuck at roughly 65 per cent for six years.
- In 2025 it signed more than twelve gigawatts of net-new clean energy, over eight times its 2019 volume. The hourly number did not move. Procurement is not the constraint.
- We introduce the Attribute Spread: the gap, in percentage points, between an annual-matched claim and hourly-matched delivery, by grid. It runs from four percentage points in Oklahoma to ninety-six in Singapore.
- We reject “100 per cent renewable” as an investable statement. It is an annual accounting identity, and the GHG Protocol’s Scope 2 revision proposes to replace it with hourly matching and deliverability.
- The position: own firm carbon-free generation on grids that possess both a third-party transmission access regime and an accredited hourly certificate registry. That combination is far rarer than either component, and it is not currently priced.
RENOWNED VINTAGE
The genius of the appellation system was never the wine. It was the date on the label.
Before vintage, a claim about quality was a claim about a producer’s reputation, which is to say a claim about nothing falsifiable. After vintage, a bottle asserted something about a specific harvest in a specific place, and the assertion could be checked against the weather records. The result was not better wine — the 1816 was still dreadful, being the year without a summer — but it was better information, and better information reorganised the entire trade around it. Négociants who had traded on relationships found themselves trading on evidence.
Electricity has spent two decades being sold without a vintage. A megawatt-hour of wind generated in Texas in April and a megawatt-hour of coal burned in Singapore at three in the morning have, for accounting purposes, been permitted to cancel one another out. Nobody has been defrauded. Everybody has been slightly misled, including the buyers.
That is about to end, and the reason is arithmetic rather than virtue.
PURCHASING POWER
Corporate renewable procurement is generally regarded as a solved problem, and the evidence for that view is substantial.
Google has matched 100 per cent of its global annual electricity consumption with renewable energy purchases every year since 2017, nine consecutive years, through more than 240 agreements for over 35 gigawatts of clean energy since 2010. It is also, by some distance, the most efficient operator of the assets consuming that power: a fleet-average power usage effectiveness of 1.09 against an industry average of 1.54, and a claimed six times more computing delivered per unit of electricity than five years ago. Whatever one thinks of the sector, this is not a company that has been idle.
Nor is it alone. The corporate power purchase agreement has become one of the more successful financial instruments of the past decade, underwriting genuinely additional generation at a scale that public subsidy never achieved. The consensus that this works is not foolish. It is simply measuring the wrong thing.
FLAT LINING
In 2020 Google set itself a different target: to match its electricity demand with carbon-free supply every hour of every day, on every grid where it operates, by 2030. It then did something unusual, which was to publish the result.
The result is a flat line. Sixty-one per cent in 2019. Sixty-seven in 2020. Sixty-four in 2023. Sixty-six in 2024. Approximately sixty-five in 2025.
Chart 1. The Flat Line. Hourly carbon-free energy against cumulative contracted clean capacity.
Sit with the shape of that chart for a moment, because it is the whole argument. Over the same period the company signed more than twelve gigawatts of net-new clean energy in 2025 alone — over eight times its 2019 volume — and cumulative contracted capacity rose to above thirty-five gigawatts. The procurement line goes up and to the right at a slope that would satisfy the most demanding sustainability committee. The delivery line does not move at all.
The rest of the 2025 disclosure is of a piece. Electricity consumption rose 37 per cent, the largest annual increase the company has recorded. Total emissions rose 18 per cent to roughly 14.5 million tonnes of carbon dioxide equivalent, and now sit 81 per cent above the 2019 baseline. Electricity-related emissions fell 3 per cent, against 12 per cent the year before. Google’s own language is that the buildout is accelerating faster than the grid is decarbonising, and that the goal is getting harder.
We would put it more bluntly. Buying more of something that arrives at the wrong hour does not make it arrive at the right one.
THE ATTRIBUTE SPREAD
This is the series we intend to run and we have not seen it published anywhere.
The Attribute Spread is the gap, expressed in percentage points, between an annual-matched claim of 100 per cent and hourly-matched delivery on a named grid. It is the quantity that a buyer is implicitly asserting does not matter, and the quantity a regulator is about to decide does.
Chart 2. The Attribute Spread. Hourly delivery by grid region, against a uniform annual claim.
The dispersion is the point. Latin America runs at roughly 92 per cent, Europe in the low eighties, North America around seventy. Asia-Pacific averages about twelve. Japan and Taiwan sit near seventeen. Singapore has been as low as four. Oklahoma has reached ninety-six. In 2024, nine of twenty grid regions achieved at least 80 per cent, which is another way of saying eleven did not.
One company. One annual claim. A range of ninety-two percentage points in what that claim actually delivers.
Read commercially rather than environmentally, that dispersion is a rent. It says that carbon-free hours are abundant in some places and effectively unobtainable in others, and that the market has not yet troubled to price the difference because no rule has required it to. Rents that exist but are not priced are the most reliable source of return we know of, and they persist precisely until the accounting changes.
THE NUMBER WE ARE KILLING
“One hundred per cent renewable” is not a fact about electricity. It is an annual accounting identity, and we propose to stop treating it as investable information.
The reason this matters now, rather than as a philosophical objection, is that the accounting is being rewritten. The GHG Protocol’s Scope 2 revision proposes an hourly matching and deliverability requirement for market-based reporting, together with eligibility criteria covering generation vintage, additionality and data precision — a decisive move away from the position that any certificate counts. Its first consultation closed on 31 January 2026 with over 400 responses, and a second runs through 2026. The Science Based Targets initiative’s draft second version makes Scope 2 a separate target line with hourly matching required. The European Union’s Renewable Energy Directive already mandates granular guarantees of origin.
