Reliability Is the Real Constraint Summers evaluates every large-scale power decision against four pillars: legal, economic, sustainable, and reliable. In the current market, one dominates. Reliability — defined not merely as uptime, but as certainty of project execution — has become the industry’s most pressing problem. “There’s a lot of noise in the market,” Summers says. “The question is whether a project is real; whether it can be delivered on time, and whether it can maintain multiple nines once it’s operating.” Legal frameworks for behind-the-meter generation are largely settled. Economics matter, particularly across multi-year development cycles. Sustainability factors in, though in many cases it has been deferred behind more immediate concerns. Execution, by contrast, is now existential. Hyperscalers are no longer evaluating power sources alone: they are evaluating delivery credibility. From Megawatts to Certainty, Speed, and Risk Transfer Historically, data centers relied on utilities to supply three things together: energy, predictable timelines, and manageable risk. That bundle has broken down. Utilities face long interconnection queues, uncertain delivery dates, and rising infrastructure costs. For developers, that uncertainty has created what industry observers and stakeholders are starting to call a “power certainty premium,” i.e. a willingness to pay more for guaranteed timelines. GPC’s customers, Summers says, are no longer buying megawatts alone. They are buying speed to market, certainty of delivery, and risk transfer. “Even if the timeline isn’t shorter, they want a date certain,” he notes. “Utilities often can’t provide that today.” That evolution is driving demand for on-site, behind-the-meter generation, where developers control timelines and cost structures rather than waiting on grid expansion. Supply Chain as the New Critical Path Remove the grid and a new constraint appears: equipment availability. For GPC, the primary gating factor is supply chain; specifically the “prime mover,” the generation equipment itself. Large industrial turbines

According to a report that Backblaze released this morning, traffic from content delivery networks and hosting and Internet services providers have stayed largely within historical norms over the past year. But traffic from hyperscalers and neoclouds fluctuated dramatically, with steep climbs in September and October and another uptick in March. Another network traffic change related to AI is geography. “Traditionally, it didn’t matter where cloud infrastructure was located,” says Nowak. But with AI workloads, if storage is close to compute, enterprises get lower latency and higher throughput. Today, Virginia and California have a high concentration of AI compute providers. This, in turn, brings in more storage companies. “In July, we chose to double our footprint in US East to increase the proximity to hyperscalers and neoclouds,” says Nowak. And that, in turn, leads to even more demand for compute, and even greater concentration. “There’s a snowball effect,” Nowak says. Why neoclouds for AI? Enterprises might think that they don’t need to worry about network traffic details if they’re using a hyperscaler for their AI workloads because the data and the processing both stay within the cloud. But there are advantages to using a third-party storage provider combined with neoclouds for the GPUs. According to a report released by Synergy Research Group in early April, neocloud revenues hit $9 billion in the fourth quarter of 2025, a 223% year-over-year increase. Revenues passed $25 billion for the whole year and are expected to hit $400 billion by 2031.

Enterprise AI adoption is no longer emerging. It is already embedded and beginning to scale in ways that will reshape data center demand. The latest TD Cowen GenAI Adoption Survey makes that clear. Across 689 U.S. enterprises, 92% are now using at least one major AI platform, with Microsoft Copilot, Google Gemini, and ChatGPT forming the core triad of daily enterprise tooling. That’s the baseline. The more important story is what comes next. AI is moving quickly from assistive software to autonomous systems, and that shift carries direct implications for compute demand, power consumption, and infrastructure design. From Copilots to Autonomous Systems Today’s enterprise AI footprint is already broad, but it is still largely human-in-the-loop. That is beginning to change. Roughly a third of respondents say they already have semi-autonomous AI agents running in production, while another large cohort is piloting or planning deployments over the next 12 to 18 months. By 2027, more than three-quarters expect to be running AI agents capable of executing multi-step workflows without human intervention. This is not incremental adoption. It is a step-function shift. Autonomous agents don’t just respond to prompts; they execute tasks, interact with enterprise systems, and continuously access data. For data centers, that translates into more persistent, baseline load: exactly the kind of demand profile that stresses power delivery, increases utilization, and accelerates capacity planning timelines. To wit: AI is moving from a bursty workload to a continuous one. ROI Is No Longer the Question At the same time, the debate around AI return on investment is effectively over. Three-quarters of respondents report positive ROI, while only a small minority report negative outcomes. A meaningful share is already seeing multiples of return on their investments. The implication seems straightforward: AI budgets are becoming durable. This is no longer experimental spend that

