Microsoft Promised Carbon-Free AI by 2030. Then It Built 4.75 Gigawatts of Gas Plants.
Microsoft is considering abandoning its 100/100/0 hourly clean energy matching target, the most rigorous corporate decarbonization pledge ever made, because AI data centers are growing faster than clean power can follow. Meanwhile, three gas-powered data center projects announced in a single month would add 15.52 million metric tons of annual CO2 emissions, increasing the company's data center carbon footprint by 160 percent. An original analysis of the gap between annual and hourly energy matching reveals why corporate clean energy claims deserve far more scrutiny than they receive.
Four point seven five gigawatts. That is how much new methane gas power capacity Microsoft committed to build in a single month this spring, across three projects in Texas and West Virginia, each designed to run behind the meter at dedicated AI data center campuses where the electricity never touches the public grid.
Here is the number that should accompany it everywhere: 15.52 million metric tons of CO2 equivalent per year. According to Stand.earth Research Group analysis of the three facilities, that figure represents a 160 percent increase over Microsoft's existing data center carbon footprint, enough annual emissions to match those of roughly 3.4 million passenger cars, all to power servers running language models and image generators.
In February 2026, Microsoft celebrated a milestone: it had matched 100 percent of its 2025 electricity use with renewable energy. Three months later, Bloomberg reported the company was considering abandoning or delaying its far more ambitious 2030 target, the so-called 100/100/0 pledge to match every kilowatt consumed with carbon-free electricity on an hourly basis, on the same regional grid where the power is used.
These two facts are not contradictory. They reveal something more important about how corporate clean energy accounting actually works, and why the distinction between annual matching and hourly matching is not a technicality but the difference between climate accounting fiction and physics.
Annual Matching Is an Accounting Trick. Hourly Matching Is Physics.
When a company says it is "powered by 100 percent renewable energy," here is what typically happens. It tallies total electricity consumption for the year, then purchases an equal quantity of renewable energy certificates, tradeable vouchers that prove a corresponding amount of clean electricity was generated somewhere, at some point, by someone. A data center in Virginia running at 2 AM on a windless January night draws power from a grid that is 60 percent fossil fuel at that hour, but the company can buy solar RECs from a farm in New Mexico that overproduced at noon in July and call the whole arrangement carbon-free.
Annual matching directs capital toward renewable projects. Nobody disputes that. But it does not decarbonize the grid in the hours and places where the data center actually consumes electricity, because electrons do not carry passports and a certificate from a different time zone does not change the fuel mix feeding a specific server rack at 3 AM.
Hourly matching forces a harder question. You need clean power available during every hour of operation, on the same grid, which means buying or building not just solar and wind but also batteries, geothermal, nuclear, or hydro to fill the gaps when the sun sets and the wind drops.
Wilson Ricks, a manager in Clean Air Task Force's electricity program, told Heatmap News the goal was always going to be difficult: "There's going to be a big difference between them saying, 'We're going to keep trying as hard as we can to go as far as we can, but acknowledge we may not hit it,' versus saying, 'Well, we can't hit this extremely ambitious goal we set for ourselves, therefore we're just giving up on the overall mission.'"
The Clean Energy Honesty Gap: A Calculation
Nobody has published the math on what annual matching actually obscures compared to true hourly matching for a hyperscale data center operator. Here it is.
Start with a stylized 1 GW data center running 24 hours a day, 365 days a year, consuming 8,760 GWh annually. Under annual matching, the operator buys 8,760 GWh of renewable certificates. Solar farms produce at roughly 25 percent capacity factor; wind at roughly 35 percent. Blended, call it 30 percent. To generate 8,760 GWh per year at 30 percent capacity factor, you need about 3.3 GW of nameplate renewable capacity under contract. Microsoft has locked in over 40 GW of renewable PPAs across 26 countries, easily enough to cover annual matching for its current fleet.
But the data center does not sleep. At night, roughly 12 hours per day in most US regions, solar output is zero and wind output is variable. During those hours, the grid draws on whatever is available: natural gas, coal, nuclear, hydro. In PJM (the grid serving much of the eastern US, where Microsoft operates major facilities), the marginal overnight fuel mix is 40 to 60 percent fossil. On a calm winter night, it can be higher.
