US Data Center Gas Demand Could Reach 18 Bcf a Day by 2035

US data center gas demand is now projected to reach roughly 18 billion cubic feet per day by 2035. That would exceed the combined consumption of Germany and Japan.

The BloombergNEF forecast, reported in a new 2035 forecast, is nearly twice its estimate from nine months earlier. It also accounts for announced data center projects that analysts expect will never reach operation.

The comparison with two industrial economies is dramatic, but the deeper conflict sits inside the United States. AI companies want vast amounts of reliable electricity faster than utilities can build transmission and connect new generation. Natural gas can fill that gap, but doing so could raise energy costs and weaken corporate climate commitments.

The Forecast Turns Data Centers Into a Major Gas Consumer

The important change is not that some AI facilities will burn gas. It is the projected scale of the entire sector.

BloombergNEF estimates that data centers could consume about 18 billion cubic feet of natural gas daily by 2035. That total includes gas burned at dedicated facilities and at power plants serving data centers through the grid.

For perspective, US data centers currently account for roughly 3 billion cubic feet per day. Reaching 18 billion would represent about a sixfold increase within a decade.

The updated projection is also nearly double BloombergNEF's estimate from nine months earlier. Such a large revision shows how quickly planned AI infrastructure has outgrown previous energy assumptions.

The forecast does not assume that every announced campus will open. BloombergNEF reportedly discounted projects based on the probability that developers will complete them. That matters because data center announcements often exceed the capacity that utilities can realistically connect.

Grid-connected facilities would create most of the projected increase. By 2035, their electricity demand could drive about 15 billion cubic feet per day of additional power-sector gas consumption.

That growth would be five times greater than the increase attributed to every other grid-connected sector combined. Data centers would become the second-largest source of US gas-demand growth, behind liquefied natural gas exports.

Dedicated power plants would add another substantial load. Projects that generate electricity on-site could consume between 2.9 billion and 3.4 billion cubic feet per day by 2035.

On-site generation is also called behind-the-meter power. It supplies a customer directly instead of first delivering electricity through the public transmission system.

That category has attracted attention because Meta, Microsoft, Google, and Amazon have pursued projects involving dedicated gas generation. Yet it would still represent a minority of the forecast total.

The distinction matters for public debate. A visible gas plant beside an AI campus is easy to identify, permit, and criticize. Gas burned across a regional grid is harder to connect to any single data center.

Both forms ultimately respond to the same requirement. AI campuses need large quantities of electricity during nearly every hour of the year.

Training clusters can contain tens of thousands of accelerators, supported by networking, storage, and cooling equipment. Inference systems add another persistent load when people and software use deployed models.

Efficiency gains can reduce the electricity needed for each computation. They do not guarantee a decline in total demand when companies keep adding chips, users, and larger workloads.

That expansion is why US data center gas demand has moved from a utility-planning question into a national energy issue. At 18 billion cubic feet per day, the sector would compete with exports, industry, and households for fuel.

Why AI Infrastructure Is Turning Toward Natural Gas

Natural gas is gaining ground because it can provide continuous power sooner than many lower-carbon alternatives can reach commercial scale.

Data centers cannot operate only when wind or solar generation is available. Their servers require a stable supply, while backup systems must respond immediately to power interruptions.

Renewables can serve part of that load. However, an around-the-clock campus also needs transmission, storage, flexible generation, or another source of firm power.

Firm power refers to electricity that operators can reliably schedule when customers need it. Gas turbines can supply that electricity regardless of sunlight or wind conditions.

New transmission lines can take years to approve and build. Large generation projects must also wait for equipment, permits, fuel connections, and interconnection studies.

Grid interconnection is the process for connecting a new power plant or major customer to the electricity network. Queues have become a central obstacle for both generators and data centers.

A developer that secures land, chips, financing, and water can still fail to open on schedule without an approved power connection. That makes energy availability a competitive constraint.

Natural gas offers several advantages under those conditions. The United States has extensive production, storage, and pipeline networks. Gas plants can also provide high output from relatively compact sites.

A dedicated plant may let a data center bypass part of the grid connection process. The operator still needs turbines, pipelines, permits, and local approval, but it gains more control over the schedule.

A separate gas infrastructure analysis illustrates the same pressure through a different model. It projects incremental AI-linked gas demand of 7.6 to 11.5 billion cubic feet daily by 2035.

That range is lower than the BloombergNEF total because the studies use different baselines, scopes, and assumptions. It still describes a historically large new source of demand.