Self-Sufficiency Becomes a Feature, Not a Risk Consider Wyoming’s Project Jade, where county commissioners approved an AI campus tied to 2.7 GW of new natural gas-fired generation being developed by Tallgrass Energy. Reporting from POWER described the project as a “bring your own power” model designed for a high degree of self-sufficiency, with a mix of natural gas generation and Bloom fuel cells. The campus is expected to scale significantly over time. What stands out is not only the size, but the positioning. Self-sufficiency is becoming a selling point both for developers seeking to de-risk timelines, and for local stakeholders wary of overloading existing utility infrastructure. Fuel Cells and Nuclear: The Middle Ground and the Long Game Fuel cells occupy an important middle ground in this shift. Bloom Energy’s 2026 report positions fuel cells as a leading onsite option due to shorter lead times, modular deployment, and lower local emissions. Market activity suggests that interest is real. For developers, fuel cells can be easier to permit than large turbine installations and can be deployed incrementally. That makes them effective as bridge-to-grid solutions or as permanent components of hybrid architectures. Advanced nuclear remains the most strategically significant, but least immediate, BYOP pathway. Companies including Switch and other data center operators have explored partnerships with Oklo around its Aurora small modular reactor design. Nuclear holds long-term appeal because it offers firm, low-carbon power at scale. But for current AI buildouts, it remains a future option rather than a near-term construction solution. The immediate reality is that gas and modular onsite systems are closing the time-to-power gap, while nuclear is being positioned as a longer-duration successor as licensing and deployment timelines evolve. The model itself is also evolving. BYOP is beginning to blur the line between developer, energy provider, and compute customer. Reuters

So far in 2026, across the United States and overseas, Microsoft is building an infrastructure portfolio at full hyperscale. The strategy runs on two tracks. The first is familiar: sovereign cloud expansion involving new regions, local data residency, and compliance-driven enterprise infrastructure. The second is larger and more consequential: purpose-built AI factory campuses designed for dense GPU clusters, liquid cooling, private fiber, and power acquisition at a scale that extends far beyond traditional cloud infrastructure. Despite reports last year that Microsoft was pulling back on data center development, the company is accelerating. It is not only advancing its own large-scale campuses, but also absorbing premium AI capacity originally aligned with OpenAI. In Texas and Norway, projects tied to OpenAI’s infrastructure plans have shifted back into Microsoft’s orbit. Even after contractual changes gave OpenAI greater flexibility to source compute elsewhere, Microsoft remains the market’s most reliable backstop buyer for top-tier AI infrastructure. It no longer needs to control every OpenAI build to maintain its position. In 2026, Microsoft is still the company best positioned to turn uncertain AI demand into deployed capacity, e.g. concrete, steel, power, and silicon at scale. Building at Industrial Scale The clearest indicator of Microsoft’s intent is its capital spending. In its January 2026 earnings cycle, Reuters reported that Microsoft’s quarterly capital expenditures reached a record $37.5 billion, up nearly 66% year over year. The company’s cloud backlog rose to $625 billion, with roughly 45% of remaining performance obligations tied to OpenAI. About two-thirds of that quarterly capex was directed toward compute chips. To be clear: this is no speculative buildout. Microsoft is deploying capital against a massive, committed demand pipeline, even as it maintains significant exposure to OpenAI-driven workloads. The company is solving two infrastructure problems at once: supporting broad Azure and Copilot growth, while ensuring

At Data Center World 2026, the industry didn’t need convincing that something fundamental has shifted. “This feels different,” said Bill Kleyman as he opened a keynote fireside with Phill Lawson-Shanks and Amber Caramella. “In the past 24 months, we’ve seen more evolution… than in the two decades before.” What followed was less a forecast than a field report from the front lines of the AI infrastructure buildout—where demand is immediate, power is decisive, and execution is everything. A Different Kind of Growth Cycle For Caramella, the shift starts with scale—and speed. “What feels fundamentally different is just the sheer pace and breadth of the demand combined with a real shift in architecture,” she said. Vacancy rates have collapsed even as capacity expands. AI workloads are not just additive—they are redefining absorption curves across the market. But the deeper change is behavioral. “Over 75% of people are using AI in their day-to-day business… and now the conversation is shifting to agentic AI,” Caramella noted. That shift—from tools to delegated workflows—points to a second wave of infrastructure demand that has not yet fully materialized. Lawson-Shanks framed the transformation in more structural terms. The industry, he said, has always followed a predictable chain: workload → software → hardware → facility → location. That chain has broken. “We had a very predictable industry… prior to Covid. And Covid changed everything,” he said, describing how hyperscale demand compressed deployment cycles overnight. What followed was a surge that utilities—and supply chains—were not prepared to meet. From Capacity to Constraint: Power Becomes Strategy If AI has a gating factor, it is no longer compute. It is power. “Before it used to be an operational convenience,” Caramella said. “Now it’s a strategic advantage—or constraint if you don’t have it.” That shift is reshaping executive decision-making. Power is no