U.S. data center operators are increasingly building their own power generation systems rather than waiting for grid connections, with developers planning roughly 56 gigawatts (GW) of on-site capacity as of early 2026. This represents 30 percent of all planned data center builds nationwide and reflects a fundamental shift in how the AI infrastructure boom is being powered.

Why Are Data Centers Turning Away From the Grid?

The bottleneck is straightforward: electricity demand is growing far faster than the grid can handle. U.S. electricity consumption is projected to grow by 25 to 50 percent by 2050, yet grid interconnection queues have swelled to over 2,200 GW of planned capacity, far exceeding the nation's total installed capacity of 1,400 GW. Meanwhile, new transmission lines take six to seven years to build, and some projects face delays of up to 15 years.

For AI data center developers who need power now, waiting is not an option. The mismatch between demand timelines and infrastructure buildout has forced a strategic pivot toward self-sufficiency. Rather than queue up for grid connections that may take years, hyperscalers and data center operators are investing in solar arrays, battery storage systems, and other distributed energy resources located directly at their facilities.

How Are Data Centers Powering Themselves?

• On-Site Solar and Storage: Developers are deploying utility-scale solar panels and battery energy storage systems (BESS) co-located with data centers to generate and store power independently of the grid.
• Virtual Power Plants (VPPs): Tech companies are partnering with residential solar and battery aggregators to create flexible, dispatchable capacity that can be called upon when needed, with partnerships like Sunrun, Tesla, and Renew Home planning a 16 GW VPP using home batteries and smart devices.
• Grid-Enhancing Technologies: Demand flexibility tools, including smart devices and battery storage, could unlock up to 200 GW of demand response capacity across utilities and wholesale markets, according to research cited in the report.

Texas is leading this trend with over 20.6 GW of planned behind-the-meter capacity, followed by New Mexico (9.2 GW), Pennsylvania (7.5 GW), and Utah (6.0 GW). Grid operators are actively encouraging this shift. For example, the Southwest Power Pool introduced its High Impact Large Load Generation Assessment (HILLGA) framework, which offers connection agreements in under 90 days for data centers that bring their own generation to support their load.

What Does This Mean for AI Infrastructure Beyond Data Centers?

The networking infrastructure supporting AI is evolving alongside the power challenge. Industrial networking provider Belden reported record second-quarter results driven partly by demand for AI-ready infrastructure, including a new approximately $20 million fiber connectivity win with a Tier-1 hyperscaler for AI data center deployments. The company booked roughly $40 million in hyperscaler orders during the quarter, demonstrating that AI infrastructure extends far beyond just computing power.

Belden's results also highlighted how AI infrastructure is spreading into physical industrial environments. A major global retailer selected Belden technology to support autonomous systems across U.S. distribution facilities, representing an estimated $20 million opportunity over four years. As industrial environments increasingly deploy robotics, machine vision, and autonomous systems, reliable wired and wireless networking becomes critical for real-time decision-making and operational performance.

Belden's July 1 acquisition of RUCKUS Networks significantly expands the company's ability to deliver complete networking solutions across both operational technology and information technology environments. The combined portfolio now spans passive infrastructure, industrial networking, enterprise switching, wireless connectivity, and cloud-based management software, with solutions representing approximately 20 percent of combined revenue.

How Are Global Players Scaling AI Infrastructure?

Beyond the U.S., regional players are building massive AI infrastructure platforms. Indosat Ooredoo Hutchison, in collaboration with Ooredoo Group, Nokia, and NVIDIA, launched Zankore by Indosat, targeting 1 gigawatt (GW) of NVIDIA DSX AI Factory capacity to serve Southeast Asia. The platform will deliver approximately 200 megawatts (MW) of AI capacity in the first half of 2027, powered by NVIDIA GB300 NVL72 processors.

"AI infrastructure is becoming a fundamental layer of the digital economy, and we believe the greatest opportunities will be created by investing early, at scale and with the right partners," stated Ulf Ewaldsson, Chief Executive Officer of Zankore by Indosat.

Ulf Ewaldsson, Chief Executive Officer, Zankore by Indosat

Zankore will also incorporate NVIDIA DSX MaxLPS, which dynamically optimizes power across the GPU fleet to recover stranded power and increase compute capacity within the available facility power envelope. This technology is designed to enable up to 40 percent more compute within the same power budget, increasing Zankore's capacity to serve customers and generate revenue.

The shift toward on-site power generation and distributed energy resources reflects a broader recognition that AI infrastructure requires not just computing power and networking, but also reliable, fast-deployable energy solutions. As greenfield transmission projects continue to face permitting and supply chain constraints, the ability to deploy behind-the-meter solar, quick-to-interconnect storage, and aggregated virtual power plant capacity is moving from an alternative strategy to an essential path forward for powering U.S. digital infrastructure.