Microsoft Community Commitments – A Global Promise Kept, A Local One Still Owed

The Standard Applied: A Test That Must Survive Being Turned Around

Greyhound Research puts that two-part test to Microsoft in what follows. Applying it is easier to write about than to live with, because the person asking depends on the same infrastructure. The column for The Hindu BusinessLine admitted as much before it made a demand of anyone, in a passage about a hillside cut for a road. That admission belongs at the head of this report too, in the column’s own words:

This is not an argument for an untouched, utopian world; none of us opines with clean hands. The desks we write from rest on quarried stone, and the hill roads we love on holiday were blasted through slopes like that one.

Growth is part of being. The argument is only about how we grow, and what we will be proud to hand down.

That dependence is not a reason to go easy on the account. It is why the account matters, and it fixes the terms of what follows, because a standard worth applying to Microsoft has to survive being applied to the people applying it.

Supporting cloud and AI growth means asking how it improves the lives of the people who host it. It also means holding companies and public authorities responsible for the decisions they control. Those terms have to be settled while a company can still change what it plans to build, and written down so they can be checked later. Exhibit 2 sets out the two tests this report applies and the evidence that would settle each. They apply to any operator, and Microsoft’s own foreword has already accepted the first half for water, saying that replenishing global volumes is not enough.

↑ Back to the section guide

Microsoft’s Commitments: Specific Enough To Be Held To

Greyhound Research reads the commitments in the terms Microsoft has set for itself, and Microsoft’s own accounts of individual sites show why those terms cannot be identical everywhere. In the Netherlands, it began planting native trees and shrubs around its Middenmeer campus in 2022, the start of a long-term effort to blend its datacenters into the natural landscape of the Wieringermeer polder. For its Western Sydney site, the Indigenous-owned organization Indigital worked with the Dharug people on the site’s artwork and landscape design. Microsoft’s Canadian commitments include support for a national network of Indigenous AI Fluency and Workforce Readiness Hubs.

Consulting early enough to change a facility’s size, design, or location gives the engagement a purpose. The same sensitivity has to shape the benefits a company promises. A community worried about drinking water needs to know what will happen to its supply. Training may be valuable, but its success cannot settle the water question. Each commitment has to answer the concern that gave rise to it.

Microsoft’s commitments are now specific enough to be tested one by one. The January 2026 Community-First AI Infrastructure announcement made five to communities in the United States, and an April announcement applied the approach in Canada, where Microsoft notes that electricity systems are provincially governed and solutions vary by region while the commitment itself stays consistent nationally. Earlier and later statements add a cooling design, a leased-facility scope, a reported milestone, and, in March 2026, an end to non-disclosure agreements with local governments.

Exhibit 3 places all eleven commitments on one page, by where each stands on Microsoft’s published evidence. Tables 2 to 4 then take them one at a time. Each row sets a commitment against what Microsoft has published for it, what that evidence cannot show, and Greyhound Research’s assessment, with the one disclosure that would change it. Water comes first because it is where Microsoft has published most.

Exhibit 3 and the three tables use the same three states, each named in words as well as shown in color: kept on the published evidence, kept in part or unreconciled, and not testable on public evidence.

Water

Five commitments run from a pledge to replenish water in the same districts to a cooling design that uses none. They are the most specific and the best documented of the eleven, and the only ones a reader can test against published numbers today. Table 2 finds one kept for the owned estate, three kept in part, and one that cannot be tested at any leased site.

Greyhound Research reads four of the five as having a number behind them and the fifth as having only a statement. The pattern across all five is the same. What Microsoft reports is company-wide, what it promised is local, and the two have not yet appeared on the same page.

Electricity

Two commitments cover electricity, both from the January announcement. They are specific, and they answer a question about buying power rather than the question the price pledge raises. Table 3 shows one kept in part on Microsoft’s own measures, and a price pledge no ratepayer can test.

Greyhound Research treats the record behind them as real and large. That record establishes that Microsoft bought as much renewable energy in the year as it used. It does not establish that a local grid had power in the hour a facility needed it. On connection costs Microsoft has stated its practice, undertaking to pay for the transmission and substation improvements its own expansion requires, and it publishes no connection where that can be checked.

The community’s return, and how the decision is made

Three commitments name jobs, the tax base, and local investment. The fourth is Microsoft’s March 2026 decision to share its datacenter plans openly and end non-disclosure agreements with local governments, which governs whether the other three are ever discussed before a site is fixed. Table 4 shows that three of the four turn on a number Microsoft already holds and does not publish by location.

Of the eleven, the tax pledge is the easiest to audit, and nobody has audited it, this report included. The openness commitment matters most and is the least documented, because it decides whether a community is consulted while the design can still change.

↑ Back to the section guide

The Local Water Account: What Microsoft’s Own Table Shows

Greyhound Research tests the water commitments first, because Microsoft has published the most against them. A credible water account says which water it counts, where replenishment happens, when the benefit arrives, and how it is measured. Residents should be able to follow the line from the corporate total to their own tap.

That takes two connected views: the individual facility, and the watershed or supply system it draws on. A city boundary or a cloud region will not always capture the link. Facilities use different cooling systems and water sources, and sites in different towns may draw on the same river or aquifer. In comments to Network World in June, I argued for keeping efficiency and replenishment in separate accounts and for making water sources and drought plans visible. Reporting by facility makes that scrutiny precise.

Microsoft’s 2026 Environmental Data Fact Sheet is the closest thing to a local water account that the company publishes. Its Table 15, referred to here as Microsoft’s location table, lists 29 owned data center locations with FY25 electricity use, water withdrawal, the share of non-potable supply, and a replenishment column. Read closely, the table says three things the headline does not.

Where the replenishment landed

The table shows where the replenishment milestone landed, as Exhibit 1 sets out in full. Seven locations show a replenishment volume in the FY25 table: Quincy, Phoenix, San Antonio, Des Moines, Cheyenne, Querétaro, and Madrid. Eleven are marked as priority locations expected to deliver volumetric benefits by FY30 and not yet doing so, among them Boydton, Ashburn, Chicago, Dublin, Hollands Kroon, and Jakarta. The remaining eleven carry no current replenishment plan, including Singapore, Busan, Taipei, and every listed site in Sweden, Denmark, Austria, Poland, and Italy.

By Greyhound Research’s arithmetic from the table, the seven locations that show replenishment withdrew 3,179 ML and show 6,247 ML replenished. The other 22 withdrew 1,538 ML and show none. Replenishment shown exceeds local withdrawal at Phoenix, San Antonio, Madrid, and Querétaro. At Quincy, Des Moines, and Cheyenne it does not. That is a more useful picture than “replenishment surpassed withdrawals,” and it cuts both ways. There is real progress, more than covering local withdrawal, at four sites, including the two US locations with the highest withdrawals after Quincy. Nothing shows yet at 22 of the 29.

Two features of that distribution matter, and Greyhound Research reads both as the table’s own arithmetic rather than as a verdict on any site. The first is concentration. Quincy, Phoenix, Singapore, San Antonio, and Boydton hold 74 percent of the 4,717 ML. Sixteen of the 29 locations withdrew under 20 ML each, and the median site withdrew 19 ML.

The second follows from the first: the largest replenishment multiples sit on the smallest sites. Querétaro shows 59 times its local withdrawal and Madrid 34 times, on 19 ML and 15 ML respectively. Those multiples describe the small denominator more than the project. The same arithmetic makes Singapore the more consequential entry. It has the third largest withdrawal in the table at 423 ML, 99 percent of it non-potable, and no current replenishment planned.

How much water the table covers

The table also shows how much of Microsoft’s water it leaves out. The 29 locations withdrew about 4,717 ML in FY25. Microsoft’s company-wide water table, Table 8 of the same fact sheet, reports 13,266 ML withdrawn across the offices, labs, and data centers, owned and leased, under Microsoft’s operational control. The location table therefore accounts for roughly 36 percent of the water Microsoft reports withdrawing, as Exhibit 4 shows. Microsoft does not publish how the remaining 8,550 ML divides between leased data centers, offices, and labs. Leased capacity outside its operational control sits in neither table.

This is a boundary Microsoft set for itself. Its inventory boundary, stated in Section 1.7 of the same fact sheet, covers owned and leased real estate facilities and data centers under its operational control. Section 1.8 applies that boundary to water by name, covering owned and leased offices, data centers, and labs. So leased capacity sits inside the company-wide total and outside the location table, in the same document. On the public record, a community that hosts Microsoft capacity in a leased building has no local account at all.

