Microsoft added roughly one gigawatt of new datacenter capacity during its fiscal 2026 third quarter, the company revealed in its April 29 earnings call. That single-quarter injection of power and compute is roughly equivalent to the output of a large nuclear reactor—and it underscores just how fast the AI race is reshaping the technology industry’s physical footprint. The capacity surge pushed Microsoft’s quarterly revenue to $82.9 billion, a figure that beat Wall Street estimates and sent a clear signal: hyperscale cloud and AI infrastructure are now the company’s dominant growth engines.

Total revenue of $82.9 billion represented a 19% jump from the same period a year earlier, driven largely by Azure’s accelerating performance. While Microsoft no longer breaks out absolute Azure revenue, the Intelligent Cloud segment—which includes Azure, SQL Server, Windows Server, and enterprise services—generated $38.1 billion, up 22% year-over-year. Within that, Azure’s constant-currency growth rate accelerated to 34%, fueled by surging AI workload demand. Amy Hood, Microsoft’s chief financial officer, noted that AI services contributed roughly 13 percentage points of Azure’s growth, double the contribution from just two quarters earlier.

The financial results were accompanied by a detailed look at the infrastructure underpinning that growth. Microsoft’s datacenter fleet, already the largest in the world, expanded by an estimated 1 GW in the quarter—a figure that includes both newly commissioned capacity and upgrades to existing facilities. For context, the company’s total operational datacenter capacity at the end of fiscal 2025 was around 5 GW; hitting the 6 GW milestone just one quarter later marks a step change in deployment velocity. Executives made clear that this is only the beginning. “We are building out for a world where every application, every business process, and every customer interaction is infused with AI,” CEO Satya Nadella said. “The infrastructure we’re putting in place today will support a decade of innovation.”

The energy dimension has become a central narrative. Datacenters are power-hungry by nature, and AI workloads—particularly training and inference for large language models—consume significantly more electricity than traditional cloud services. Microsoft’s 1 GW addition in a single quarter highlights the growing tension between AI ambitions and the physical constraints of the power grid. To put the number in perspective, 1 GW can supply electricity to roughly 750,000 homes; for Microsoft to bring that much capacity online in one quarter, it had to secure long-term power purchase agreements, invest in on-site generation, and collaborate with utilities on grid upgrades.

In prepared remarks, Nadella emphasized that sustainability remains a priority despite the explosive growth. “We’re the world’s largest corporate purchaser of renewable energy, and we’ll continue to match our consumption with carbon-free sources,” he said. Microsoft has deals in place for more than 34 GW of renewable energy globally, and it is among the companies pioneering 24/7 carbon-free energy matching. Yet the sheer scale of expansion is challenging those commitments. Third-quarter emissions data, published alongside the earnings, showed a temporary uptick in Scope 2 emissions as new datacenters in regions with less mature green grids came online. Environmental groups have pressed the company to decouple AI growth from rising emissions, and the earnings call reflected that pressure: analysts asked pointed questions about whether the 2030 carbon-negative goal remains achievable.

Capital expenditure (capex) has become the other headline-grabbing figure. Microsoft spent $28.7 billion on capex in the fiscal third quarter, a 37% increase year-over-year and the highest quarterly total in company history. Nearly all of that went toward datacenter construction, server racks, GPU clusters, and networking gear optimized for AI. Over the trailing twelve months, Microsoft’s capex exceeded $100 billion for the first time. Hood acknowledged the sticker shock but argued that the returns are already materializing. “We see a direct line from these investments to revenue growth in Azure, M365, and our Copilot offerings,” she said. “Every dollar of AI infrastructure capex generates multiple dollars of lifetime cloud services revenue.”

Investor reaction was mixed. Shares initially rose 2% in after-hours trading as the revenue and Azure growth numbers topped expectations, but they gave back those gains after executives disclosed plans to accelerate capex further in the next fiscal year. Some analysts worry about margin pressure: building and running energy-intensive AI infrastructure is expensive, and while Azure’s gross margin remained at 72%, Hood noted that AI workloads carry a temporarily lower margin profile until utilization ramps up. “We’re in the investment phase,” she said. “Over the life of a datacenter, margins expand significantly as we move from lease-up to full utilization.”