The analysis starts with 280 to 340 gigawatts of announced US data center capacity. It then adjusts for construction delays, permitting failures, efficiency improvements, and substitution by non-gas power.

Its base case assumes that 72% of announced data center capacity reaches completion. Under that scenario, incremental AI-linked gas demand reaches 5.8 billion cubic feet per day by 2030.

The infrastructure problem extends beyond gas production. Fuel must reach the correct location at sufficient pressure during periods of high electricity demand.

Pipelines, compressor stations, storage fields, and existing rights-of-way can therefore determine whether a site receives power on time. A region can have abundant gas while an individual project still faces a delivery shortage.

Equipment availability creates another bottleneck. Data center developers are competing for large turbines, transformers, switchgear, and other components already required by utilities.

These constraints help explain why companies are pursuing several energy routes at once. Big Tech has signed contracts involving nuclear power, geothermal energy, renewables, storage, and natural gas.

The energy supply outlook from the International Energy Agency expects natural gas to provide the largest increase in electricity generation for data centers through 2030. It projects more than 130 terawatt-hours of additional annual gas-fired generation.

Gas is not necessarily the preferred long-term resource for every technology company. It is often the resource available within the construction schedule that an AI strategy demands.

That schedule has become more aggressive as cloud providers compete for model developers, enterprise customers, and consumer traffic. Delaying a campus can mean delaying revenue and surrendering scarce computing capacity.

The rush therefore changes the energy decision. Companies are no longer selecting power only through cost and climate targets. They are increasingly selecting it through time to operation.

US Data Center Gas Demand Puts Utilities and Customers Under Pressure

The forecast transfers part of AI's infrastructure risk from technology companies to regional energy systems.

Utilities must decide how much generation and network capacity to build for customers whose future needs remain uncertain. Building too little risks shortages and delayed connections. Building too much can leave customers paying for underused assets.

A large data center can arrive much faster than a conventional utility forecast anticipates. Its electricity demand may also rival that of an established city or industrial facility.

The challenge becomes greater when several projects target the same region. They can request connections simultaneously, even though not every developer has firm financing or customers.

Utilities are responding with new tariffs, deposits, contract terms, and minimum payment requirements. These measures aim to keep households and small businesses from absorbing costs created by speculative projects.

Natural gas adds another layer of exposure. Power plants serving AI facilities will compete for fuel with residential heating, manufacturing, and liquefied natural gas terminals.

LNG exports convert domestic gas into liquid form for overseas shipment. Export terminals can operate as large, continuous sources of demand, much like data centers.

BloombergNEF expects LNG exports to remain the strongest source of gas-demand growth over the next decade. Data centers would rank second under its latest projection.

Those two sources can reinforce each other. Export terminals increase the connection between US gas prices and global markets, while data centers add concentrated domestic consumption.

A high-price scenario from energy analytics company Noreva describes periods when certain US gas hubs exceed $10 per million British thermal units. The company identifies LNG expansion and power demand as central pressures.

That is a scenario, not a consensus price forecast. Gas markets remain sensitive to weather, production growth, pipeline construction, storage levels, and export-terminal schedules.

However, the direction of pressure is clear. If data center gas use approaches 18 billion cubic feet daily, suppliers must increase production and delivery capacity.

Higher fuel costs would flow into electricity markets. Gas plants frequently set the marginal price, meaning their operating costs influence the price paid to multiple generators.

Technology companies can negotiate long-term contracts or build dedicated plants. Even then, their consumption can affect the regional market by increasing demand for fuel, equipment, and pipeline capacity.

Households have fewer options. A residential customer cannot relocate a home or negotiate a bespoke gas-supply agreement when regional costs increase.

Industrial users also face exposure. Chemical plants, fertilizer producers, glassmakers, and other manufacturers rely on gas as fuel or feedstock.

This creates the primary conflict surrounding the forecast. AI companies can use large balance sheets to secure power, while other customers remain tied to the same constrained system.

The impact will vary by location. Regions with existing pipelines, nearby production, and spare generation may absorb growth more easily.

Areas requiring new transmission or long-distance gas delivery will face longer schedules and larger infrastructure bills. Local regulators will decide who carries those costs.

The issue is therefore broader than whether a technology company pays for its own power plant. Regulators must also examine shared pipelines, reserve capacity, transmission upgrades, and market-wide fuel effects.

