Psst: NatGas Futures Haven鈥檛 Priced in AI Power Demand Yet

Gas-fired power demand is coming for AI-enhanced data generation as Microsoft, Amazon and others race to stay on top鈥攁nd not go the way of IBM鈥攁nalysts said at a Pittsburgh energy forum.

Nissa
Darbonne
Oil and Gas Investor

Gas futures prices don鈥檛 yet reflect growth in AI demand for power, according to market analysts.

鈥淚s it factored in the natural gas pricing in the back end of the curve? I don't think so,鈥� said Dennis Kissler, senior vice president of trading for BOK Financial, in a recent Pittsburgh energy forum.

鈥淢aybe a very small percentage.鈥�

January 2026 gas was $4.07/MMBtu on Oct. 22 at the CME Group exchange. The January 2030 price was $4.29/MMBtu.

The market doesn鈥檛 fully understand the phenomenon yet, Kissler said.

Data centers currently pull some 3.5% of power off the U.S. grid. 鈥淚f that goes to 8% or 9%, it's exponential,鈥� Kissler said.

鈥淎nd that's what people don't understand. They say, 鈥榃ell, if it's 3.5% and it goes to 8%, that's not such a big deal.鈥�

But 鈥渘o, 鈥� that is huge,鈥� Kissler said in the annual BOK Financial-hosted program Northeast Energy Summit.

The Federal Energy Regulatory Commission calculated 2023 data-center power demand was some 19 gigawatts (GW). It expects this to grow to 35 GW by 2030.

Meanwhile, S&P Global Research estimates 61 GW as soon as 2028.

Wind and solar can supplement power supply. But data centers need reliable, 24/7 electrons, which are expected to come from gas-fired power generation and, at some point, newbuild nuclear.

Meanwhile, more natural gas for powergen could be needed in the coming years for another reason: Anticipated new EPA rules will result in about one-third fewer net electrons put onto the grid, a Vistra Corp. power company executive said at an energy forum at the University of Texas last fall.

This will be due to Btu loss from powering mandated, adjacent carbon capture and storage infrastructure.

鈥楬uge number鈥�

The futures market may still be trying to fully comprehend the magnitude before placing money on gas demand growth鈥攁nd whether it will outstrip supply growth鈥攃oncurred Matt Stephani, president of Cavanal Hill Investment Management.

鈥淚 don't think that there's a tight enough range of estimates as to how much incremental demand data centers would provide to natural gas,鈥� Stephani said.

The market 鈥渉as to see those [natgas-fueled] plants built and running and gas being pulled there鈥� before wagering on higher futures.

For perspective, projections for growth in power demand for AI data centers would result in the equivalent of up to 70 roughly 1 GW nuclear reactors, he expects.

鈥淭hat's a huge number.鈥�

To generate the equivalent power from natural gas, 鈥渢hink of a 3:1. You need [up] to 200 natural gas turbines. That is a massive investment.鈥�

Microsoft has signed a deal that will result in the restart of the Three Mile Island nuclear plant鈥檚 Unit 1 that was capable of producing some 837 megawatts before it was turned off in 2019.

Stephani counts 18 other reactors that aren鈥檛 yet fully decommissioned that could be brought back online. 鈥淭here are very few of them,鈥� he said.

Toby Rice, EQT Corp.鈥檚 president and CEO, told Hart Energy recently after the Three Mile Island news that the current estimate for new U.S. power demand for data centers is 75 GW.

The Three Mile Island restart is 鈥渘ot a needle-mover,鈥� he added. And 鈥渢here鈥檚 only 3 GW of nuclear potential if we restart all facilities that could be restarted,鈥� he added.

鈥楶erfect place鈥�

Currently, Appalachian gas output is capped at about 35 Bcf/d due to limited pipe to Gulf Coast and other U.S. demand centers.

Could Appalachian Basin gas producers directly participate in power-generation buildout in the U.S. Northeast where the world鈥檚 largest data center concentration is located in nearby Virginia?

鈥淚t's a very good possibility,鈥� Kissler said. 鈥淎nd it is a kind of a [way] out if they [the federal government] continue the politics where you can't increase the pipeline infrastructure.鈥�

He added, 鈥淚 would think EQT could lead something like that.鈥�

Forum attendees鈥擜ppalachian gas producers鈥攚ere mum about their plans, though, with none raising their hand in a quick poll of whether they were working directly with data center developers.

Stephani said, 鈥淚 think [while] no one raised their hands in the room, they are considering it.鈥�

He told them, 鈥淵ou should be considering that.鈥�

Meanwhile, 鈥淚 can't believe that with all the negotiation taking place for power right now [by] Microsoft and Amazon and [others], that someone isn't considering this.鈥�

The ideal locations for a data center are where there is trunkline broadband, a mild climate, low humidity, land and plentiful water for cooling.

Northern Virginia is a popular spot due to trunkline broadband and temperatures milder than in the western U.S. desert, which carries low humidity.

鈥淪omewhere between Philadelphia and D.C. is a perfect place to put a data center just based on access to trunkline broadband,鈥� Stephani said.

鈥楶henomenal鈥�

One typical Google search, such as 鈥淢cDonald鈥檚 nearby鈥� uses about 0.33 watt of power, Stephani said.

鈥淎I is going to change that. If using an AI-enhanced Google app search, it is 25 times more intense.

鈥淪o the power demand from AI searching sounds like this little thing, but understand there are 8.5 billion global searches a day. So the demand coming from the search side is going to be phenomenal.鈥�

Yet, 鈥渢hat's the smaller piece,鈥� he added.