When that lands, a cohort of “100 per cent renewable” claims will require restatement. Not because anyone lied, but because the label will finally carry a vintage.
TESTIMONIAL
“We have known for years that annual matching was a convention rather than a measurement. What has changed is that auditors now know it too.”
— Navigate Portfolio Advisers
VIRTUE, THEN RULE
The discipline requires the rate of change on both sides, and both are moving in the same direction for once.
On demand, the buyer of granular certificates has until now been a volunteer. Shortly it will be a defendant. Demand for hourly attributes is about to be manufactured by regulation rather than by conscience, which is a considerably more durable demand curve. Hourly matching is at present largely the preserve of data centres and companies with the resources to capture very granular consumption data; a rule change makes it everybody’s problem.
On supply, the infrastructure barely exists. EnergyTag accredited its first granular certificate issuers only in June 2025. Hourly issuance is live through M-RETS and PJM in the United States, Energinet in Denmark, Taiwan’s T-REC system on fifteen-minute intervals since 2017, and the I-REC standard across more than fifty countries. That is a short list against the number of grids on which multinationals operate. Africa, to take the case we know best, has functioning annual certificate markets — zaRECs in South Africa, governed alongside the I-TRACK standard — and effectively no accredited hourly issuance at all.
A demand curve created by rule, meeting a supply of issuance infrastructure that is four years behind it, is not a subtle setup.
LET ELECTRONS BE ELECTRONS
The consequence for generation is more interesting than the consequence for certificates and it is already visible in what the most sophisticated buyer is doing with its money.
Intermittent renewables cannot close an hourly gap; adding a fifth solar farm to a grid that is already saturated at noon changes nothing about three in the morning. So procurement has pivoted toward firm carbon-free generation. Google has contracted advanced nuclear through Kairos Power and the Tennessee Valley Authority, delivering fifty megawatts from 2030. It has agreed with NextEra to bring Iowa’s only nuclear plant back into service. It has taken enhanced geothermal at Fervo’s Cape Station in Utah, at nearly four hundred megawatts, building on the first corporate next-generation geothermal agreement signed in 2021. It is buying long-duration storage, funding transmission, and placing early bets on fusion.
None of that is a sustainability gesture. It is a buyer discovering that the marginal product it requires is not clean energy but clean energy at four in the morning and that almost nobody sells it.
FIRMING UP ARRANGEMENTS
The asset to own is not renewable generation. It is the combination of three things, and the combination is much rarer than any one of them.
Firm, dispatchable carbon-free generation — geothermal, hydro with storage, nuclear, or renewables genuinely firmed with storage — rather than intermittent capacity sold at an annual average.
On a grid with a working third-party transmission access regime, because an electron that cannot be wheeled to the buyer cannot be sold to them at any price.
In a jurisdiction with accredited hourly certificate issuance, or a credible path to it, because without that the clean attribute cannot be proved by the hour and therefore cannot be paid for by the hour.
South Africa has the wheeling regime. Kenya has the clean grid. Zambia has legislated third-party access. The United States has the certificate registries. Very few places have all three, and the places that acquire all three first will find a queue of buyers with a compliance obligation and no alternative.
The risks are worth naming. The Scope 2 revision may soften; consultations frequently do, and the current position is that it does not yet mandate hourly matching and does not move procurement targets. Several buyers are visibly holding back until the rules settle, which delays the repricing. And a sufficiently aggressive build-out of storage could compress the Attribute Spread faster than we expect, which would be excellent for the planet and inconvenient for the trade.
CONCLUDING THOUGHTS
Vintage did not improve the wine. It improved the questions, and the questions reorganised the money.
We have written elsewhere, in The Long Lead, that the megawatts required by the artificial intelligence buildout cannot physically arrive on the announced schedule, because the equipment that energises them is ordered two and a half years in advance. This is the same constraint approached from the other side. Even where the megawatts do arrive, they cannot presently be clean at the hour they are consumed, on most of the grids where they are consumed, and the accounting that has concealed that fact is being withdrawn.
Energy competition, as much as transition; displacement as much as replacement. The house has said as much for some time. The Attribute Spread is what it looks like with a number attached.
The vintage is coming to electricity. Most cellars have not been audited.
TESTIMONIAL
“Nobody is going to build an hourly registry out of goodwill. It will be built the week after the first company is told its disclosure is non-compliant.”
— Navigate Portfolio Advisers
SOURCES AND NOTES
Google environmental reporting, 2020 to 2026, including disclosed carbon-free energy percentages by year and by grid region. Uptime Institute Global Data Center Survey 2025 for industry-average power usage effectiveness. GHG Protocol Scope 2 revision consultation materials, first consultation closed 31 January 2026. Science Based Targets initiative draft version 2.0. EU Renewable Energy Directive provisions on granular guarantees of origin. EnergyTag scheme and matching standards, and first issuer accreditations, June 2025. M-RETS, PJM GATS, Energinet, Taiwan T-REC and I-TRACK Foundation issuance disclosures. zaRECs and South African EAC market documentation. Google announcements with Kairos Power and the Tennessee Valley Authority, NextEra, and Fervo Energy. Chart 1’s cumulative contracted capacity series is a Navigate estimate derived from disclosed cumulative and annual volumes and is flagged as such on the chart; 2021 and 2022 carbon-free percentages are not disclosed on a comparable basis and are not plotted. Chart 2 mixes disclosed regional averages with individually disclosed grid regions to show the observed range, and is labelled accordingly. Prior in this argument: “The Long Lead” (The Energisation Gap, No. 1); “How To Invest For An Electrified Society”; “Trading Places: the Net Zero Fiscal Trilemma”.
