Summary

  • Meta reported 5,637 ML withdrawn, 3,123 ML consumed and 6,017 ML of volumetric restoration benefits for fiscal 2024, each quantity defined in a different document.
  • The 200% restoration goal is stated per watershed, while the tables print aggregate restoration across seven watersheds and withdrawal only per facility — so the goal's ratio cannot be recomputed for any watershed or any data centre.

Three numbers wearing the same word

The 2025 Environmental Data Index prints fiscal 2024 withdrawal of 5,637 ML, of which 4,145 ML at data centres and 1,492 ML at offices, and consumption of 3,123 ML — 2,974 ML at data centres, 149 ML at offices. It prints 1,704 ML withdrawn from areas of high or extremely high baseline water stress, and 748 ML consumed in those same areas. Restoration for 2024 is 6,017 ML, against 5,889 ML in 2023 and 2,352 ML in 2022 (Meta 2025 Environmental Data Index).

The same index draws its own perimeter line in words worth carrying: Not included in our 2024 water withdrawal numbers are an additional 1,019 mega liters of water withdrawn for the construction of Meta data centers. That is 1,019 ML of water taken, printed by the company, sitting outside the figure that pairs with everything else here.

Three definitions in one note

Meta's restoration approach paper defines the objects the numbers attach to. Withdrawal, or water use, is the total amount of water withdrawn from its source. Consumption is the portion of water use that is not returned to the original water source after being withdrawn, and occurs when water is lost into the atmosphere through evaporation or incorporated into a product and is no longer available for reuse. Restoration, sometimes called replenishment, is an activity resulting in restoration of a volume of water to a local watershed. Volumetric water benefits are the volume of water resulting from water stewardship activities, relative to a unit of time, that modify hydrology beneficially and/or help reduce shared water challenges (Our approach to water restoration).

Read those four sentences together and the arithmetic stops being obvious. Water taken and water not returned are different physical amounts at the same site. Water returned is an activity in a watershed, credited over time, produced by projects that may be funded rather than built. A single percentage joining the second and third is a statement about a portfolio, not a measurement of a place.

How each number is made

Withdrawal is the metered-or-proxied quantity: utility or meter data from local utilities and aquifers, using existing activity data or public gallons-per-square-foot intensity proxies where activity data is incomplete or unavailable. Consumption is derived from it — calculating withdrawal minus wastewater discharge, or by calculating consumption from cooling-system cycles of concentration — and at offices it is estimated from industry averages (Meta 2025 Environmental Data Index).

So the denominator a consumption-based goal rests on is partly a subtraction, partly a cooling-system calculation, and at offices not Meta's measurement at all. That is disclosed, not hidden. It does mean the figure has three production methods inside one total, and the total prints no split between them.

The same quantity, counted twice, in two units

The data index prints 2024 restoration as 6,017 ML. The volumetric benefits report prints the 2024 total as 1,589.5 million gallons (Volumetric Water Benefits: 2024 Report). At 264,172 US gallons to the megalitre, 6,017 × 264,172 = 1,589,542,000 gallons. That is 1,589.5 million gallons — my arithmetic, and it closes exactly. The water page rounds up to more than 1.6 billion gallons (water page); the sustainability report rounds down to over 1.59 billion gallons (Meta 2025 Sustainability Report). Three printings, one quantity.

What the goal actually names

The target is stated against consumption: 200% restoration in high-water-stress regions and 100% in medium-water-stress regions (Our approach to water restoration), repeated as a 2030 goal on the sustainability home page (sustainability home). Projects must pass a hydrological-connection test to Meta's operations, with Hydrosheds Level 3 or Level 4 as the starting point for watershed boundaries and the location of the source water consumed also considered. Water-risk ratings use WRI Aqueduct metrics plus local conditions, utility reliability and third-party knowledge, refreshed every three years.

That is a goal written in the unit of a place, with a boundary rule and a risk instrument attached. It is the most specific thing in the set.

What the tables print, and what they do not

Restoration is 23 Meta-supported projects generating 1,589.5 million gallons in 2024, across seven watersheds; through June 2025 the company had funded over 40 projects in the source watersheds of 10 data centres and facilities with medium to high water risk. Results are presented by watershed, with named data centres under each watershed section — but the retrieved text does not print a restoration volume per watershed.

On the other side, the facility table prints 2024 withdrawal only: Altoona 242 ML, Clonee 571 ML, Fort Worth 311 ML, Prineville 328 ML, leased data-centre facilities 883 ML. No facility-level consumption. No stress band attached to each facility.

My arithmetic on the printed aggregates, and nobody else's: 6,017 ÷ 3,123 = 1.93, restoration equal to 193% of total fiscal 2024 consumption; 6,017 ÷ 748 = 8.04 against consumption in the combined high and extremely-high band. Neither is the target, neither is a per-watershed ratio, and neither tells you anything about the watershed beside a named data centre. They show only that the portfolio total clears 100% comfortably and the band-level figure clears the 200% the goal specifies for high-stress regions.

The clock and the cost share

Volumetric benefits are first counted in the year the project begins achieving volumetric water benefits, and in each subsequent year, for up to a maximum of ten years, provided the project is maintained and confirmed to function as intended (Volumetric Water Benefits: 2024 Report). Where a project has several funders, Meta adjusts the benefit to reflect its financial contribution relative to total project cost. Quantification follows Reig et al. (2019), with methods depending on project objectives, activities and available data; Meta acknowledges estimation uncertainty and says it applies conservative assumptions.

A numerator with a ten-year window and a funding-share adjustment is a different kind of object from a denominator billed by a utility. The rule is stated plainly. Its consequence is arithmetic: a portfolio credit has to keep being renewed to stay level, while consumption does not.

Two restatements on two clocks

The 2026 report landing page gives performance through fiscal 2025 as cumulative figures — 62 restoration projects in 19 priority watersheds since 2017, and 2.9 billion gallons restored to high- and medium-water-stress regions since 2017 (sustainability report landing page). The water page still says over 40 projects across nine watersheds and cites 2024 restoration above 1.6 billion gallons. Different periods, different counts — cumulative since 2017 against a single-year flow — and the page appears not to have been updated. No fiscal 2025 absolute withdrawal or consumption total was obtained. The Meta Platforms entry in our directory is here.

What this reading covers

Four PDFs and three web pages. All PDF text came through search indexing; page layout was not visually inspected, and per-category rows in the extracted table came out garbled, so only the totals, the data-centre and office split, the stress-band figures and the named-facility withdrawal values are used. Neither the restoration approach paper nor the volumetric benefits report prints an explicit accounting rule stating that credits are booked only in the exact watershed where a project sits; that absence is scoped to the seven documents read, not to everything Meta has published.

The gap is the granularity, not the water

Set the documents side by side and the same structure appears in each: a company-authored definition, a company-selected perimeter, a company-funded credit, all internally consistent and all stated. What no document does is print the goal's two sides at the goal's own scale. Aggregate restoration against aggregate consumption is a division a reader performs; per-watershed restoration against per-watershed consumption is the division the pledge asks for, and it is not in the set.

A reader who wants to know what that pledge means for the watershed beside the Altoona data centre can hold all seven documents and still not run the numbers. That is a granularity gap. It says nothing about whether the water came back.