What the accountant reviewed

Finally, the table shows what the accountant did not check. Microsoft’s independent accountant carried out a limited-assurance review, which is a narrower check than an audit. The review covers specified information in Section 1 of the fact sheet for the year ended June 30, 2025. It excludes information outside Section 1, information relating to prior periods, and information relating to targets, goals, and progress against goals.

Table 5 sets out what that means figure by figure. The location table sits in Section 2 and carries its own footnote saying the section was not subject to that review. As progress against a goal, the replenishment milestone is outside the review twice over. The FY25 withdrawal and consumption totals are inside it, while the FY24 and FY23 columns beside them are not, because they are prior periods.

On Greyhound Research’s reading, then, the figure Microsoft leads with is outside the review. The reviewed FY25 figure is a withdrawal of 13,266 ML, against 11,590 ML the year before. That is an increase of 14.5 percent, with consumption up 22.1 percent to 8,170 ML. The comparison year is itself outside the review, so the word “reviewed” covers the FY25 figure and not the growth rate. Exhibit 5 puts the reviewed series beside the unreviewed one.

Third-party quantification of individual replenishment projects is real and creditable. It is not assurance of the milestone, and Microsoft’s own documents make that distinction available. The accountant’s report sits in the fact sheet, and the 2026 report’s reporting-disclosure note records that Deloitte & Touche LLP performed a review relating to specified information within Section 1 of that fact sheet.

What the record credits

Two things in the year run in Microsoft’s favor, and Greyhound Research counts them. The 2026 report records that Microsoft strengthened its replenishment target to focus on total water withdrawals rather than consumption. That is a more rigorous basis, because it covers the full volume withdrawn and not only the portion evaporated for cooling, and values for previous fiscal years were recalculated on it. The milestone was therefore reached against a harder test than the one it replaced.

Microsoft also published more than it promised. The January commitment to publish by region was made for the United States, and the table Microsoft published is global.

One boundary remains open. The table does not say whether its replenishment volumes are the year’s or cumulative to date, or which of them are estimates rather than quantified benefits. Microsoft’s portfolio notes say lifetime estimates are used unless marked otherwise. That is the first of the ten questions in section 10. Until it is answered, read the local comparisons above as the table’s own arithmetic, not as verified annual balances.

↑ Back to the section guide

Credible Disclosure: Three Requirements, None Needing New Science

Greyhound Research proposes three requirements for a local water account, and each can be checked against what Microsoft and AWS publish today. None of the three requires new science, and one requires no new measurement at all.

1/ Publish the volumes and their local context. Water withdrawal and consumption are different measures. Withdrawal is water taken from a river, lake, or aquifer, which hydrologists call a surface-water or groundwater source, and a facility’s account should also identify water delivered by a utility or other supplier. Part of it returns to the water system. The rest is consumed through evaporation or other processes that make it unavailable for immediate use.

Microsoft’s company-wide account shows withdrawal, discharge, and consumption, and consumption was 62 percent of withdrawal in FY25. Its location table shows withdrawal only. A resident cannot tell from it how much of the water taken at Quincy or Phoenix returned to the system, in what condition, or where.

Efficiency also needs context. Greyhound Research reads an efficiency ratio and a total volume as answers to different questions. A data center can use fewer liters per unit of IT electricity while drawing more water overall as capacity grows, much as a car can get better mileage and still burn more fuel on longer trips. Microsoft reports a global average water usage effectiveness of 0.27 liters per kilowatt-hour in FY25. It also reports a 25 percent reduction at its owned data centers since 2022, against a 40 percent target for 2030 that covers the owned fleet.

Between FY22 and FY25, company-wide withdrawals rose from 8,698 ML to 13,266 ML. The improvement is real, and so is the additional demand. Residents need total volumes alongside efficiency ratios.

Amazon Web Services (AWS) deserves credit for publishing data that makes these distinctions easier to examine. Its regional disclosures, read again on October 19, 2026, report water usage effectiveness averages of 0.02 liters withdrawn per kilowatt-hour of IT load in Ireland and 2.85 in Jakarta for reporting year 2025. The year before, the figures were 0.03 and 2.75.

The table now carries explicit year columns, which it did not when this report was first drafted. AWS has since added a geography view giving a global average of 0.12 liters per kilowatt-hour for 2025. These regional averages show how much location matters. On their own they cannot rank the environmental harm of individual facilities, and the regional table gives efficiency without the volume beside it.

The two accounts are not like for like. Microsoft’s intensity target covers the fleet it owns, while AWS’s water-positive goal covers owned, leased, and colocation capacity. The two companies also draw the local unit differently, Microsoft as water districts and data center regions, AWS as river basins. Exhibit 6 sets out what each company publishes at each stage of the water account, and what neither publishes by site.

Greyhound Research applies the same standard to AWS as to Microsoft. Regional intensity and a stated methodology do not, on their own, establish the outcome at a particular site. For both companies the test is coverage, local relevance, timing, and verification. Because their definitions and boundaries differ, these sources cannot support a single league table of providers.

Utilities should use facility information to assess the combined demand from operating sites, proposed developments, and other users before supporting expansion. Separate approvals are not enough if nobody adds up the combined demand on the shared water system.

2/ Show when replenishment delivers a benefit. An annual total can conceal a difficult summer. A project that improves long-term storage may be valuable while doing little to relieve an immediate shortage. Microsoft’s same-district commitment in the United States now has a location table to be tested against. The next disclosure should connect it to seasonal demand, delivered benefits, and drought response in each district. A global FY25 result cannot show whether residents’ supply is protected when pressure is greatest.

Greyhound Research also expects replenishment to be described accurately. Repairing a leaking municipal pipe can leave more usable water in the supply network. Where local hydrogeology supports it, restoring a wetland can help rain soak back into the ground and refill underground water, which hydrologists call groundwater recharge. These interventions have different beneficiaries and timescales.

Microsoft’s FY25 portfolio lists 98 projects by type, partner, basin, duration, and estimated benefit, and its notes single out leak-detection projects, of the kind Microsoft has funded around Phoenix and Dublin, as first-of-their-kind work with large volumes and potential future variation. A claim about the local water table needs evidence of recharge, with rainfall and other withdrawals accounted for.

There is useful guidance to build on. The World Resources Institute’s Volumetric Water Benefit Accounting 2.0 emphasizes local conditions, defensible methods, appropriate timescales, attribution, and avoiding double counting. It is voluntary guidance, rather than proof that a particular project has met its claims.

Greyhound Research expects a local account to distinguish contracted projects from those delivering benefits, label estimates, and explain what a contribution adds beyond what would otherwise have happened. Where several organizations fund a project, their shares of the claimed benefit need to be clear. Future replenishment cannot justify taking water today from a source that cannot sustain the demand. Where that is the constraint, the design, supply arrangement, or site needs to change.

3/ Assess water savings alongside electricity demand. Cooling choices need to be assessed against both resources. In its December 2024 cooling-design announcement, Microsoft explained that replacing evaporation with mechanical cooling involved an energy tradeoff, and it described engineering measures that limited the expected increase. That is a design-specific assessment, not a rule that every reduction in cooling water raises electricity use by the same amount.

The boundary of a claim matters too, and Greyhound Research does not read a zero-water cooling design as a zero-water footprint for the whole operation. Administrative uses remain, and electricity generation can carry its own water demand, a relationship examined in Berkeley Lab’s data center energy research, updated in June 2026. The local assessment should show the cooling design’s electricity implications and account separately for relevant indirect water use.

Microsoft already holds the fact that would make this visible. Its fact sheet estimates water for locations without primary data from electricity use and cooling technology type, so the cooling type of every listed location is known to the company. A cooling-type column in the location table would let residents see which sites still evaporate water and which do not, at no research cost to Microsoft.

The practical approach will differ by location. AWS reports that it has piloted and deployed a water recycling system in Singapore with local company Hydroleap, with plans to scale it across the Asia-Pacific region. That is the kind of local partnership that changes what an operating facility draws. It does not establish that any facility is water positive, and AWS names no site, so the benefit cannot be placed on a local account. Singapore’s August 2026 standard for liquid cooling in tropical data centers takes the local engineering principle further, addressing operation in heat and humidity alongside water-use measurement, with outcomes still to be measured.

