U.S. data center construction tied to artificial intelligence is increasingly being limited by access to electricity rather than the availability of chips, creating a projected 32-gigawatt net power shortfall by 2028, according to Morgan Stanley. The bank estimates the deficit would equal roughly 34% of expected electricity supply even after mitigation measures, placing delivery schedules for servers, networking hardware and supporting infrastructure under greater pressure.
The constraint changes the economics of AI expansion. Semiconductor makers can manufacture and ship accelerators, but data center operators cannot put full systems into service until utilities deliver enough generation, transmission and substation capacity. That leaves project developers, equipment suppliers and cloud customers exposed to delays measured in years rather than quarters.
Morgan Stanley separately estimated that new U.S. data center demand between 2026 and 2028 could reach about 68GW, compared with roughly 30GW of built or contracted capacity. That methodology implied a 38GW gap, underscoring how sharply demand forecasts have moved beyond the grid’s current connection pipeline.
Grid connection queues clash with data center timelines
The challenge is not simply building more power plants. Interconnection—the process of linking new generation, storage and large customers to the grid—has become a central obstacle.
Lawrence Berkeley National Laboratory’s 2026 Queued Up report found that more than 2,060GW of generation and storage capacity was awaiting U.S. grid interconnection at the end of 2025. Typical projects can require five to seven years to clear studies, approvals and construction, while major data center customers frequently seek access to power within two to three years.
That mismatch makes local conditions particularly decisive. A developer may have land, financing and contracted computing equipment, yet remain unable to operate until a utility expands a substation, installs transformers or completes transmission work. Delays can also cascade through the supply chain because server racks, memory, optical equipment and power-management components are often scheduled around a planned energization date.
U.S. electricity demand is already setting records. The Energy Information Administration forecasts total consumption of 4,135 billion kilowatt-hours in 2026 and 4,211 billion kWh in 2027. Commercial-sector electricity use is expected to rise 3.3% in 2026 and 2.7% in 2027, accounting for an estimated 63% and 56%, respectively, of total load growth in those years.
Texas pause shows how approvals can reshape project calendars
Texas has become an immediate example of the permitting and grid-planning risk facing large computing campuses. On Aug. 3, 2026, Governor Greg Abbott ordered a pause on new data center connections to the Electric Reliability Council of Texas grid while audits are conducted. The order affects about 49.8GW of pending load, according to the supplied reporting.
The national data center development pipeline was estimated at about 253GW, putting the affected Texas projects at close to one-fifth of that total. ERCOT’s large-load interconnection queue stood at around 474GW, with data centers representing roughly 90% of the queue.
BloombergNEF estimated that extended delays could produce cumulative cost impacts of $8 billion to $15 billion by the first quarter of 2027. The costs would not fall evenly across the technology sector. They would include postponed construction, idle or rescheduled equipment, and deferred revenue for suppliers whose products cannot be installed until entire facilities are ready.
Texas also illustrates a problem for firms attempting to treat power as a readily available input. Large-load requests can exceed the capacity utilities have planned for in a specific region, forcing regulators and grid operators to weigh reliability concerns alongside economic-development promises.
Core chip suppliers have more flexibility than rack-level vendors
Morgan Stanley’s assessment distinguishes between companies selling priority computing components and suppliers dependent on fully commissioned racks and servers.
Nvidia and Broadcom were described as relatively insulated from individual site delays because customers are likely to prioritize accelerators and crucial networking hardware when capacity is constrained. Large cloud operators can also shift workloads, and in some cases equipment deployment, toward regions where power is available.
Nvidia reported fiscal 2027 second-quarter revenue of $96.2 billion, up 106% from a year earlier. Its data center revenue rose 117% to $89 billion, and the company forecast $108 billion in third-quarter revenue. Broadcom reported fiscal third-quarter revenue of $29.6 billion, up 86%, including $16.7 billion in AI semiconductor revenue, a 221% increase. It guided for $21.7 billion in AI semiconductor revenue in its fiscal fourth quarter.
