AI Data Centers Strain U.S. Grids as Transmission Lags

Wood Mackenzie warns that power demand from AI data centers is growing faster than U.S. grids can support. The consultancy estimates major transmission upgrades will be five to ten years away, leaving projects, markets and customers exposed while operators pursue backup generation and conditional interconnection agreements.

Utilities and grid operators are prioritizing flexible interconnection arrangements and onsite generation to keep new data centers running. The firm cautions those fixes carry technical, regulatory and economic risks that many in the sector do not fully recognize. Ben Hertz-Shargel, head of grid transformation and large loads at Wood Mackenzie, warned: “The power sector is fixated on data center flexibility, but that is not the end-game for grid operators or data center operators. Firm grid service is the goal, backed by new transmission superhighways.” He said there is limited awareness of the risks facing colocation projects and conditional interconnections.

Wood Mackenzie’s modeling finds the grid would need about 16.4 gigawatts of new gas-fired capacity per year through 2035 to match projected demand, compared with roughly 4 gigawatts added per year between 2023 and 2025. Chris Seiple, vice chairman for energy transition and power and renewables at Wood Mackenzie, noted: “Load growth and affordability are in direct opposition in the deregulated markets.” He added that if capacity prices rise enough to attract new generation, consumer bills could increase and prompt political pushback.

Regional data show the imbalance. PJM, the mid-Atlantic grid operator, lists about 78 gigawatts of committed data center load while only 36 gigawatts of accredited generation sits in its pipeline. In Texas, market prices of about $30 to $40 per megawatt-hour remain far below the $78 to $100 per megawatt-hour Wood Mackenzie calculates would be needed to draw new gas plants.

Many operators are pursuing colocated generation and flexible interconnection models. Wood Mackenzie found more than 90 gigawatts of colocated generation in U.S. interconnection queues, but it said those approaches are mainly achievable for the largest, best-capitalized hyperscalers. The report lists several technical challenges: near-instantaneous swings in AI workloads can stress reciprocating engines and gas turbines; lithium-ion batteries can buffer swings but have limited lifespans; ultra-fast response battery technologies are not widely commercialized; irregular cooling and GPU loads create power harmonics that can overheat equipment if not filtered; and sub-synchronous oscillations pose stability risks for local and remote generators.

Regulatory and business problems layer on the technical issues. Data center operators seek protections against downtime that can lead to demands for firm service commitments and contractual terms that are hard to fit within existing grid rules. Wood Mackenzie also highlights a cost-allocation dilemma: grid operators plan close to $100 billion in transmission investments to support large new loads, but current allocation methods could spread those costs across all ratepayers rather than assign them to the large consumers driving the need.

Developers are responding by concentrating capacity in regions with more available power and land or by shifting some capacity to emerging markets with fewer grid constraints. Research cited by Wood Mackenzie shows 850 megawatts of data center power capacity was delivered across Europe, the Middle East and Africa in the first three quarters of 2025, an 11 percent decline from the same period a year earlier.

Wood Mackenzie says lead times for new transmission-often five to ten years-mean choices now about interconnections, onsite generation and cost allocation will affect the pace of data center deployment and the financial impact on other electricity customers.