↑ Back to the section guide

Electricity Procurement: A Real Record Answering A Different Question

Greyhound Research reads Microsoft’s electricity commitments against the same standard, and the cooling discussion leads straight to them. A design choice inside the facility has to work with the network outside it. In Ireland’s December 2025 connection-policy decision, the Commission for Regulation of Utilities required system operators to consider whether a requested data center connection is in a constrained location, where the network is already near its limit, or an unconstrained one. ESB Networks’ implementation process, effective May 2026, assesses constraints at the specific substation as well as the wider planning zone.

Greyhound Research weighs Microsoft’s electricity commitments against that precedent. “We’ll pay our way to ensure our datacenters don’t increase your electricity prices” is, in the announcement’s own terms, a request to utilities to set rates high enough to cover data center costs, collaboration on infrastructure expansion where it is needed, efficiency work, and policy advocacy. Each of those depends on a regulator or a utility, and none carries a metric or a disclosure.

Microsoft has said publicly how it intends to keep the wider promise. The January announcement commits it to work with local utilities to add electricity and the supporting infrastructure to the grid when its data centers need it.

The 2026 report records 100 percent of annual global electricity consumption matched with renewable energy in FY25, with agreements for up to 40 GW of new renewable capacity across 26 countries, 19 GW of it online, and hourly matching at one new Washington site.

The January announcement puts numbers behind the second of its four electricity undertakings. In the Midcontinent Independent System Operator market, Microsoft says it has contracted to add 7.9 GW of new electricity generation to the grid, more than double its current consumption there, and it undertakes to keep paying for the transmission and substation improvements its expansion requires.

One new campus takes a different route. Microsoft’s June 2026 announcement for Pecos, Texas says the campus will open on a co-located natural gas plant that serves it directly and independently of the public grid, with a connection to the wider grid anticipated over time. Microsoft is paying for that generation, and says the arrangement means the campus’s demand does not take from the current grid. It also means the power at that site comes from gas in every hour it runs on the plant, whatever the company-wide annual matching total shows.

Greyhound Research counts the quality of that matching in Microsoft’s favor, because not every 100 percent is the same. Microsoft’s foreword records that the company paused its use of non-additional, unbundled renewable energy certificates. These are certificates bought apart from any power purchase, which add no new generation to the grid. Microsoft paused them to prioritize investments that bring new power to grids, and accepted higher reported emissions in the near term as a result.

The foreword states that more than 90 percent of the renewable energy applied to the 2025 target came from contracted projects rather than grid mix, and that the target excludes short-term certificates bought on the spot market. Microsoft’s fact sheet puts numbers on that split. Contracts and on-site generation cover 34.5 million MWh of the 37.0 million MWh Microsoft used in FY25, about 93 percent. Its 100 percent metric also counts the renewable portion of the grid mix, which the contracted figure leaves out. A company can reach 100 percent on paper. Microsoft took the more demanding route to it, and that deserves recording before anything is said about what the figure does not settle.

Those are real commitments with a public record behind them, and they answer a different question from the one the price pledge asks. Annual matching means Microsoft bought as much renewable energy in a year as it used. It does not show that the local grid had power, renewable or otherwise, in the hour the facility needed it. Exhibit 7 shows the difference on one illustrative day. On who pays for the substation, Microsoft has answered in principle, undertaking to continue paying for the transmission and substation improvements its expansion requires. No connection is published where that undertaking can be checked.

In Greyhound Research’s assessment, the price pledge cannot yet be tested by a ratepayer, the household or business that pays the utility bill. What would make it testable is ordinary: the tariff class Microsoft is served under at each utility, the cost-allocation terms of each new connection, and any rate case, a regulator’s review of what a utility may charge, in which Microsoft has taken a position. Published alongside the regional water table, they would settle it. None of the three appears in the documents that carry the pledge.

An electricity commitment should be tested against the hours when the local system is under pressure. Annual procurement figures do not answer every question about whether the network can serve the facility at those times. The community and utility should be able to assess peak demand, the upgrades needed before each development phase, who pays for them, and any dependable reduction in demand the operator can offer during system stress. Proposed capability should remain distinguishable from a tested operating response.

This is also where Greyhound Research insists on precision about the enterprise consequence. A local issue affects a buyer’s capacity plan when it changes an identifiable dependency: a utility upgrade, an approval, an operating condition, or a delivery obligation. The buyer then needs to establish whether that dependency affects the location, amount, and date of capacity it requires, and whether a usable alternative exists. Community opposition alone establishes none of those links.

Better engagement can reveal a constraint early enough to change the design or sequence of investment. It cannot create spare water or electricity by itself. Equally, a dispute around a proposed expansion should not automatically cast doubt on an existing service with demonstrated supply and operating readiness. Physical adequacy, fulfillment of obligations, and public confidence are connected questions, each requiring its own assessment.

↑ Back to the section guide

Commitments Through Operation: A Record That Outlives The Consultation

Greyhound Research’s second test is durability. Water and electricity commitments leave communities with the same practical problem: how to follow what was promised once construction gives way to operation. Disclosure should begin with a baseline before construction. It should then let residents compare the promise with delivery and obtain a response when it falls short. The record needs to continue through expansion and changes of ownership.

CNA reported in September 2026 that a Microsoft infrastructure executive described a data center as a commitment to a market lasting twenty or thirty years. A consultation lasts months, and the record has to last as long as the building. The power contracts behind it are shorter: Microsoft describes its power purchase agreements as typically ten- to fifteen-year commitments. Exhibit 8 sets those timescales on one line, with a contract renewal falling partway through the life of the building.

Greyhound Research reads the durability problem as a problem of ownership rather than of data. Almost every figure a resident would want is already produced somewhere, by the operator, the utility, or the regulator. What is missing is a place where it stays attached to the site and to the promise made about that site. It has to survive a change of operator, a capacity increase, and the twenty or thirty years a site is expected to run.

A record that survives that long has to be built for people who were not in the room when the commitment was made. Greyhound Research proposes three features that make one usable. They draw on what regulators already require and what operators already measure, so the cost falls on assembling the record rather than on new instrumentation.

1/ Keep a continuing public record. Each facility should have a stable identifier connecting its owner and operator with resource demand, community commitments, measured outcomes, and corrective actions, updated when ownership or capacity changes. The record should distinguish announced capacity, reserved utility connections, construction, and actual operation.

Greyhound Research’s earlier global analysis of data center capacity examined the gap between speculative demand and real development. Communities should know which phase a benefit estimate describes and when the associated obligations become due. Microsoft’s dated FY30 promises are a small example of the form: a named site, a dated promise, and a column where the delivered figure will have to appear.

2/ Use comparable definitions and proportionate reporting. A public annual account with independent verification should be supported by quarterly updates, and monthly figures where seasonal stress warrants them. The schedule should be agreed locally, and methodology changes need an explanation so that comparisons remain meaningful.

3/ Define a proportionate response to missed commitments. Greyhound Research expects a shortfall to identify a responsible owner, funded corrective action, and a deadline, with safeguards in operating agreements, contracts, or permit conditions. The response should reflect the obligation and its materiality: a delayed community grant and an unmet water-supply condition require different remedies. Residents need an escalation route, agreed drought thresholds should trigger a defined response, and attendance at a consultation meeting cannot be treated as consent.

The rules already in force

Existing rules need to be described precisely, because they run on two channels and only one of them is well known. Under the EU data center reporting regulation, operators of data centers at or above 500 kW of installed IT power submit information, including total water input, to a European database. That database publishes in aggregate at Member State and Union level, and keeps individual-facility submissions confidential.

The second channel points the other way. The same regulation records that, under Article 12 of the Energy Efficiency Directive, Member States are to require owners and operators of data centers to make publicly available the information set out in Annex VII to that directive. A per-facility publication duty therefore already exists in European law, discharged by the operator rather than by the Commission. Neither obligation, by itself, establishes delivery of the community commitments attached to a site.

National initiatives can provide further public visibility. Germany announced the launch of its public data center register platform in August 2026, with aggregate statistics and the individual annual reports of facilities whose operators choose that publication route. The next step, in Germany and anywhere else, is to connect usable resource information with the community commitments, delivery record, and corrective actions attached to each site. Public information should be detailed enough to assess local impacts, with customer identities, security arrangements, and sensitive real-time operational information protected.