So a data center that is "100 percent renewable" on an annual basis is powered by fossil fuels for roughly half the hours in a year, sometimes more. Multiply across Microsoft's expanding fleet, and the gap between the annual matching claim and physical reality is enormous.
Hourly matching closes that gap. It requires pairing renewables with storage, firm clean power, or both. Battery storage at 4-hour duration costs roughly $150 to $300 per MWh of throughput. Adding 500 MW of battery capacity per GW of data center to cover nighttime hours approximately doubles the effective energy procurement cost, from $15 to $25 per MWh for annual RECs up to $50 to $80 per MWh for true 24/7 carbon-free electricity. For a 10 GW fleet growing by 1 GW per quarter, the incremental cost of hourly matching over annual matching runs $3 billion to $5 billion per year, which begins to explain why Microsoft is reconsidering.
4.75 Gigawatts of Gas in 30 Days
Each of the three projects has a name, a location, and an emissions estimate.
| Project | Location | Capacity | Annual CO2e | Partner |
|---|---|---|---|---|
| Chevron/Engine No. 1 | Pecos, Texas | 2.5 GW | 8.31 Mt | Chevron, Engine No. 1 |
| Monarch Compute Campus | Mason County, W.V. | 1.35 GW | 4.31 Mt | Nscale |
| Crusoe Facility | Abilene, Texas | 900 MW | 2.90 Mt | Crusoe |
| Total | 4.75 GW | 15.52 Mt |
Source: Stand.earth Research Group, April 2026. Emissions calculated from air permit applications and industry benchmarks.
All three are behind-the-meter installations, meaning the gas turbines connect directly to the data center without passing through the public grid. This matters for accounting. Behind-the-meter generation occupies a gray zone in corporate emissions reporting: it is Scope 1 (direct combustion) rather than Scope 2 (purchased electricity), and it exists outside the framework of grid-based renewable matching entirely. You cannot buy a REC to offset a gas turbine you own and operate on your own property. It simply burns methane and produces CO2.
Behind-the-meter gas generation for data centers barely existed two years ago. At the end of 2024, on-site data center projects accounted for 5 percent of all new US methane gas capacity under development. By the end of 2025, that figure had reached 39 percent, with on-site data center demand for gas turbines in a single year exceeding all demand for new gas power plants in the entire preceding year. Data centers have become the fastest-growing category of fossil fuel demand in America, and almost nobody outside the energy industry has noticed.
Why Microsoft Cannot Have Both
Microsoft is adding approximately 1 gigawatt of data center capacity every three months. Capital spending on data centers now exceeds $40 billion per year. Brad Smith, the company's president, told the Associated Press in March that he remained "confident in our ability" to meet the 2030 carbon-negative target through nuclear, solar, and hydropower investments.
Consider those numbers against each other: 40 GW of renewable PPAs at 30 percent capacity factor yields roughly 12 GW of average clean power. Existing data center capacity plus 4.75 GW of new gas-powered facilities, running at 85 percent capacity factor, produces about 4 GW of average dirty power around the clock, during precisely the hours when renewables are weakest. And Microsoft's fleet is expanding by 4 GW of nameplate capacity per year, each gigawatt consuming about 8.76 TWh annually.
At this growth rate, the renewable procurement pipeline cannot keep pace with demand in real time, only in annual retrospect. That is why the company is retreating from hourly matching while maintaining its annual matching claim. Both statements can be true simultaneously, which is exactly the problem.
Google Got There First, Then Flinched
Google introduced 24/7 carbon-free energy as a corporate target in 2020, before Microsoft or anyone else. After six years of effort, the company has achieved roughly 64 percent hourly matching across its global fleet, with some data centers reaching above 90 percent in favorable locations like Finland and Oregon where hydro and geothermal are abundant. In regions like Singapore and parts of the eastern US, the number is far lower, because clean firm power simply does not exist at sufficient scale in those grids yet.
Only two other companies followed Google and Microsoft in committing to hourly matching: Iron Mountain and a small handful of members of a coalition organized by The Climate Group. Most corporations still rely on annual matching with unbundled RECs, the approach that a company operating in coal-heavy Indiana can satisfy by purchasing certificates from a decade-old Texas wind farm.
