Data Centers Are the New Anchor of Gas Demand, IGU's Ydreos Says

NextFin News - Data centers have emerged as a new driver of global gas demand, International Gas Union Secretary General Mel Ydreos said in an October 26 interview, as the industry body argues that natural gas is not a transitional bridge but a structural, long-term anchor of AI-era power systems. The claim rests on a stark timing mismatch: a hyperscale data center can be energized in months, while the generation and transmission projects needed to power it routinely take years to permit and build.

The International Gas Union laid out its case in a January 2026 report, "The Role of Gas in Powering AI-Driven Energy Demand," and Ydreos returned to the theme on Bloomberg's "The China Show" this week. The numbers behind the argument are large enough to move a commodity market that already serves a record 4,202 billion cubic metres of annual global demand.

Data centre electricity consumption is projected to double to 800-1,000 TWh by 2030, according to the IGU report. The International Energy Agency's base case is even higher: roughly 945 TWh by 2030, growing at about 15% a year — more than four times faster than total electricity consumption across all other sectors. By 2030, data centres would account for just under 3% of global electricity use, up from about 1.5% in 2024.

The IGU's central claim is that gas-fired generation serving data centres could nearly double by 2035, equivalent to roughly 60 billion cubic metres of incremental gas supply. In the report's base trajectory, gas-fired data-centre demand expands by about 30 bcm by 2030. For scale, that 2030 increment alone is roughly half of Italy's total annual gas imports, and the 2035 figure is close to Italy's entire annual gas import volume.

The mechanism the IGU points to is not that AI is simply "energy hungry." It is that AI load is flat, always-on, and unforgiving — a profile that renewables alone cannot serve. Renewables are expected to provide about half of data centres' electricity by 2030, but their inherent variability creates a mismatch with the 24/7 load profile, which requires not just more generation but significantly more dispatchable capacity. Gas, with modern combined-cycle turbines running at 50-55% efficiency, fast ramping, and far lower emissions than coal, is positioned to supply the bulk of that flexible, firm capacity.

Ydreos framed the issue as a race between two construction clocks.

Data centres can be built and energised much more rapidly than new generation and transmission projects which often face permitting, approval and construction constraints. With speed-to-power as the binding constraint, Gas will play an indispensable role in responding to this challenge, providing rapid ramping, firm capacity, and system stability.

The market has not fully priced this. Here is why the timing mismatch matters more than the fuel mix.

The Binding Constraint Is Not Fuel — It Is Time

The first-order reading of the data-center story is a fuel-share argument: gas versus renewables versus nuclear. That framing misses the actual bottleneck. A data center's economics are governed by time-to-revenue. A facility that cannot be powered cannot bill. Hyperscalers have therefore shifted from signing power purchase agreements for intermittent renewables toward firm, on-site, or co-located generation that can be connected on the same timeline as the compute build-out.

This is why the IGU's argument is mechanism-driven rather than ideology-driven. Gas turbines can be ordered, delivered, and commissioned inside the window that a grid interconnection queue cannot match. In the United States, the median wait time for a project to reach commercial operation is approaching five years, with some data-center connection requests facing delays of up to 12 years, while a large data center campus can be built in 18 to 30 months. The mismatch is not a market failure that prices will quickly fix; it is a permitting and infrastructure reality.

The evidence is already visible in procurement behavior. Natural gas supplied over 40% of electricity for U.S. data centers as of 2024, according to the IEA, with renewables at roughly 24%, nuclear around 20%, and coal about 15%. Data centers accounted for about 4% of total U.S. electricity use in 2024. Gas turbine new orders were expected to reach 1,025 units in 2025, a 27% year-over-year increase, driven by data center electricity demand, according to an industry report on data-center energy. These are not projections — they are orders already placed.

The second-order implication is what the market has not fully absorbed. The conventional wisdom prices data-center demand as a renewables-and-storage story, with gas as a declining transitional hedge. The IGU's timing argument inverts that: if speed-to-power is the binding constraint, gas is not the hedge — it is the default. Every month a transmission line is delayed is a month of gas-fired generation that becomes economically rational regardless of its position in a 2050 decarbonization pathway.