What moved in August and September

Between mid-August and late September 2026, that second channel moved. On September 21, 2026 the European Commission published its first report on the energy efficiency of data centers in the Union, drawn from the first reporting period. It carries the first official Union-wide water benchmark Greyhound Research is aware of: an average water usage effectiveness of 0.58 cubic meters per megawatt-hour across 458 data centers, with national averages ranging from 0.07 to 1.28.

On the same day the Commission adopted a delegated act creating an A to G sustainability label generated from data operators already report. The act redefines water usage effectiveness against freshwater input rather than potable water input, and sets the first labels for August 15, 2027. A public consultation on minimum performance standards opened alongside it and closes on December 14, 2026, with a legislative proposal expected in 2027.

The same weeks moved several United States jurisdictions, and they moved toward commitments rather than resources alone. California enacted seven data center statutes on September 21, 2026. Among them are a requirement to report actual annual water use and cooling-system type at license renewal, and a route to expedited environmental review that is conditioned on an enforceable community benefits agreement.

Massachusetts conditioned data center permits on a community benefits agreement by executive order on September 8, 2026. The order also prohibited non-disclosure agreements between state agencies and data center projects, except as the law provides. New Jersey legislated semiannual energy and water reporting with no minimum size threshold, signed on August 27, 2026. Texas paused state environmental permits for data centers on September 21, 2026, until grid and water audits are published. Tables 6 and 7 set the rules side by side, Europe first.

These rules are young and mostly untested. The newest of them, in California and Massachusetts, supply what the voluntary record still lacks: a named obligation attached to a named site.

↑ Back to the section guide

Leased And Partner Capacity: Where The Public Account Stops

Greyhound Research applies the same test to capacity Microsoft does not own, because following the record means following who operates the infrastructure. Cloud capacity can sit in a company’s own building, a leased facility, or a site run by a local partner. A community still depends on the same water supply and electricity network when the contract changes.

Greyhound Research expects community commitments to follow the capacity through the delivery chain. The company selling the service should explain which facilities are covered, who measures performance, and who must act when a commitment is missed. Exhibit 9 shows how far Microsoft’s public account currently follows its capacity.

Owned data centers appear in every instrument except the accountant’s review of the location table. Leased data centers under Microsoft’s operational control are inside the company-wide totals and inside the reporting boundary Microsoft states in Sections 1.7 and 1.8 of the fact sheet, and nowhere else. Capacity outside that control, leased or colocation, appears in none of them. Local project partners appear in the replenishment portfolio by name, with basins and durations, which is the strongest partner disclosure in the set.

AWS’s water-positive methodology covers owned, leased, and colocation data centers. It sets a target in every water-scarce basin where AWS withdraws more than 25 million gallons a year, alongside its global target. Microsoft’s stated boundary covers owned and leased capacity under its operational control, so the difference in declared scope between the two companies is colocation rather than leasing. That difference matters less than what the two have in common. Scope in a methodology is not evidence at a site, and neither company publishes the per-location numbers that would turn either boundary into an account a community can check.

Greyhound Research expects agreements with operators to require metering, consistent data, audit access, and corrective action. Shared facilities need a disclosed allocation method so that several customers cannot each claim the whole benefit. Reporting gaps should stay visible until resolved. Local organizations delivering water or community projects need scrutiny and support too. Money awarded, projects completed, and benefits sustained are separate measures, and a signed declaration cannot establish that a benefit continues to be delivered years later.

↑ Back to the section guide

The Community’s Economic Return: Right Evidence, Missing Baselines

Greyhound Research counts the community’s return separately from the facility’s own needs, because the question from the column remains: what does the community gain? The account should distinguish protecting existing users, funding the infrastructure the facility’s own demand requires, and creating an additional local benefit. Each contribution matters. Paying to connect a campus does not, by itself, establish a gain for its neighbors.

Microsoft’s January announcement offers two documented cases against which those distinctions can be tested. It cites more than USD 25 million of water and sewer improvements in Leesburg, Virginia, funded so that the cost of serving Microsoft’s facilities does not fall on local ratepayers. The announcement also cites the Quincy Water Reuse Utility in Washington. Microsoft’s location table separately shows that 74 percent of the water withdrawn at Quincy in FY25 was non-potable. Quincy is one of only three listed locations, with San Antonio and Singapore, where the table shows a non-potable share at all.

The questions for both cases are the ones the assessment should ask anywhere. Which households or other users benefit? What improvement would they have received without Microsoft’s contribution? When does the additional service become available, and who funds its maintenance? If the same work was already fully funded and would have happened unchanged, the claim of additional benefit needs to be reconsidered.

Where funding enables a shared improvement that would not otherwise have happened, Greyhound Research credits Microsoft for its attributable share. These are questions, not findings. The announcement gives the amounts and the places, not the baselines.

One case runs the other way, and it sits at a location the water table credits. Querétaro shows the largest replenishment multiple in Microsoft’s own table, 59 times its local withdrawal. Reporting by the Thomson Reuters Foundation in February 2026 examined a community program announced for the surrounding municipalities, 21 projects funded at USD 3.9 million and administered through a United Nations agency. Its reporters found the program almost entirely unbuilt when they visited in late 2025, and Microsoft did not answer their questions about it.

Greyhound Research has not independently verified the state of those projects. The point for this report is narrower and does not depend on doing so. A location can lead the water account and lag the community account at the same time, which is precisely why the two accounts have to be read separately. Exhibit 10 sets them side by side.

A public authority may legitimately choose shared infrastructure that serves several groups. The relevant question is whether the contribution produces an identifiable improvement for those beneficiaries while the development’s own obligations remain met. More jobs or a training program cannot be added to a total that cancels an unresolved drinking-water burden.

Greyhound Research holds the economic account to the same standard as the resource account. Residents should be able to assess what their community receives alongside the public resources committed to the project. Three habits would make that account credible.

1/ Count employment and local opportunity accurately. Separate construction work from sustained local employment, assess training against the opportunities it creates, and report local procurement with a clear definition of what counts as local. Microsoft’s Washington figures are the right kind of number and need that split. They are more than 1,300 skilled trades workers building data centers, and more than 650 full-time employees and contractors expected across its operational facilities there by the end of next year. The AI skilling investments Microsoft has announced, in the United States and in Canada, need the same treatment.

2/ Show the public contribution alongside the return. Explain actual tax receipts, subsidies, concessions, and publicly funded infrastructure, and identify who pays for maintenance and future expansion. Microsoft’s pledge to add to the tax base is, on Greyhound Research’s reading, the easiest commitment in its set to audit. A list of host jurisdictions, and any abatement or exemption received after January 2026, would settle it. Quincy’s county revenue, cited in the announcement as rising from roughly USD 60 million to more than USD 180 million, is one data point, not the test.

3/ Make demonstrated performance count in expansion decisions. Projects that meet agreed resource and community commitments should have a clearer basis for subsequent phases. Where an enabling condition is late, the affected phase should remain conditional on its completion. If a community benefit is late, the response should follow its agreed obligations and consequences. A credible delivery record can support confidence in expansion. Physical constraints or unresolved legal conditions can still determine the outcome.

↑ Back to the section guide

The Greyhound CXO Runbook: Turning Evidence Into A Decision

Greyhound Research turns the same evidence into buying decisions here, because these distinctions belong in the relationship between a cloud provider and its customers. It does not treat a disclosure gap as an outage, and a buyer that treats the two alike will misprice both. A buyer should respond differently to missing disclosure, an unfinished upgrade, and a demonstrated inability to support demand. A local issue reaches a buyer only through an identifiable dependency, so the first question is always which dependency, which capacity, and which date.

Greyhound Research proposes the decision rules in Table 8 for deciding how far to rely on a provider’s capacity. They sit alongside the separate obligation to deliver promised community benefits. Applying them to a particular facility requires site and service evidence beyond what this report provides.

Greyhound Research holds that a decision rule without a named owner is a preference rather than a control. The third column of Table 8 fails in one predictable way: it arrives after the workload has been placed, in a document nobody is accountable for reading. The sorting only works where a person with a budget and a deployment date can act on it, which is a question of process design rather than of evidence.

Greyhound Research therefore assigns each response to a role and a moment in the buying cycle, from qualification through renewal. Table 9 sets the five actions against their timing, owner, and the evidence that settles each, and the steps below explain them. They are written for the enterprise buyer rather than for the provider, because the buyer holds the only lever that operates on this evidence: what gets committed to which capacity, and when.

