Speed Is the New Capacity: Why AI Infrastructure Needs a New Power Strategy

Lawrence Berkeley National Laboratory identified more than 40 approaches to accelerating large-load connections for AI data centers, as power delivery speed replaces raw capacity as the primary constraint on infrastructure development. Reuters reported in August 2026 that planned European AI data centers coming online between 2026 and 2028 sit an average of 175 kilometers from major cities, versus 46 kilometers for projects delivered between 2022 and 2025, while PJM Interconnection says more than 50 GW of generation projects already hold grid connection agreements but face permitting, equipment, and workforce delays. Elon Musk said in August that SpaceX would begin casting gas-turbine blades and vanes in-house as turbine generator demand strains the supply chain.

Insight and analysis on the data center space from industry thought leaders. Speed Is the New Capacity: Why AI Infrastructure Needs a New Power Strategy For AI infrastructure, power delivery speed now matters more than capacity alone, reshaping site selection and requiring phased, flexible power strategies. For large infrastructure projects, power capacity has traditionally been treated as a quantity. A site needs 100 MW, 500 MW, or a gigawatt, and developers plan around whether the local power system can ultimately supply that amount. AI infrastructure places greater emphasis on timing. A gigawatt promised years from now has little value to a campus whose compute, customers, and capital are ready today. For developers racing to bring capacity online, the useful measure is no longer planned megawatts alone. It is the number of megawatts that can actually be delivered on schedule. Power Planning Moves Upstream That reality pulls power planning much earlier in the development process. Lawrence Berkeley National Laboratory LBNL recently identified https://emp.lbl.gov/news/beyond-bottleneck-new-report-identifies-solutions-accelerating-large-load-connections more than 40 approaches to accelerating large-load connections, covering load forecasting, interconnection, resource planning and procurement, market operations, and cost allocation. The range of solutions reflects the number of steps that can delay the availability of usable power. Grid studies, transmission upgrades, generation development https://www.datacenterknowledge.com/energy-power-supply/why-data-centers-produce-their-own-power , equipment procurement, permitting, and construction all sit on the same critical path. Developers Are Siting for Speed-to-Power Developers are already choosing sites around that constraint. Reuters reported https://www.reuters.com/business/europe-ai-data-centres-seek-cheaper-quicker-energy-land-2026-08-19/ in August 2026 that planned European AI data centers coming online between 2026 and 2028 are located an average of 175 kilometers from major cities, compared with 46 kilometers for projects delivered between 2022 and 2025. JLL attributed much of the movement to the search for cheaper land, available energy, and faster grid connections. Power is beginning to determine where digital infrastructure gets built. Procurement Models Are Evolving Procurement is changing too. In PJM Interconnection PJM , more than 50 GW https://insidelines.pjm.com/new-interconnection-process-delivers/ of generation projects already have grid connection agreements, yet permitting, equipment, and workforce constraints continue to delay development. PJM is responding by facilitating bilateral agreements between generators and large loads that can support new capacity outside the traditional procurement model. The Federal Energy Regulatory Commission FERC has also directed regional grid operators https://www.datacenterknowledge.com/build-design/ferc-targets-grid-rules-for-data-centers-and-large-loads to address co-location, behind-the-meter generation, and more flexible transmission service for large loads. These strategies can shorten the wait for grid-supplied power, but they move more of the development burden onto the project itself. A developer pursuing onsite generation now has to secure turbines https://www.datacenterknowledge.com/sustainability/replacing-diesel-in-ai-scale-data-centers-gas-engines-turbines-and-steam , fuel, permits, and supporting infrastructure. A project co-located with generation needs the right site and commercial structure. A bilateral agreement still depends on the generator being built on time. The Supply Chain Is the Schedule SpaceX offers an extreme example of how a single dependency can become the schedule constraint. In August, Elon Musk said the company would begin casting gas-turbine blades and vanes in-house as demand for turbine generators strains the existing supply chain. Tom’s Hardware https://www.tomshardware.com/tech-industry/data-centers/spacex-starts-in-house-turbine-blade-manufacturing-to-boost-gas-powered-generator-output-for-elons-ai-data-centers-new-manufacturing-strategy-cuts-generator-delays-by-18-months reported that specialized turbine blades can require 60 to 90 weeks to manufacture. Musk said https://www.businessinsider.com/elon-musk-spacex-build-gas-turbine-parts-meet-ai-demand-2026-8 bringing production in-house could shorten turbine delivery by as much as 18 months. The lesson is simple: speed-to-power https://www.datacenterknowledge.com/energy-power-supply/speed-to-power-how-developers-are-restructuring-for-ai-demand depends on the entire supply chain behind the megawatt. Moving around the interconnection queue helps only if the generation equipment can be delivered. Securing turbines helps only if fuel and permits are available. Co-location helps only if the site can support the required infrastructure. Each project needs to identify the constraint most likely to determine when power actually reaches the servers. Build Power in Phases with a Flexible Portfolio A more flexible power strategy spreads risk and accelerates first power. Utility service, bilateral contracts, onsite generation, storage, energy recovery, and other distributed resources can come online on different schedules. Developers can use that mix to build power capacity in phases, bringing an initial block online while larger grid or generation projects continue in parallel. Modular power and cooling systems can follow the same model, adding infrastructure as compute capacity grows rather than requiring the entire campus energy system to be complete on day one. Separate Real Demand from Noise Faster development also requires credible demand forecasts. Reuters reported https://www.reuters.com/business/energy/us-power-use-beat-record-highs-2026-2027-ai-use-surges-eia-says-2026-09-09/ on Sept. 9 that large-load requests, driven primarily by data centers, have grown far beyond existing US data center demand in several major power markets. Texas and other states are imposing deposits, milestones, and other requirements intended to distinguish financed, shovel-ready projects from speculative requests. Utilities cannot plan generation and transmission efficiently when the same prospective demand appears in several queues or projects reserve capacity they may never use. Speed with Certainty Wins The strongest AI infrastructure projects combine speed with certainty. Developers who control land, financing, and permits understand their load ramp. They have a realistic plan for bringing power online in stages and give utilities and generation partners something concrete to build around. They also give themselves more options when one part of the power supply chain falls behind. Bottom Line Power planning now belongs in the first stage of site development alongside land, connectivity, cooling, and permitting. The relevant question is not simply how much capacity a market can eventually provide. Developers need to understand when each power source can arrive, what physical infrastructure it requires, and which constraints could delay its arrival. AI infrastructure has put electricity directly on the project’s critical path. The competitive advantage will go to developers that can turn planned megawatts into operating megawatts faster and continue adding capacity as demand grows. In that sense, speed has become part of capacity itself.