The market is pricing the fuel mix. The constraint is the construction schedule.

Cyclical Demand or Structural Regime Shift?

This is the call that determines the whole thesis. Is data-center gas demand a cyclical wave that will crest as AI capex normalizes, or a structural regime shift that will not revert on its own?

The evidence points to structural, for three reasons.

First, the demand base is compounding, not episodic. The IEA projects data center electricity use growing at roughly 15% a year through 2030 — more than four times the growth of all other electricity demand. The United States alone is expected to add about 240 TWh by 2030, a 130% increase over 2024; China adds roughly 175 TWh, up 170%. Together the two account for nearly 80% of global growth. This is not a one-off capacity cycle; it is a new load class being added to grids that were not designed for it.

Second, the load profile is fundamentally different from historical industrial demand. Traditional industrial load could be shifted, curtailed, or time-of-use managed. AI inference load is continuous and latency-sensitive. A data center cannot "wait for the wind to blow." That flat, 24/7 profile structurally favors dispatchable capacity — gas, nuclear, geothermal, or storage at a scale that does not yet exist commercially. Renewables can provide the energy; they cannot, without an order-of-magnitude expansion of storage, provide the firm capacity profile.

Third, the infrastructure lock-in is durable. A gas-fired generation asset built to serve a data center campus has a 20- to 30-year operating life. Once that turbine is installed and the fuel pipeline is connected, the load does not revert to the grid when the next AI model ships. The capital is sunk, the offtake contract is signed, and the operating economics favor continued use.

The cyclical counter-argument is real and must be taken seriously. AI capex has been called into question by investors who see spending outrunning monetization. If hyperscalers slow their build-out — or if chip efficiency improves faster than workload growth — the demand curve flattens. Efficiency gains kept data center power flat for much of the past decade despite rising workloads; that trend reversed only after 2020, but it could return if a technological breakthrough decouples compute growth from electricity draw.

The structural call survives this, but only partially. The base rate of growth could moderate — that is the cyclical leg — while the underlying regime shift remains intact. Even at half the IGU's projected growth, data centers add a load class larger than any new industrial sector in a generation, and the timing mismatch that favors gas does not disappear with slower growth; it only stretches out. The correct read is a structural shift with a cyclical growth rate riding on top of it.

The Second-Order Trade: Gas Is Repricing as a Capacity Asset, Not a Commodity

The first-order effect of the IGU's argument is straightforward: more gas demand. The second-order effect is more important and less priced. Gas is being revalued from a commodity whose price is set by weather and LNG arbitrage into a capacity asset whose value is set by reliability.

Under the old model, Henry Hub prices were set by winter heating demand, shoulder-season maintenance, and LNG export volumes. Prompt-month Henry Hub settled near $3.02 per MMBtu in late September 2026, and the forward curve expected a gradual rise toward roughly $5.40 by December 2027, per the U.S. Energy Information Administration's latest short-term outlook. That curve is built on weather-normalized residential and industrial demand plus LNG feed gas, which is expected to climb above 20 billion cubic feet per day as new liquefaction capacity comes online.

Data center demand changes the shape of that curve, not just its level. Unlike heating demand, which peaks in winter and collapses in summer, data center load is flat across the year. That flattens the seasonal price spread and puts a floor under summer prices — precisely the shoulder season when gas prices historically break down. It also localizes demand: a data center cluster in Texas, Virginia, or the Southeast bids for gas and power in a specific node, tightening local basis spreads even when the national hub is well supplied.

This is where the equity market has already moved ahead of the commodity. Power producers with data-center exposure have re-rated sharply as forward power prices in data-center hubs began reflecting tighter grid conditions. Around-the-clock forward power prices for 2026 in northern Texas had risen roughly 13% since the prior November, according to Vistra's earnings-call commentary. The gas producers and midstream companies that serve those same nodes have lagged that re-rating. That gap — power assets priced for scarcity, gas assets priced for a weather cycle — is the second-order opportunity the IGU's thesis points toward.