1/ Qualify the capacity before committing the workload. Technology and procurement teams should establish which locations and resource dependencies support the service. For Microsoft, that means asking which regions and availability zones run on capacity outside the location table and outside the company-wide operational boundary, under confidentiality where needed. Separate evidence gaps, which call for further diligence, from demonstrated constraints, which call for a different deployment decision. Every later step depends on this one, and the decision sits with the technology leader who approves the placement.

2/ Make readiness milestones part of rollout planning. The program owner should tie future capacity to the relevant utility connections, approvals, and commissioning milestones. If a necessary milestone slips, rephase the dependent workload or activate the agreed alternative instead of keeping an unsupported delivery assumption. The decision sits with the program owner.

3/ Negotiate disclosure and partner responsibility. Procurement should seek a named accountable provider, coverage of leased and colocation arrangements, and notification of material changes. Where a capacity commitment depends on a location that carries an FY30 replenishment promise, such as Dublin, Hollands Kroon, or Boydton, the disclosure schedule should name it and report against it. Agree what requires corrective action and which service consequences justify escalation. The decision sits with procurement.

4/ Test the independence of the fallback. Resilience teams should examine shared grid, water-supply, and critical partner dependencies, because Greyhound Research does not count an alternative exposed to the same limiting condition as an alternative. Where it is exposed, establish another delivery path or explicitly accept and manage the residual exposure. The decision sits with the resilience lead.

5/ Reassess at decisions that increase exposure. A named owner, supported by technology, procurement, and risk teams, should review delivery before renewals and capacity increases. Verified correction should count in the provider’s favor. Unresolved material dependencies should remain visible in the approval decision, which sits with the named owner of the provider relationship.

↑ Back to the section guide

Questions Ahead For Microsoft: Cheap To Answer, Expensive To Leave Open

Every finding in this report rests on what Microsoft has published, and every gap in it can be closed by Microsoft alone. Greyhound Research puts the ten questions below in the order the evidence raises them. Most ask for a column, a scope note, or a figure the company already holds. Two would change what this report can conclude: question 1, on the basis of the replenishment volumes, and question 3, on the two water intensity figures. Where Microsoft permits attribution, Greyhound Research will publish its answers as given, beside these questions, in the next revision of this report.

1/ What basis do the replenishment volumes in Microsoft’s location table use? Are they FY25 volumes or cumulative to date, and which are third-party quantified rather than estimated? The portfolio notes say lifetime estimates are used unless a project is marked otherwise, and the location table does not carry that boundary. Until it does, the local comparisons that make Microsoft’s strongest case, at Phoenix and San Antonio, cannot be relied on by the communities they concern.

If answered: The first row of Table 2 stops reading “Kept globally. Unproven in most districts.” and becomes a district-level claim a reader can test. Exhibit 1 can then show delivered volumes rather than a mix of volumes, promises, and blanks.

2/ Why is no replenishment planned at eleven of the listed locations? Greyhound Research reads this as the plainest gap in the table. Eleven of the 29 rows in Microsoft’s fact sheet Table 15 carry a blank in the replenishment column, which the fact sheet’s own footnote defines as no current replenishment efforts planned. Those eleven account for 487 ML of withdrawal. Is the reason low local water stress, a supply arrangement that makes replenishment unnecessary, or a site not yet assessed? A one-line reason code beside each blank would settle it, and it decides whether the blank records a considered position Microsoft can defend or a part of the program that has not started.

If answered: The eleven blank rows in Exhibit 1 acquire either a date or a stated reason. The coverage this report puts at seven of the 29 listed locations becomes a figure Microsoft has explained rather than one a reader has to infer.

3/ How do the two published intensity figures relate to each other and to the 40 percent target? The 2026 report states a 25 percent reduction in water usage effectiveness at owned data centers since the 2022 baseline. The local explainer states an 18 percent reduction in water intensity at operational data centers on the same baseline. The two differ in more than the number. Water usage effectiveness has a defined denominator, liters per kilowatt-hour of IT load, and water intensity does not, while owned and operational describe different populations of site. The explainer was modified again on October 15, 2026 and still carries the lower figure, so a stale page is not the obvious explanation. Greyhound Research holds that one target needs one metric, one baseline, and one scope, published where the rest of the water data sits.

If answered: The third row of Table 2 moves off “Reported twice, reconciled nowhere.” One metric on one baseline would also let a reader judge progress against the 40 percent target, which neither published figure supports on its own.

4/ How much of the water outside the location table is leased data center capacity? About 8,550 ML of the 13,266 ML withdrawn company-wide sits outside Microsoft’s location table, in leased data centers, offices, labs, and sites below the reporting threshold. That one split decides whether the local account covers about a third of Microsoft’s data center water or most of it, and it is the difference between a partial disclosure and a nearly complete one.

If answered: Section 7 stops resting on an estimate. The split decides whether Microsoft’s location table covers about a third of its data center water or most of it, which is the single number that sets the reach of every finding here.

5/ Do the community commitments travel to leased and partner capacity, and who answers for them there? Microsoft’s own reporting boundary covers leased facilities under its operational control, while the location table reports the owned estate alone. For capacity Microsoft leases or reaches through a partner, is the commitment set written into the agreement, is Microsoft or the operator the accountable party, and where is that recorded? Communities host sites whose operator they cannot name and enterprise buyers place workloads on capacity they cannot see, so the answer decides whom a resident or a customer approaches when a commitment is not met.

If answered: The fifth row of Table 2 gains an owner. Whoever answers for a commitment at a leased site is the person a community can hold to it, and until that person is named the commitment has no address.

6/ Which listed locations still evaporate water for cooling? This is the cheapest disclosure Greyhound Research asks for anywhere in this report. Microsoft estimates water for locations without primary data from electricity use and cooling technology type, so the cooling type of every listed location is already known to the company. California now requires actual water use and cooling-system type at license renewal for facilities in that state, so the disclosure is becoming routine whether or not operators volunteer it. A cooling-type column in the location table would show which sites the zero-water design has reached, at no research cost, and would turn a design announcement into a local fact that can be checked.

If answered: The fourth row of Table 2 moves off “A credible direction, not yet a local outcome.” A cooling-type column would also let any reader test the zero-water design claim at the site they live beside, which is where the claim was made.

7/ What are the tariff class and cost-allocation terms behind the price pledge? At which utilities is Microsoft served under a tariff designed to recover the full cost of its load, and where has it accepted the cost of a connection upgrade that would otherwise fall on other customers? Those two answers are the pledge in a form a ratepayer can check, and published beside the regional water table they would make the most quoted of the January commitments testable.

If answered: The first row of Table 3 moves off “Not testable by a ratepayer.” The price pledge is the only commitment in this report that a household can feel on its own bill and currently cannot check.

8/ Where does the renewable record meet the constrained hour? Of the 19 GW of contracted capacity now online, how much sits on the grids that serve the 29 listed locations, and beyond the Washington agreement, where is hourly matching in place? Annual matching is a procurement result and Microsoft has met it. Constrained-hour supply is what a utility and a neighbor actually experience, and it is the harder claim worth making.

If answered: The second row of Table 3 stops answering a different question from the one the pledge raises. Hourly matching on the grids that serve the listed locations is what turns a procurement result into a local one.

9/ What does the engagement record show, site by site? Greyhound Research treats this as the question that decides whether the practice is real. Microsoft’s March 2026 commitment to share its datacenter plans as openly as possible becomes testable the moment there is a record: what was disclosed, when, what the community asked, what changed as a result, and which announced site has not proceeded. There is now a candidate. On October 18, 2026 Bloomberg reported, citing The Telegraph, that Microsoft had withdrawn from a United Kingdom site where the grid connection was not expected until the 2030s, and Microsoft did not comment. If that withdrawal followed the open process Microsoft describes, saying so, with what the community was told and when, would establish the practice more firmly than the commitment that describes it.

Two things make the question sharper. An October 2026 report by three United States senators records that Microsoft has stopped seeking non-disclosure agreements with local governments while continuing to seek them with state-level agencies, and Massachusetts has now prohibited such agreements with its own agencies outright.

If answered: The fourth row of Table 4 moves off “Ended locally, kept with states and utilities.” A per-site record is also the only evidence that would survive a change of personnel at either Microsoft or the municipality.

