Data centre power hunger drives GW-scale generation bets from gas plants to renewables
BloombergNEF estimates that global data centre electricity demand will nearly quadruple by 2035. The pipeline backing that trajectory is already visible: 181 GW of data centre projects worldwide are tied to natural gas power generation, according to BloombergNEF data. A single week of announcements spanning three continents illustrates how AI load growth is pulling forward commitments in gas, renewables, storage, and grid regulation simultaneously.
Gas as the fast path
Homer City Generation secured an operator for its planned 4.4 GW natural gas plant in Indiana County, Pennsylvania. That single facility, if built to full capacity, would rank among the largest gas plants anywhere. GridAI, meanwhile, aims to deliver its first 300 MW of power in 2027. Both projects treat gas as the dispatchable resource that can meet data centre load profiles on a timeline renewables and nuclear alone struggle to hit.
The 181 GW of gas-tied data centre projects globally suggests the same calculus is playing out far beyond any one jurisdiction. If even a fraction of that pipeline reaches final investment decision, gas turbine manufacturers and pipeline developers stand to gain a demand anchor that persists well into the next decade. Grid operators, conversely, would face concentrated load additions at a pace their planning cycles were not designed for.
Renewables and storage at matching scale
Not every new project defaults to gas. In Australia, AirTrunk and European Energy Australia signed an exclusivity deal to develop more than 1 GW of solar, wind, and battery capacity in New South Wales to support a 550 MW data centre campus. The ratio matters: roughly 1 GW of variable generation and storage paired with 550 MW of firm data centre demand implies a build that accounts for capacity factors and intermittency from the outset.
In Malaysia, a multi-year plan targets at least 5 GW of AI compute capacity across Sarawak and Peninsular Malaysia over the long term. At that scale, the generation fleet serving those facilities - whatever its fuel mix - would need to be comparable to an entire mid-sized national grid.
The contrast between the Australian and gas-heavy approaches points to a divergence. Where renewable resources are strong and co-located storage is commercially viable, data centre operators can anchor long-term power purchase agreements to clean generation. Where speed to market or around-the-clock firm supply dominates the calculus, gas wins on deliverability. Both paths are attracting capital at GW scale.
Grid operators start to push back
ISO New England's proposed transmission services, study processes, and reliability requirements aim to prevent large new electricity loads from compromising resource adequacy. That intervention signals a grid operator recognizing that concentrated demand additions - data centres chief among them - carry system-level risk if they outrun generation and transmission build.
The tension is structural. Data centre developers optimize for site selection, fiber connectivity, and speed. Grid planners optimize for long-run adequacy across an entire balancing area. When a single campus draws 550 MW or a single gas plant is sized at 4.4 GW, the load or generation is large enough to shift the resource adequacy equation for the region it sits in. ISO New England's rule proposal would, if adopted, force large new loads through a reliability screen before interconnection, adding time to projects whose economics depend on rapid deployment.
Who is exposed
Three groups face material consequences. Gas turbine suppliers and gas infrastructure developers are positioned to capture orders from the 181 GW pipeline of gas-tied data centre projects. Their order books could thicken if even a modest share of that pipeline converts to firm contracts. Renewable developers with co-located storage, like the AirTrunk-European Energy Australia venture, compete for the same campus-level deals but carry technology risk around storage duration and curtailment.
Grid operators and regulators sit in the middle. ISO New England's rule proposal is an early signal that system operators will impose new requirements on large loads. If other grid operators follow that lead, data centre developers face a longer and less predictable permitting path, which in turn could favor behind-the-meter generation models - gas or otherwise - that bypass grid interconnection queues entirely.
The 5 GW target in Malaysia and the 4.4 GW gas plant in Pennsylvania share a feature: both are large enough that the generation or load they represent would reshape the local grid. If reliability screening rules like those ISO New England proposes become standard, the commercial structure of data centre power supply may shift from grid-connected offtake toward dedicated, self-supplied generation. That outcome would accelerate gas build while limiting the transmission investment that renewables at distance require.
BloombergNEF's near-quadrupling forecast by 2035 sets the demand envelope. The generation and grid decisions being made now - gas in Pennsylvania, renewables in Australia, compute-first planning in Malaysia, reliability rules in New England - determine how that demand gets served, by whom, and at what carbon cost.
Sources
- The Environmental Impact of AI: Data Center Energy and Water Consumption Challenges - Archyde (opens in a new tab) - Event Registry
- Operator selected for Homer City Generation natural gas plant (opens in a new tab) - Event Registry
- GridAI to buy Wyoming gas firm Pronghorn to power AI data centres (opens in a new tab) - Event Registry
- AirTrunk, European Energy lock in $1.5bn renewable project for Hunter region (opens in a new tab) - Event Registry
- Grab and Softbank to seed $2.37b AI push in Malaysia (opens in a new tab) - Event Registry
- Open ISO-NE board meeting considers strategic plan, large loads, winter preparedness (opens in a new tab) - ISO New England Newswire