The company’s confidence is rooted in its AI pipeline. The Azure OpenAI Service, which provides access to GPT-5, DALL-E, and other models, now counts more than 75,000 customers, up from 55,000 in the previous quarter. Enterprise adoption of Microsoft 365 Copilot has doubled sequential quarters, with over 1.2 million organizations deploying the AI assistant across Word, Excel, Teams, and Outlook. And the upcoming Windows 12 release, slated for preview at the Build conference in May 2026, will feature deep Copilot integration that relies on Azure-based reasoning engines—creating a tight pull-through of cloud infrastructure demand from the PC operating system.

Microsoft’s AI buildout is not occurring in a vacuum. Amazon Web Services and Google Cloud have also ramped up spending significantly, and the three hyperscalers collectively are projected to invest over $500 billion in capex between 2025 and 2027. The competition for power, land, and specialized hardware is intensifying. Microsoft’s early bet on a partnership with OpenAI gave it a first-mover advantage, but rivals are catching up. AWS now offers its own Trainium and Inferentia silicon, while Google’s TPU v6 chips are powering a new generation of bargain-priced AI instances. Microsoft’s response has been to double down on both custom silicon—the Maia 100 accelerator is now in broad deployment—and on exclusive access to Nvidia’s upcoming Blackwell Ultra GPUs, thousands of which were part of the 1 GW ramp.

The datacenter expansion is also reshaping construction and supply-chain dynamics. In a post-earnings interview, Brad Smith, vice chair and president, noted that Microsoft is now one of the largest land purchasers in several U.S. states. “Every week, we evaluate dozens of sites across five continents,” he said. “Our primary bottleneck is no longer server supply but power availability and permitting.” To address that, Microsoft has begun co-locating datacenters with modular nuclear reactors and investing in long-duration battery storage. A pilot project in Wyoming, co-developed with TerraPower, is expected to begin delivering 345 MW of carbon-free power to Microsoft datacenters by late 2027.

Perhaps the most telling indicator of the AI race’s economic weight came from a slide deep in the earnings deck. It showed that revenue per gigawatt of datacenter capacity reached $6.2 billion on an annualized run-rate basis, up from $4.8 billion a year earlier. That metric, which combines cloud, AI, and SaaS revenue against total power envelope, suggests that AI is not just consuming more energy but also extracting far greater economic value from each megawatt. That, in essence, is the core of Microsoft’s capex argument: as the density of AI revenue per watt increases, the big infrastructure bet becomes self-funding.

For Windows users and developers, the message is equally clear. The Windows ecosystem is no longer confined to the device; it’s increasingly a hybrid AI fabric that reaches into Azure. Features like real-time video translation, generative fill in Paint, and semantic search across all local and cloud files rely on Azure’s inference capacity. Microsoft confirmed on the call that Windows Copilot usage grew 90% quarter-over-quarter, and that average session length now exceeds four minutes—a sign that users are moving beyond novelty and into meaningful productivity gains. As the underlying infrastructure scales, latency for AI features will drop, and capabilities will become more context-aware, Microsoft executives said.

Still, risks loom. Regulatory scrutiny over the energy consumption and environmental impact of AI datacenters is intensifying, particularly in the U.S. and European Union. If new rules slow datacenter construction or impose carbon taxes, Microsoft’s expansion plans could face headwinds. Additionally, the enormous capital outlay means that any slowdown in AI demand—whether from enterprise fatigue, open-source model commoditization, or macroeconomic pressures—would leave Microsoft with stranded assets. Hood conceded that the company stress-tests against a “muted demand scenario,” but she expressed confidence that AI adoption is in the early innings. “We’re not building for next quarter; we’re building for 2030,” she said.

Looking ahead, Microsoft reiterated its guidance for fiscal Q4 revenue of $87.5 billion to $89.0 billion, implying continued Azure acceleration. The company also teased the coming launch of “Azure AI Zones”—geographically localized AI instances that will reduce latency for time-sensitive applications such as autonomous vehicle coordination and immersive AR. Meanwhile, the engineering teams are pushing toward liquid-cooled server designs that can pack more compute into the same megawatt footprint, which would further improve the revenue-per-gigawatt metric.

As the AI boom enters its third full year, Microsoft’s quarterly numbers make one thing plain: the line between a software company and an energy-infrastructure company has blurred. The race to power AI is now the race to build, connect, and sustain datacenters at a scale never seen in the technology sector. With a 1 GW splash in a single quarter, Microsoft is betting that this hybrid future—where watts and tokens are two sides of the same coin—will define the next decade of computing.