Microsoft’s latest environmental update lands with a familiar kind of tension: the company is still talking about ambitious climate outcomes, but its newest accounting shows emissions rising again—this time sharply enough to force a closer look at what “progress” means when the world’s most energy-hungry technology keeps accelerating.
In its 2026 Environmental Sustainability Report, Microsoft reports that its carbon emissions increased 25 percent in 2025, reaching 34 million metric tons. The figure is presented as the total “without select interventions,” a phrase that matters because it signals that the number is not simply a raw snapshot of operations. It’s an accounting outcome shaped by what Microsoft chooses to include—or exclude—when it evaluates progress against its own targets. In other words, the headline increase is real, but the context around how the company measures and manages emissions is equally important.
According to reporting summarized by GeekWire and referenced by The Verge, Microsoft attributes the rise primarily to the expansion of its datacenter infrastructure. That explanation will sound unsurprising to anyone tracking the AI boom: more compute requires more servers, more power delivery, more cooling, and often more construction. But what makes this update worth reading closely is how Microsoft pairs that operational reality with a second driver that is less intuitive—its decision to stop purchasing certain types of renewable energy certificates.
That combination—physical growth plus procurement changes—creates a story that goes beyond “emissions went up.” It’s about how climate strategies can collide with the pace of infrastructure buildout, and how the rules of the game for renewable energy claims are tightening.
To understand why, it helps to start with what Microsoft says it did. In February 2025, Microsoft decided to stop purchasing “non-additional, unbundled renewable energy certificates.” The wording is technical, but the implication is straightforward: Microsoft is moving away from renewable credits that it considers less credible because they don’t represent new renewable generation or aren’t bundled with the electricity being used. In climate accounting, not all renewable energy certificates are treated equally. Some are viewed as supporting additional clean power capacity; others are seen as more like a financial instrument that can be purchased without guaranteeing that new renewable electricity is actually added to the grid.
So when Microsoft stops buying those certificates, it may reduce the ability to offset emissions in the way it previously could. Even if the company is still buying renewable energy in some form, changing the quality of the instruments used to support claims can shift the reported emissions totals. That doesn’t necessarily mean Microsoft’s underlying energy use got worse overnight. It can mean the company’s measurement and offset strategy became stricter, and the emissions accounting reflects that stricter standard.
This is where the phrase “without select interventions” becomes more than a footnote. It suggests that Microsoft’s reported emissions are not just a single number pulled from meters. They’re the result of a framework that includes interventions—actions that can reduce or counterbalance emissions under specific conditions. When those interventions are excluded, the number can look more alarming even if the company is actively working on mitigation. The key point is that Microsoft is essentially telling readers: here is what emissions look like under a particular lens, and here is why that lens produces an increase.
Still, even with that nuance, the core message remains: Microsoft’s emissions rose in 2025, and the company itself points to datacenter expansion as the primary reason.
Why datacenter expansion is such a big deal isn’t just because of electricity consumption. It’s because datacenters are complex systems with multiple layers of energy demand. There’s the obvious part—powering servers—but there’s also the less visible part: cooling systems, networking equipment, storage, and the infrastructure required to deliver electricity reliably at scale. As companies build out more capacity, they often need new facilities or expansions that take time to come online, and those facilities may initially run at different efficiency levels than older ones. Even when new sites are designed to be efficient, the sheer scale of growth can overwhelm improvements in efficiency.
And then there’s the AI-specific factor. AI workloads tend to be both compute-intensive and bursty. Training runs can require large amounts of power for sustained periods, while inference can create continuous demand depending on usage patterns. That means the energy profile of AI-driven services can differ from traditional cloud workloads. It’s not only that more compute is needed; it’s that the pattern of demand can change how quickly new capacity must be built and how much power must be available at any given time.
Microsoft’s report, as summarized in the coverage, frames the emissions increase as driven primarily by datacenter infrastructure expansion. That aligns with what many observers have been saying: the AI economy is not just a software story—it’s a physical infrastructure story. And physical infrastructure has a carbon footprint that can lag behind policy intentions.
But the renewable energy certificate decision adds another layer: it highlights how climate progress can be constrained by the credibility of offsets and the evolving standards for what counts as “real” decarbonization.