Developers may argue that new data centers expand tax bases and create construction work. Critics will ask whether those benefits justify long-lived infrastructure and potential rate increases.

Both claims require project-level evidence. A national demand forecast cannot determine whether one campus produces a fair local bargain.

The Gas Buildout Collides With Big Tech's Climate Promises

The fastest route to new AI capacity can also become the hardest route to reconcile with corporate emissions targets.

Natural gas plants release carbon dioxide when they generate electricity. Production and transportation can also release methane, which has a strong warming effect over shorter time periods.

According to the TechCrunch account, burning the additional gas in BloombergNEF's scenario would produce roughly 1 million metric tons of greenhouse gas emissions each day. That would equal about 12% of current US emissions.

That estimate reportedly includes emissions from extraction, processing, distribution, and combustion. Its precise result depends on assumed plant efficiency and methane leakage.

The potential increase would not be limited to companies that build power plants beside their campuses. Grid-connected data centers can drive additional fossil generation even when operators sign renewable energy contracts.

Annual renewable matching does not guarantee that a facility receives carbon-free electricity during every operating hour. A company may buy enough clean electricity across a year while drawing gas-fired power at night.

Hourly matching sets a stricter standard. It seeks carbon-free electricity in the same region during each hour of consumption.

That goal becomes difficult when load grows faster than clean generation, transmission, and long-duration storage. AI expansion has widened that timing mismatch.

Technology companies continue to invest in lower-carbon resources. Meta has announced agreements involving existing and advanced nuclear projects. Google and Microsoft have backed geothermal, nuclear, and carbon-management efforts.

Those investments may reduce future dependence on gas. Most advanced projects still need time to complete licensing, construction, fuel arrangements, and commercial validation.

Gas plants can operate for decades. Building them to solve a short-term electricity shortage creates the risk that temporary bridge capacity becomes permanent infrastructure.

Operators could later run those plants less often, add carbon capture, or replace them with other resources. Each route carries additional cost and technical uncertainty.

Carbon capture attempts to prevent carbon dioxide from reaching the atmosphere, then transports it for storage or use. Capture rates and project economics vary significantly across facilities.

The climate concern is not theoretical. A permit-based analysis of planned projects connected to 11 US data center campuses found potential annual emissions exceeding 129 million tons.

Air permits describe the maximum emissions a facility is authorized to produce. They do not prove that plants will operate continuously at those levels.

Still, permits reveal the scale developers are preparing to build. They also show how emissions can become concentrated around communities hosting AI infrastructure.

Local effects extend beyond carbon. Gas turbines can emit nitrogen oxides and other pollutants, while pipelines and compressor stations add land-use and safety questions.

New facilities may also require water, depending on the generating technology and cooling design. Communities must evaluate the combined demands of the power plant and data center.

Corporate climate accounting can obscure some of these tradeoffs. Market-based reporting lets companies claim renewable purchases that may occur at different times or locations from consumption.

Physical grids work differently. Operators balance supply and demand continuously, using the generators available within network constraints.

If incremental AI load keeps gas plants running longer, the resulting emissions remain part of the system impact. Renewable contracts can finance cleaner generation, but they do not erase every hour of fossil consumption.

This tension does not mean every data center project will follow the same path. Some locations can access abundant hydroelectric, nuclear, geothermal, wind, or solar resources.

The forecast instead indicates that gas remains the default balancing resource across much of the country. That role could expand before cleaner firm-power options reach sufficient scale.

What the 18 Bcf Forecast Does Not Prove

The projection is a warning about infrastructure direction, not a guaranteed measurement of demand nine years from now.

Forecasts depend heavily on how many proposed data centers reach operation. Developers regularly announce more capacity than utilities can connect or markets can support.

Capital availability could weaken. AI revenue might grow more slowly than expected. Chip efficiency could improve faster, while customers could shift workloads toward smaller models.

Electricity supply could also change. Faster transmission approvals, new nuclear capacity, geothermal projects, or cheaper long-duration storage would reduce gas demand.

Even the definition of AI-linked gas demand can vary. One study may count only new gas burned for incremental AI load. Another may include total consumption at all data centers.

Some models separate on-site generation from grid power. Others estimate the data center share of regional generation through dispatch assumptions.

That is why BloombergNEF's roughly 18 billion cubic feet per day should not be treated as directly interchangeable with PwC's 7.6 to 11.5 billion range.

The first figure represents projected total gas consumption associated with US data centers. The second describes incremental AI-linked demand across three scenarios.