10/ What did host communities receive, and what did they give up? Microsoft undertook in January not to ask municipalities to reduce local property tax rates and to pay its full and fair share, and at Mount Pleasant in September 2026 it gave up an incentive worth up to USD 5 million a year. The senators’ report cited in question 9 records that the undertaking covers future facilities rather than existing ones, and that Microsoft will continue to seek state sales tax exemptions on chips and equipment.

A list of host jurisdictions since January 2026, with the rate applied, any incentive received from any level of government, and whether the site is treated as existing or future, would settle the pledge, which is still the most auditable commitment in the set. Per-site headcount split by construction, operations, and local hires would do the same for the jobs pledge, and disbursement by data center community would do it for the training and nonprofit investment.

If answered: The first two rows of Table 4 move off “Plausible, unaudited.” and “Kept once, and narrower than it reads.” The tax pledge would then be settled rather than sampled, which is what Microsoft’s own January language invites.

None of these is a gotcha, and none asks Microsoft to concede a point. Each converts a company-wide claim into a local fact that a resident, a utility, or an enterprise buyer can check, which is the direction the January commitments already point. Greyhound Research would rather test the answers than assume them, and the offer to publish them as given stands.

↑ Back to the section guide

A Global Promise Kept, A Local One Still Owed: The Final Greyhound Standpoint

Greyhound Research’s final judgment is that Microsoft’s community commitments are now specific enough to be tested, and that its own 2026 disclosures allow a first test. On that test, replenishment has reached seven of the 29 data center locations Microsoft lists, and the location table covers about a third of the water Microsoft reports withdrawing. The milestone the company leads with sits outside its accountant’s review, and the reviewed FY25 withdrawal figure is 14.5 percent above the year before.

The same table earns Microsoft specific credit, and Greyhound Research gives it. It attaches dated FY30 promises to eleven named locations, it shows replenishment above local withdrawal at Phoenix, San Antonio, Madrid, and Querétaro, and its project portfolio names partners, basins, and durations in more detail than the corporate total suggests.

Microsoft’s March 2026 decision to end non-disclosure agreements with local governments and share its datacenter plans as openly as possible is the right place to begin. At Middenmeer and in Western Sydney, Microsoft has already shown that it can adapt a site to its setting. The discipline behind the 2015 article still applies: measure carefully, and design for the place.

Microsoft does not need to be persuaded of the principle. Its own foreword states that replenishing global volumes is not enough and that the next phase of the work is increasingly local. The question this report leaves with the company is therefore not whether the local test is the right one, which Microsoft has already answered, but when the published record will be able to carry it.

The account stops where most of the water is. Two disclosures would strengthen Microsoft’s position: the leased estate in the location table, or a stated reason it cannot be there, and an FY26 table that shows delivered volumes at the locations carrying FY30 promises, with the quantification basis stated. Evidence of unsupported demand or an unmet material obligation at a named site would make this judgment more critical. The evidence behind it is Microsoft’s own published disclosure, verified against the documents and current through October 20, 2026. It is not measured data from any facility.

For enterprise buyers, the consequence runs through the specific dependency, and Table 8 says what to do with each kind of evidence. A community’s right to expect promised benefits stands even where a shortfall creates no cloud outage.

Greyhound Research believes providers that demonstrate delivery should be better placed to earn confidence in their next development. A year after opening, and again five years later, residents should be able to return to the commitments recorded during consultation and see what happened. That is how a conversation about welcoming a data center becomes a lasting account of what it gave back. The action this report asks of Microsoft is small and specific: publish the cooling type and the leased estate in the next edition of the location table, and answer the ten questions beside it. For an enterprise buyer, the action is to establish which regions and availability zones run on capacity outside that table before the next renewal, because every other answer depends on it.

↑ Back to the section guide

Sources And References: Every Document Behind Every Claim

Public sources reviewed through October 20, 2026 and grouped by what each class of evidence can establish. Every source named in a sentence is hyperlinked there as well as listed here. Where this research report states that something is not published, the statement is bounded by the documents listed below, read on the dates shown, and does not extend to material Microsoft may hold, may have filed with a regulator, or may publish elsewhere. Sources that refused automated retrieval are identified as such and were not reached by any other route.

A. Microsoft’s own environmental disclosures

These establish what Microsoft reports about its own operations. They are company statements, and the table-level figures carry the assurance boundary described in this research report.

Microsoft (2026a) 2026 Environmental Data Fact Sheet. Undated. Cover title 2026 Environmental Sustainability Report, Data Fact Sheet, reporting on FY25. The independent accountant’s review report inside it is dated July 9, 2026. Company-wide water, Table 8. The location table, Table 15, and its definitions. The accountant’s limited-assurance report and its scope. The inventory boundary and its application to water, Sections 1.7 and 1.8. Available at: https://cdn-dynmedia-1.microsoft.com/is/content/microsoftcorp/microsoft/msc/documents/presentations/CSR/2026-Microsoft-Environmental-Data-Fact-Sheet-PDF.pdf

Microsoft (2026b) 2026 Environmental Sustainability Report. Undated. Released with the foreword of July 9, 2026, reporting on FY25. Water reporting, the reporting-disclosure note, and the organizational boundary. Available at: https://cdn-dynmedia-1.microsoft.com/is/content/microsoftcorp/microsoft/msc/documents/presentations/CSR/2026-Microsoft-Environmental-Sustainability-Report-PDF.pdf

Microsoft (2026c) Understanding water use at Microsoft datacenters. No date shown on the page. Page metadata records publication on March 30, 2026 and modification on October 15, 2026. Local explainer and its water-intensity figure. Available at: https://local.microsoft.com/blog/understanding-water-use-at-microsoft-datacenters/

Microsoft (n.d.) Microsoft water replenishment project portfolio. Undated. As of end of FY25. Accessed September 9, 2026. Project tables and the methodology notes on estimated and quantified volumes. Available at: https://cdn-dynmedia-1.microsoft.com/is/content/microsoftcorp/microsoft/msc/documents/presentations/CSR/Water-Portfolio-FY25-Water-Replenishment.pdf

B. Microsoft’s commitments and statements

These establish what Microsoft has undertaken to do. A commitment is evidence of intent, not of delivery.

Microsoft (2022) Blending datacenters into Wieringermeer nature with biomimicry. No date shown on the page. Page metadata records publication on April 7, 2022 and modification on June 16, 2025. Local Community Team, Microsoft Local. Native trees and shrubs planted around the Middenmeer campus from 2022, the start of a long-term effort to blend it into the Wieringermeer polder. Available at: https://local.microsoft.com/blog/datacenters-inpassen-in-de-wieringermeer/

Microsoft (2023) Microsoft collaborates with Indigital and traditional owners on artwork and landscape design for its new Western Sydney data centre. June 30, 2023. Microsoft Source. Artwork and landscape design shaped with the Dharug people, the Traditional Custodians of the land, through the Indigenous-owned organization Indigital. Available at: https://news.microsoft.com/source/asia/features/microsoft-collaborates-with-indigital-and-traditional-owners-on-artwork-and-landscape-design-for-its-new-western-sydney-data-centre/

Microsoft (2024) Sustainable by design: Next-generation datacenters consume zero water for cooling. December 9, 2024. Steve Solomon, Vice President, Datacenter Infrastructure Engineering. The cooling design and its energy tradeoff. Available at: https://www.microsoft.com/en-us/microsoft-cloud/blog/2024/12/09/sustainable-by-design-next-generation-datacenters-consume-zero-water-for-cooling/

Microsoft (2026d) Building Community-First AI Infrastructure. January 13, 2026. Brad Smith, Vice Chair and President. United States commitments and named examples. Available at: https://blogs.microsoft.com/on-the-issues/2026/01/13/community-first-ai-infrastructure/

Microsoft (2026e) 6 projects that helped Microsoft meet its renewable energy goal. February 18, 2026. Sally Beatty, Microsoft Source. Power purchase agreements described as “typically 10- to 15-year commitments”. Available at: https://news.microsoft.com/source/features/sustainability/6-projects-that-helped-microsoft-meet-its-renewable-energy-goal/