Renewable energy certificates have long been a tool for companies trying to claim that their electricity use is matched with renewable generation. However, the market has been criticized for allowing companies to purchase certificates that don’t necessarily correspond to additional renewable capacity. If a certificate is “non-additional,” it may represent renewable energy that would have been generated anyway. If it’s “unbundled,” it may be separated from the actual electricity consumption, meaning the company isn’t necessarily using renewable electricity at the time it’s consuming power—at least not in a direct, traceable way.
Microsoft’s February decision to stop purchasing these types of certificates suggests it wants its climate claims to align more closely with additionality and tighter linkage between consumption and clean generation. That’s a meaningful shift because it implies Microsoft is willing to accept less favorable accounting outcomes in exchange for higher integrity in its climate strategy.
This is one of the more interesting angles in the story: sometimes emissions rise not because a company is doing less, but because it is doing more of the right things—and measuring them differently. A stricter approach can make reported emissions look worse even if the company’s actual trajectory is improving. That doesn’t erase the need to reduce emissions, but it changes how you interpret the numbers.
There’s also a strategic communication element. Companies often face pressure to show progress toward targets. But if the target is carbon negative by 2030, the path to that outcome depends on both reducing emissions and removing carbon. Microsoft has talked for years about being carbon negative by 2030, which implies that at some point it expects to remove more carbon than it emits. That requires not only operational reductions but also carbon removal at scale—something that remains difficult, expensive, and dependent on technologies and markets that are still maturing.
So when Microsoft reports an emissions increase, it’s not just a setback; it’s a signal that the company’s decarbonization plan is operating under stress. The stress comes from two directions: the rapid growth of energy demand from datacenters and the tightening of what counts as legitimate renewable matching.
The result is a kind of double bind. If Microsoft expands datacenters to meet AI demand, emissions can rise. If Microsoft tries to offset those emissions using renewable certificates, it may find that the most credible options are harder to source or more expensive—or that the accounting rules no longer allow certain certificates to count. Meanwhile, the company still needs to keep moving toward carbon negativity by 2030, which means it can’t simply pause growth or rely indefinitely on offsets.
This is why the “without select interventions” framing is so important. It suggests Microsoft is actively using interventions to manage emissions, but the reported increase indicates those interventions weren’t enough to fully counterbalance the emissions impact of growth in 2025. Or, depending on how the interventions are defined, it may indicate that the interventions were not applied in the same way as in prior years.
Either way, the story points to a broader industry challenge. Many tech companies are building AI capacity faster than the grid can decarbonize, and faster than the supply chain for clean energy and carbon removal can scale. Even if a company buys renewable energy, the timing and additionality questions remain. Even if a company improves efficiency, the growth rate can outpace gains. And even if a company invests in carbon removal, the removals may not be immediate enough to offset near-term emissions increases.
Microsoft’s report, as described in the coverage, effectively lays out that reality in corporate language: emissions rose, driven by datacenter expansion, and influenced by changes in renewable energy certificate purchasing.
What makes this update feel particularly relevant now is that it arrives at a moment when AI is moving from novelty to infrastructure. The early phase of AI adoption was often framed as a software revolution. But the last couple of years have made it clear that AI is also a construction project: new campuses, new power substations, new transmission agreements, new cooling systems, and new operational practices. Each of those steps has a carbon footprint, and each step takes time to implement and optimize.
In that context, Microsoft’s emissions increase can be read as a symptom of a larger transition. The world is building the hardware for AI faster than it is building the clean energy and carbon removal capacity to match it. That mismatch doesn’t mean decarbonization is impossible. It means the timeline is harder than many people assumed.
There’s also a subtle but important point about how corporate climate reporting works. Emissions totals are not just a measure of what happened; they’re also a reflection of what a company chooses to count. When Microsoft changes its approach to renewable certificates, it can change the emissions accounting even if the physical reality hasn’t changed as dramatically. That’s why readers should treat the number as both a measurement and a statement of policy.
Microsoft’s decision to stop purchasing “non-additional, unbundled renewable energy certificates” suggests it is trying to improve the integrity of its climate claims. That can be seen as a positive development, even if it makes the emissions number look worse. It’s similar to how some organizations move from broad offsetting to more targeted