Nvidia has also moved closer to securing power capacity directly. In disclosures related to SB Energy’s PORTS-Pike project in Ohio, Nvidia said it provided credit support connected to land, power and facility construction. The arrangement covers an initial 4.25 IT-GW, with an option for another 3.75 IT-GW, and includes a $1.5 billion Nvidia investment in SB Energy. Nvidia expects phased availability from 2028, while associated grid investment is estimated at at least $4.2 billion.
That structure gives a chip supplier a more direct stake in the physical infrastructure required to turn hardware demand into operating computing capacity.
Memory manufacturers, optical-module suppliers and power-management companies face a different timing profile. Their products are frequently delivered alongside completed rack installations, making them more vulnerable to rescheduling when sites cannot be energized. Micron reported fiscal 2026 fourth-quarter revenue of $54.23 billion, including $18 billion from its core data center business, and forecast fiscal 2027 first-quarter revenue of $61.5 billion, plus or minus $1.5 billion.
Server integrators occupy an especially exposed position between chip demand and site readiness. Dell Technologies reported $60.9 billion in AI server orders and a $95 billion AI backlog in fiscal 2027’s second quarter, while raising its full-year outlook for AI-optimized server revenue to $74 billion. A large backlog offers visibility, but it does not ensure that shipments can be recognized on the initially planned timetable if customers lack utility connections.
Transformers, cooling and generation become critical equipment
Grid hardware is becoming a practical limit on AI construction. Industry reporting cited U.S. power-transformer lead times of around four years, with prices rising about 80% over five years. Demand between 2019 and 2025 rose 274% for generator step-up transformers and 116% for substation transformers.
Manufacturers are responding with new capacity. Hitachi Energy has disclosed about $1 billion in investment, including a South Boston factory targeted for 2028. Siemens committed $421 million to a transformer plant in Charlotte. Eaton reported $8.5 billion in second-quarter revenue, with Electrical Americas trailing-12-month orders growing organically by 41% and Electrical Global backlog up 103% year over year.
Higher rack densities also require more cooling. Vertiv reported second-quarter net sales of $3.274 billion, up 24%, and raised its 2026 revenue outlook to between $13.8 billion and $14.2 billion.
Generation developers are pursuing gas, nuclear, solar and storage projects, though their delivery dates differ sharply. GE Vernova reported $24.2 billion in second-quarter orders and said its gas-power equipment backlog plus slot-reservation agreements had reached 116GW. Constellation Energy announced a 20-year agreement to supply 2GW to a hyperscale data center developer near its Susquehanna nuclear plant in Pennsylvania through behind-the-meter connections, an arrangement designed to avoid some grid-queue constraints.
Mining operators face regional competition for power
The tighter market for large-load connections also affects proof-of-work cryptocurrency miners, whose facilities require sustained electricity access and can compete with AI data centers for suitable sites. The Cambridge Blockchain Network Sustainability Index estimated global electricity consumption for Bitcoin mining at 138 terawatt-hours in 2026.
The impact will vary by location and contract structure. Mining operations with long-term power agreements, flexible load arrangements or access to generation may face a different cost profile from facilities seeking new grid connections. AI data center developers generally have greater ability to support expensive dedicated infrastructure because their customers are funding high-value computing workloads.
Claims that grid constraints would automatically weaken proof-of-work network security or force transaction fees higher go beyond the available evidence. Hash rate depends on mining economics, hardware efficiency, bitcoin prices and electricity costs across many jurisdictions, rather than on a single U.S. interconnection policy.
The immediate consequence is more concrete: electricity access has become a scarce development asset. For AI infrastructure builders and power-intensive crypto operators alike, securing dependable capacity now increasingly determines which projects can move from announced construction plans to running machines.
Want to see how AI transforms trading too? Explore automated strategies with AI copy trading today.
Disclaimer: The content on this page is provided for general informational purposes only and does not represent the views or financial advice of Toobit. We make no guarantees regarding the accuracy or completeness of this information and shall not be held liable for any errors, omissions, or outcomes resulting from its use. Investing in digital assets involves risk; users should independently evaluate their financial situation and the risks involved. For further details, please consult our Terms of Service and Risk Disclosure.