Microsoft (2026f) Putting communities first: Our decision to end NDAs with local governments. No date shown on the page. Page metadata records publication on March 18, 2026 and modification on March 23, 2026. Local Community Team, Microsoft Local. The end of non-disclosure agreements with local governments, and the commitment to share datacenter plans as openly as possible. Available at: https://local.microsoft.com/blog/putting-communities-first-our-decision-to-end-ndas-with-local-governments/

Microsoft (2026g) Building AI infrastructure the Community-First way in Canada. April 7, 2026. Matt Milton, President, Microsoft Canada. The Canadian application of the approach, including support for Indigenous AI Fluency and Workforce Readiness Hubs. Available at: https://blogs.microsoft.com/on-the-issues/2026/04/07/building-ai-infrastructure-the-community-first-way-in-canada/

Microsoft (2026h) Powering the next wave of AI: Expanding capacity with our new datacenter in Pecos. June 22, 2026. Noelle Walsh, President, Cloud Operations and Innovation. The co-located natural gas plant serving the Pecos campus off the public grid, paid for by Microsoft. Available at: https://blogs.microsoft.com/blog/2026/06/22/powering-the-next-wave-of-ai-expanding-capacity-with-our-new-datacenter-in-pecos/

Microsoft (2026i) Responsibly building the AI future. July 9, 2026. Brad Smith, Vice Chair and President, and Melanie Nakagawa, Chief Sustainability Officer. Foreword to the 2026 report: the local next phase, the replenishment milestone, and the certificate pause. Available at: https://blogs.microsoft.com/on-the-issues/2026/07/09/responsibly-building-the-ai-future/

C. Another operator’s disclosures, for comparison

These establish what a comparable operator publishes. They show what disclosure at a given granularity looks like, not a ranking of one operator against another.

Amazon (2024) 4 ways AWS is innovating on water sustainability in Singapore. March 26, 2024. About Amazon Singapore. The Hydroleap water recycling pilot and deployment, as stated by AWS. No facility is named. Available at: https://www.aboutamazon.sg/news/aws/4-ways-aws-is-innovating-on-water-sustainability-in-singapore

Amazon (2026) Water Positive Methodology. March 2026. Facility coverage, the accounting formula, and basin objectives. Available at: https://sustainability.aboutamazon.com/aws-water-positive-methodology.pdf

Amazon (n.d.) AWS Cloud. No date shown. Accessed September 9, 2026 and again October 19, 2026. The water usage effectiveness definition and the 2025 regional and geography tables. Available at: https://sustainability.aboutamazon.com/products-services/aws-cloud

D. Regulators, legislatures and official bodies

These establish what public authorities require, have found, or have resolved to do. They are the strongest class in this research report, because they do not depend on the subject’s own account.

Commission for Regulation of Utilities (2025) The CRU Publishes its Decision on New Electricity Connection Policy for Data Centres. December 12, 2025. Location-specific network conditions for data center connections in Ireland. Available at: https://www.cru.ie/about-us/news/the-cru-publishes-its-decision-on-new-electricity-connection-policy-for-data-centres/

Commonwealth of Massachusetts (2026) No. 658: Establishing Requirements for Responsible Data Center Development and Operations in Massachusetts to Protect and Support Ratepayers, Communities, and the Environment. September 8, 2026. Permits conditioned on a community benefits agreement. Non-disclosure agreements between state agencies and data center projects prohibited except as provided by law. Available at: https://www.mass.gov/executive-orders/no-658-establishing-requirements-for-responsible-data-center-development-and-operations-in-massachusetts-to-protect-and-support-ratepayers-communities-and-the-environment

ESB Networks (2026) LEU Pre-Engagement and Application Process. Effective May 20, 2026. No issue date shown. Section 6, planning-zone and substation constraints. Available at: https://media.esbnetworks.ie/media/docs/default-source/publications/leu-pre-engagement-and-application-process.pdf

European Commission (2024) Commission Delegated Regulation (EU) 2024/1364 of 14 March 2024 on the first phase of the establishment of a common Union rating scheme for data centres. March 14, 2024. Published in the Official Journal May 17, 2024. Articles 1, 3, and 5 and Annexes II and IV on reporting and confidentiality. Available at: https://eur-lex.europa.eu/legal-content/EN/TXT/PDF/?uri=OJ:L_202401364

European Commission (2026a) Report from the Commission to the European Parliament and the Council on the energy efficiency of data centres in the EU. September 21, 2026. COM(2026) 500 final. First reporting period. Union-wide average water usage effectiveness of 0.58 cubic meters per megawatt-hour across 458 data centers. Available at: https://energy.ec.europa.eu/document/download/8a87a5fe-0260-462a-b165-09f1496d1aa0_en?filename=COM_2026_500_1_EN_ACT_part1_v4.pdf

European Commission (2026b) Commission Delegated Regulation (EU) …/… of 21.9.2026 supplementing Directive (EU) 2023/1791 of the European Parliament and of the Council and amending Commission Delegated Regulation (EU) 2024/1364 as regards the establishment of a common Union rating scheme for data centres. September 21, 2026. C(2026) 3472 final. Adopted, its number still unassigned and not yet published in the Official Journal as at October 19, 2026. The A to G label, water usage effectiveness redefined against freshwater input, first labels due August 15, 2027. Available at: https://energy.ec.europa.eu/document/download/b073ba9b-82a8-45cb-b8ba-78dad149e941_en?filename=C_2026_3472_1_EN_ACT_part1_v9.pdf

European Commission (2026c) Minimum performance standards for data centres in Europe – public consultation launched. September 21, 2026. Directorate-General for Energy. The consultation closes December 14, 2026. Available at: https://energy.ec.europa.eu/news/minimum-performance-standards-data-centres-europe-public-consultation-launched-2026-09-21_en

European Parliament and Council (2023) Directive (EU) 2023/1791 of the European Parliament and of the Council of 13 September 2023 on energy efficiency and amending Regulation (EU) 2023/955 (recast). September 13, 2023. Published in the Official Journal September 20, 2023. Article 12 and Annex VII on published facility information, as restated in recital 3 of the 2024 delegated regulation. Available at: https://eur-lex.europa.eu/legal-content/EN/TXT/HTML/?uri=CELEX:32023L1791

Federal Office for Economic Affairs and Export Control (2026) Rechenzentrumsregister: Öffentliches RZReg ab sofort online. August 14, 2026. Bundesamt für Wirtschaft und Ausfuhrkontrolle (BAFA), Germany. The public register launch and its publication routes. Available at: https://www.bafa.de/SharedDocs/Kurzmeldungen/DE/Energie/260814_rzreg_livegang_portal_d.html

Infocomm Media Development Authority and Enterprise Singapore (2026) Singapore launches world’s first liquid cooling standard for data centres in tropical climates. August 27, 2026. Release and factsheet for SS 726:2026. This research report uses the standard’s scope, not the release’s ranking. Available at: https://www.enterprisesg.gov.sg/-/media/esg/files/media-centre/media-releases/2026/august/mr03826_singapore-launches-worlds-first-liquid-cooling-standard-for-data-centres-in-tropical-climates.pdf

Warren, E., Van Hollen, C. and Blumenthal, R. (2026) Power and Profits: How the AI Data Center Boom Costs Households and Communities. October 2026. No day printed. Report by three United States senators. The scope of Microsoft’s property tax undertaking, and its position on non-disclosure agreements at local and state level, drawn from Microsoft’s written responses. Available at: https://www.warren.senate.gov/wp-content/uploads/2026/10/471703-ai-data-center-boom-cost-Digital-8.5×11.pdf

E. Scientific, technical and methodological references

These establish definitions and methods. They do not establish any outcome at any facility.

Carter, V. (n.d.) Technical Aspects of Wetlands: Wetland Hydrology, Water Quality, and Associated Functions. No date shown. US Geological Survey, National Water Summary on Wetland Resources, Water-Supply Paper 2425. Groundwater recharge and discharge depend on local setting. Available at: https://water.usgs.gov/nwsum/WSP2425/hydrology.html

Collins, N. et al. (2025) Volumetric Water Benefit Accounting 2.0: Guidance for implementing, evaluating, and claiming volumetric water benefits of water stewardship projects. September 17, 2025. World Resources Institute and partners. Voluntary guidance on quantification, timing, and attribution. Available at: https://www.wri.org/research/volumetric-water-benefit-accounting-2-0

Smith, S.J. et al. (2026) United States Data Center Energy Usage Report: 2025 Update. June 2026. Lawrence Berkeley National Laboratory. Indirect water use in electricity generation. National electricity forecasts are not used. Available at: https://datacenters.lbl.gov/publications/united-states-data-center-energy-2025

US Geological Survey (2019) Water-Use Terminology. February 27, 2019. Water Resources Mission Area. Withdrawal, delivery, consumptive-use, and return-flow definitions. Available at: https://www.usgs.gov/mission-areas/water-resources/science/water-use-terminology

F. Independent reporting and legal analysis

These establish what third parties have reported or analyzed. Where a report rests on another publication, both are named, and nothing in this class is treated as independently verified by Greyhound Research.