Microsoft also recently paused its carbon removal purchasing program, which had been the largest in the corporate world. Combined with the potential retreat from hourly matching, the pattern suggests that the company's climate ambitions are being systematically subordinated to its AI ambitions when the two conflict.
The Strongest Counterargument
Microsoft has done more to advance corporate clean energy procurement than any company on earth, and it is not particularly close. Forty gigawatts of renewable PPAs across 26 countries represents real capital flowing to real wind and solar projects that would not have been built otherwise. Annual matching, even with its accounting limitations, has driven more clean energy investment than hourly matching because it is achievable at scale today, while hourly matching requires storage and firm clean power technologies that have not reached commercial deployment in most regions. Abandoning an impossible target is not abandoning ambition; it is being honest about what physics and grid infrastructure allow in 2026.
Furthermore, the alternative to building gas-backed data centers in the United States is not building no data centers at all; it is building them in countries with dirtier grids and weaker environmental standards, a climate outcome that would be categorically worse. And Microsoft's internal carbon tax, if it remains in place, means the gas plants carry an internal cost penalty that incentivizes transition to clean alternatives as they become available. Criticizing a company for building gas as a bridge while investing $40 billion in renewables simultaneously is demanding perfection in an imperfect world.
All of this is true. None of it changes the fact that 15.52 million tons of CO2 will enter the atmosphere every year from these three facilities, regardless of how many solar certificates sit in Microsoft's accounting ledger.
Limitations of This Analysis
Stand.earth is an environmental advocacy organization, and its 160 percent emissions increase figure relies on projected, not actual, emissions from facilities that are not yet operational; real-world emissions will depend on capacity utilization rates, turbine efficiency, and operational decisions that cannot be predicted with certainty before construction is complete. Microsoft's actual hour-by-hour energy consumption profile is proprietary, which means the "honesty gap" calculation above uses industry-average capacity factors rather than Microsoft-specific data, and the true gap could be larger or smaller depending on where Microsoft's data centers are located relative to its renewable assets. Bloomberg's reporting on the internal debate cites unnamed sources, and Microsoft has neither confirmed nor denied that it is considering abandoning the 100/100/0 target. Corporate clean power contracting data from BloombergNEF captures announced deals, not completed projects, and some PPAs signed in prior years may have been amended or cancelled without public disclosure. Finally, the behind-the-meter emissions accounting question is evolving: the Science Based Targets initiative and GHG Protocol are both actively revising guidance on how to classify on-site fossil fuel generation, and future standards may close the reporting gap identified here.
What You Can Do
If you are a corporate sustainability officer benchmarking against Microsoft's "100 percent renewable" claim, ask whether that figure reflects annual or hourly matching and demand the methodology, because the difference between the two can be a factor of three in both cost and climate impact. If you are a ratepayer in PJM or ERCOT territory where data center construction is accelerating, attend your utility's integrated resource planning hearings and ask specifically what share of proposed gas capacity is dedicated to behind-the-meter data center load that will never serve residential customers but will affect your rates. If you are an investor evaluating big tech climate commitments, look past the annual matching headline and examine Scope 1 emissions trends, which capture on-site combustion that RECs cannot offset, and watch for the GHG Protocol revision expected in late 2026 that may reclassify behind-the-meter gas as a Scope 2 reporting obligation. If you are a policymaker, note that behind-the-meter data center gas demand surged from 5 to 39 percent of all new US gas capacity in a single year without any corresponding update to state permitting or environmental review requirements. That regulatory gap exists in every state where data centers are proliferating.
The Bottom Line
Microsoft set the hardest clean energy target in corporate history, and it may be about to walk away from it, not because the company lacks resources or ambition but because it has chosen to grow AI infrastructure at a pace that makes the target physically unreachable with today's grid and storage technology. That is an honest acknowledgment of real constraints. But it is not the same thing as leadership. Behind-the-meter gas plants that add 15.52 million tons of annual CO2 while the company claims 100 percent renewable energy through annual certificate matching is a contradiction that no amount of procurement volume can resolve. Corporate clean energy pledges were supposed to pull the grid toward decarbonization. When the biggest pledge-maker in history starts building gas plants faster than wind farms, the question is no longer whether hourly matching is achievable by 2030. It is whether annual matching was ever more than a number on a spreadsheet.