Plant efficiency also matters. A modern combined-cycle facility extracts more electricity from the same fuel than a simple-cycle turbine.

Combined-cycle plants reuse exhaust heat to generate additional power. They are generally more efficient but can require more time, equipment, and capital to construct.

Data center utilization adds another uncertainty. A planned campus can have a high maximum capacity without consuming that amount during every hour.

AI workloads may also move between regions. Cloud operators can direct flexible computing tasks toward locations with available electricity, although latency and network limits restrict that strategy.

Policy could shift the outcome as well. Regulators may require developers to fund grid upgrades, secure cleaner power, disclose energy use, or limit emissions.

Conversely, faster permitting for gas pipelines and power plants could make the higher-demand path easier to reach. Public policy will influence which bottlenecks disappear first.

The international comparison also needs context. Saying US data centers could consume more gas than Germany and Japan combined communicates scale, but it does not compare identical activities.

National gas demand includes homes, factories, commercial buildings, and electricity generation. Data center demand refers mainly to fuel used for electricity.

Climate conditions, industrial structures, energy efficiency, and fuel mixes differ across countries. The comparison should not imply that a server campus functions like an entire national economy.

It does reveal something important. A single category of digital infrastructure is moving toward the gas consumption associated with two major industrial nations.

The speed of the forecast revision deserves equal attention. A near doubling within nine months indicates that analysts are still struggling to map announcements onto realistic electricity demand.

That uncertainty cuts in both directions. The eventual number could fall far below 18 billion cubic feet per day. It could also remain high if efficiency gains lead companies to deploy even more computing.

The appropriate response is not to dismiss the forecast or present it as destiny. Utilities and regulators should test projects against transparent assumptions and multiple demand scenarios.

Companies should disclose expected electricity use, supply arrangements, project timing, and emissions. Without comparable disclosures, communities cannot distinguish committed demand from speculative capacity.

Three Signals Will Show Whether the Gas Forecast Is Becoming Reality

The decisive evidence will come from completed projects, fuel infrastructure, and enforceable power contracts rather than corporate announcements.

The first signal is the conversion rate for announced data centers. Investors and regulators should track how many projects receive financing, construction permits, utility agreements, and operating equipment.

A high completion rate would strengthen the forecast. Repeated cancellations, downsizing, or extended delays would weaken it.

This measure matters because the projected gas demand begins with a very large pipeline of proposed computing capacity. Every later assumption depends on how much of that pipeline becomes operational.

The second signal is the approval of gas delivery and generation infrastructure. New pipelines, compressor capacity, turbines, and storage contracts would show that developers are preparing for sustained consumption.

On-site plants deserve attention, but grid infrastructure matters more to the total. BloombergNEF attributes most projected growth to data centers that continue drawing power through utility systems.

Regulatory filings can reveal the direction before plants begin operating. Utilities must describe forecast loads, proposed generators, transmission upgrades, and customer payment arrangements.

If regulators approve large amounts of gas capacity tied to data center demand, the 18 billion cubic feet estimate becomes more plausible. Rejections or cleaner substitutions would weaken it.

The third signal is the gap between electricity demand and new carbon-free supply. Nuclear, geothermal, renewables, storage, and transmission must arrive fast enough to displace gas, not merely supplement it.

Announcements alone will not close that gap. Observers should track commercial operation dates, hourly energy delivery, and verified output.

This signal also tests Big Tech's climate strategy. A company can support future clean projects while using gas for several years. The emissions impact depends on how quickly the cleaner resources replace fossil generation.

The next few utility planning cycles will provide clearer evidence than another wave of campus announcements. They will show which projects have firm commitments and who must pay for them.

For developers, the question is no longer simply where land and fiber are available. A viable location now needs deliverable electricity, credible fuel supply, and community approval.

For enterprise AI buyers, infrastructure choices can influence cloud availability, regional capacity, and the emissions associated with workloads. Those effects may eventually appear in procurement requirements and sustainability reports.

For households and businesses, the central question is whether regulators can isolate the costs created by unusually large customers. Protective contracts will matter if fuel and network expenses rise.

US data center gas demand will not reach 18 billion cubic feet per day through one decision. It will emerge through hundreds of permits, contracts, utility forecasts, and construction milestones.

Watch which projects secure real power, which clean resources arrive on schedule, and which costs enter customer rates. Those signals will determine whether this forecast becomes reality or another peak in the AI infrastructure hype cycle.