Baptista, D. and McDonnell, F. (2026) Resistance blooms in Mexico’s data centre valley. February 4, 2026. Context, Thomson Reuters Foundation. The state of a community program around Querétaro. Not independently verified by Greyhound Research. Available at: https://www.context.news/ai/long-read/resistance-blooms-in-mexicos-data-centre-valley

Day Pitney LLP (2026) New Jersey Enacts Energy and Water Reporting Requirements for Data Centers. September 14, 2026. Katharine A. Coffey, Luke S. Pontier and Michael P. Castore. P.L. 2026, c. 75, signed August 27, 2026: semiannual energy and water reporting with no minimum size threshold. Available at: https://daypitney.com/new-jersey-enacts-energy-and-water-reporting-requirements-for-data-centers

Foley & Lardner LLP (2026) Texas Governor Abbott Directs TCEQ to Freeze All Data Center Permits in a Whole-of-Government Action. September 22, 2026. The directive of September 21, 2026 pausing pending and new environmental permits for data center projects until grid and water audits are published. Available at: https://www.foley.com/insights/publications/2026/09/texas-governor-abbott-directs-tceq-to-freeze-all-data-center-permits-in-a-whole-of-government-action/

Holland & Knight LLP (2026) California’s 2026 Data Center Laws: What You Need to Know. October 5, 2026. Seven statutes signed September 21, 2026, including reporting of actual annual water use and cooling-system type at license renewal. The legislature’s own bill pages refused retrieval on October 19, 2026, so the detail rests on this analysis. Available at: https://www.hklaw.com/en/insights/publications/2026/10/californias-2026-data-center-laws-what-you-need-to-know

Lai, S. (2026) EXCLUSIVE INTERVIEW/Microsoft working on making Taiwan Azure region widely available: Executive. September 1, 2026. Focus Taiwan, CNA. A Microsoft infrastructure executive’s description of a data center as a commitment to a market lasting 20 or 30 years, as reported. Available at: https://focustaiwan.tw/business/202609010030

Plumb, T. (2026) Amazon claims its data centers are 7x more water-efficient than the industry average. June 11, 2026. Network World. Cited only for Greyhound Research’s own commentary. Available at: https://www.networkworld.com/article/4184250/amazon-claims-its-data-centers-are-7x-more-water-efficient-than-the-industry-average.html

Smith, R. (2026) Microsoft Scraps UK Data Center Deal With Nscale, Telegraph Says. October 18, 2026. Bloomberg, reporting The Telegraph, read through Yahoo Finance’s republication because the publisher’s own page refused automated retrieval. Microsoft did not comment. Available at: https://finance.yahoo.com/technology/ai/articles/microsoft-scraps-uk-data-center-083110363.html

Village of Mount Pleasant (2026) Microsoft Foregoes Data Center Incentive. September 28, 2026. Village press release, carried by Urban Milwaukee, which states it did not write or verify the release. Available at: https://urbanmilwaukee.com/pressrelease/microsoft-foregoes-data-center-incentive/

G. Greyhound Research’s own published positions

These establish the firm’s prior published view, which this research report extends.

Gogia, S.V. (2026) Data centres: The watts and the water. August 13, 2026. Column in The Hindu BusinessLine. The quoted passage is the published text. The page refused automated retrieval on October 19, 2026 and is cited from the earlier reading. Available at: https://www.thehindubusinessline.com/opinion/data-centres-the-watts-and-the-water/article71335930.ece

Greyhound Research (2015) Building Energy Efficient Data Centers. April 21, 2015. Historical analysis by Sanchit Vir Gogia. Available at: https://greyhoundresearch.com/building-energy-efficient-datacenters-press-media-cxotodayalerts/

Greyhound Research (2025) Data Center 2025: Speculation, Pricing Shocks, And The Rise of Tier-2 Markets. June 28, 2025. Announced demand and actual development. Available at: https://greyhoundresearch.com/data-center-2025-speculation-pricing-shocks-and-the-rise-of-tier-2-markets/

↑ Back to the section guide

Disclosures: Methodology, Independence, Financial, Authoring, And Copyright

The evidence in this research report is current through October 20, 2026. Any written response received from an organization named in it after that date is dated where it appears.

Research Methodology

Greyhound Research’s analysis rests on source evidence verified against the document that supports each claim. For this research report the method is documentary. Each commitment was tested against what the company itself has published, what that evidence cannot establish was recorded, and every statement that something is not published is bounded by the documents named in Sources And References and the date they were read.

For this research report, the source evidence comprises Microsoft’s 2026 Environmental Data Fact Sheet and 2026 Environmental Sustainability Report, and its water replenishment project portfolio as of the end of FY25. It also comprises Microsoft’s announcements and posts of December 9, 2024, and January 13, February 18, March 18, April 7, June 22, and July 9, 2026. Microsoft’s local water explainer, whose page metadata records modification on October 15, 2026, is part of it, as are its published accounts of site design in the Netherlands and Western Sydney. Published disclosures of Amazon Web Services, and the regulatory, scientific, and independent sources listed in full in Sources And References, complete the record. All were read on or before October 20, 2026.

Independence Disclosure

Research under the Greyhound Research brand is independent, and no organization holds a right of approval over its content or publication. This edition was not submitted to Microsoft for factual review before publication. Editorial control remained entirely with Greyhound Research throughout.

Greyhound Consilium is the B2B thought-leadership arm of The House Of Greyhound and a sister brand to Greyhound Research. It builds on Greyhound Research’s independent research through collaborations that produce thought-leadership content, research, and practical toolkits for decision-makers. Every Greyhound Consilium publication names its partner on the face of the document. This research report is published under the Greyhound Research brand.

Greyhound Research undertakes advisory and consulting engagements with enterprise buyers and technology providers. No such engagement confers control over research findings or the decision to publish.

Financial Disclosure

No organization commissioned or paid for this research report. Neither Microsoft nor Amazon, the two companies whose disclosures it assesses, is a client of Greyhound Research, and neither pays for any work by Greyhound Research or Greyhound Consilium. Greyhound Research and the author hold no shares or other financial instruments in either company.

Not Investment Advice

This research report is technology research for enterprise buyers and the communities that host data centers. It is not investment advice, and nothing in it is a view on any company’s securities.

Authoring Disclosure

This material reflects the analysis, viewpoint and editorial judgment of Greyhound Research. The position presented is developed from Greyhound Research’s research, advisory work and accumulated institutional knowledge. Depending on the assignment, AI may be used to retrieve relevant material; structure, draft or polish prose from analyst-provided concepts, dictation, notes, rough drafts and source material; support secondary research and factual cross-checking; and assist with editing, condensation and consistency with Greyhound Research’s house style. AI does not independently establish Greyhound Research’s viewpoint, analytical position or conclusions. Material factual claims are verified against source evidence before release. Greyhound Research reviews and approves the final material and retains responsibility for it.

Copyright Policy

All content in this document and on the Greyhound Research website is protected by copyright law and may not be reproduced, distributed, transmitted, displayed, published, or broadcast without the prior written permission of Greyhound Research or, in the case of third-party materials, the prior written consent of the copyright owner of that content. You may not alter, delete, obscure, or conceal any trademark, copyright, or other notice appearing in any Greyhound Research content. Quotation of up to one hundred words is permitted where it carries attribution to Greyhound Research and a link to the published page. Reproduction, redistribution or republication beyond that extent, in whole or in part, by email or in any medium including websites, newsletters and intranets, requires prior written permission, which may be requested at the address below. The published version of each content piece on the Greyhound Research website carries sharing tools for readers who wish to share it.

For questions arising from these disclosures, including licensing requests and media inquiries, contact the Greyhound Research Community Relations Team at connect@thofgr.com.

↑ Back to the section guide