Adversarial check: The strongest case against this trade is that gas is on the wrong side of climate policy. Carbon pricing, methane regulation, and corporate net-zero commitments could strand gas-fired data-center assets before they earn back their capital. The IGU itself acknowledges the industry's "continuing need to reduce emissions," and the 2026 Global Gas Report frames decarbonization as a simultaneous pressure alongside demand growth.

The answer is that policy risk is real but slow-moving relative to the construction cycle. A data center operator facing a 2027 or 2028 power deadline is not choosing between gas and a zero-carbon future; it is choosing between gas and no power. Policy moves on legislative and permitting timelines measured in years; a data center's revenue clock starts on the day it is energized. That does not make gas immune to regulation — it makes gas the near-term default while the policy environment catches up. The risk is not that gas disappears; it is that the premium for firm capacity gets capped by regulation before the assets fully monetize.

Falsifying signal: The structural thesis is wrong if data center electricity consumption growth slows to below 8% per year for two consecutive years while grid interconnection queues clear materially — meaning the timing mismatch that favors gas has resolved. Watch the IEA's annual data-center demand update and the U.S. interconnection queue data from the grid operators. If queues compress and growth decelerates together, the gas-default thesis breaks.

Conclusion: Who Benefits, Who Is Exposed, and What to Watch

The mechanism cashes out into a clear asymmetry. The beneficiaries are not only the obvious ones. Yes, gas producers with exposure to data-center-heavy basins and the midstream pipelines that serve them stand to gain. But the bigger re-rating belongs to firms that sell firm capacity — gas-fired generators, turbine manufacturers, and the utilities that can actually connect load before the customer arrives. The exposed are the merchants who sold power on the assumption that renewables-plus-storage could serve 24/7 load at scale within this decade, and the gas bears who underwrote demand on weather-normalized residential and industrial curves that no longer describe the market.

Split by time horizon:

• Short term (6-18 months): sentiment and liquidity dominate. Gas prices remain anchored near the $3 per MMBtu area with a weather-driven trading range, and the data-center premium is visible mainly in power prices and selected equities rather than the commodity itself. Volatility is the trade, not direction.
• Medium term (2-5 years): fundamentals take over. As LNG feed gas demand climbs above 20 Bcf/d and data-center load connects, the seasonal price floor rises and local basis spreads tighten in data-center corridors. This is where the gas complex re-rates toward the capacity-asset model.
• Long term (5-10 years): structure decides. If data-center demand compounds at 10-15% annually as projected, gas-fired firm capacity becomes a permanent anchor of the power system, and the "transitional fuel" narrative is replaced by a "firm capacity anchor" narrative. If efficiency breakthroughs or policy caps intervene, gas remains a large but capped market.

Scenarios:

• Base case: data-center demand grows at 12-15% annually; gas-fired data-center generation expands by roughly 30 bcm by 2030 and approaches 60 bcm by 2035, as the IGU projects. Gas prices grind higher on a flattening seasonal curve; firm-capacity assets outperform.
• Upside case: AI monetization accelerates, interconnection queues remain backlogged, and hyperscalers default to on-site gas at scale. The 60 bcm figure is exceeded; gas reprices as a reliability asset and the power-gas basis gap widens.
• Downside case: AI capex slows sharply, chip efficiency decouples compute from power, or carbon policy imposes a binding cost on gas-fired generation before 2030. Gas demand from data centers falls well short of the IGU's trajectory, and the commodity reverts to a weather-driven range.

What to watch: the IEA's annual data-center electricity demand update; U.S. grid-operator interconnection queue reports; monthly Henry Hub and LNG feed gas data from the EIA; and turbine order books from the major manufacturers. The single falsifying signal is sustained sub-8% data-center demand growth combined with materially clearing interconnection queues.

The market is still asking which fuel will win the AI power race. The better question is which fuel can be connected before the data center opens — and on that test, gas is not competing on cost. It is competing on the clock.

Explore more exclusive insights at nextfin.ai.