鈥淭he bigger piece is the demand for training the large language models, which are what create the AI.鈥�

A consumer could ask Expedia to plan a 10-day trip to Italy. 鈥淭hat's a language model focusing on travel built for Expedia.

鈥淭he training of that consumes an incredible amount of power. And that's where we are: We're at the start of the training.鈥�

Win or 鈥榖ecome IBM鈥�

He does expect some new nuclear will come online but if and when that happens, 鈥測ou're six to eight years away from those nuclear plants producing power,鈥� Stephani said.

鈥淪o in the interim, gas has to step up.鈥�

And tech companies have to step up as well or go the way of companies that lost to them in the tech race with Microsoft, Samsung, Google and others in the past 30 years.

鈥淲inners and losers are going to be decided,鈥� Stephani said.

鈥淚f Microsoft is saying, 鈥楬ey, we're looking at this $100 billion data center and we need 5 GW of power,鈥� these other guys have to be looking at the same thing.

鈥淭hey have no choice鈥�. If you lose your edge, you lose your business.鈥�

Thus 鈥渢he power demand is real. The necessity to win the AI future is the most critical decision Microsoft, Apple, Amazon and [others] face: Either they [win] or they're going to become IBM and they know that.鈥�

Comments
Comments

Add new comment

This conversation is moderated according to Hart Energy community rules. Please read the rules before joining the discussion. If you鈥檙e experiencing any technical problems, please contact our customer care team.

About text formats
原文链接/HartEnergy

Psst: NatGas Futures Haven’t Priced in AI Power Demand Yet

Gas-fired power demand is coming for AI-enhanced data generation as Microsoft, Amazon and others race to stay on top—and not go the way of IBM—analysts said at a Pittsburgh energy forum.

Nissa
Darbonne
Oil and Gas Investor

Gas futures prices don’t yet reflect growth in AI demand for power, according to market analysts.

“Is it factored in the natural gas pricing in the back end of the curve? I don't think so,” said Dennis Kissler, senior vice president of trading for BOK Financial, in a recent Pittsburgh energy forum.

“Maybe a very small percentage.”

January 2026 gas was $4.07/MMBtu on Oct. 22 at the CME Group exchange. The January 2030 price was $4.29/MMBtu.

The market doesn’t fully understand the phenomenon yet, Kissler said.

Data centers currently pull some 3.5% of power off the U.S. grid. “If that goes to 8% or 9%, it's exponential,” Kissler said.

“And that's what people don't understand. They say, ‘Well, if it's 3.5% and it goes to 8%, that's not such a big deal.’

But “no, … that is huge,” Kissler said in the annual BOK Financial-hosted program Northeast Energy Summit.

The Federal Energy Regulatory Commission calculated 2023 data-center power demand was some 19 gigawatts (GW). It expects this to grow to 35 GW by 2030.

Meanwhile, S&P Global Research estimates 61 GW as soon as 2028.

Wind and solar can supplement power supply. But data centers need reliable, 24/7 electrons, which are expected to come from gas-fired power generation and, at some point, newbuild nuclear.

Meanwhile, more natural gas for powergen could be needed in the coming years for another reason: Anticipated new EPA rules will result in about one-third fewer net electrons put onto the grid, a Vistra Corp. power company executive said at an energy forum at the University of Texas last fall.

This will be due to Btu loss from powering mandated, adjacent carbon capture and storage infrastructure.

‘Huge number’

The futures market may still be trying to fully comprehend the magnitude before placing money on gas demand growth—and whether it will outstrip supply growth—concurred Matt Stephani, president of Cavanal Hill Investment Management.

“I don't think that there's a tight enough range of estimates as to how much incremental demand data centers would provide to natural gas,” Stephani said.

The market “has to see those [natgas-fueled] plants built and running and gas being pulled there” before wagering on higher futures.

For perspective, projections for growth in power demand for AI data centers would result in the equivalent of up to 70 roughly 1 GW nuclear reactors, he expects.

“That's a huge number.”

To generate the equivalent power from natural gas, “think of a 3:1. You need [up] to 200 natural gas turbines. That is a massive investment.”

Microsoft has signed a deal that will result in the restart of the Three Mile Island nuclear plant’s Unit 1 that was capable of producing some 837 megawatts before it was turned off in 2019.

Stephani counts 18 other reactors that aren’t yet fully decommissioned that could be brought back online. “There are very few of them,” he said.

Toby Rice, EQT Corp.’s president and CEO, told Hart Energy recently after the Three Mile Island news that the current estimate for new U.S. power demand for data centers is 75 GW.

The Three Mile Island restart is “not a needle-mover,” he added. And “there’s only 3 GW of nuclear potential if we restart all facilities that could be restarted,” he added.

‘Perfect place’

Currently, Appalachian gas output is capped at about 35 Bcf/d due to limited pipe to Gulf Coast and other U.S. demand centers.

Could Appalachian Basin gas producers directly participate in power-generation buildout in the U.S. Northeast where the world’s largest data center concentration is located in nearby Virginia?

“It's a very good possibility,” Kissler said. “And it is a kind of a [way] out if they [the federal government] continue the politics where you can't increase the pipeline infrastructure.”

He added, “I would think EQT could lead something like that.”

Forum attendees—Appalachian gas producers—were mum about their plans, though, with none raising their hand in a quick poll of whether they were working directly with data center developers.

Stephani said, “I think [while] no one raised their hands in the room, they are considering it.”

He told them, “You should be considering that.”

Meanwhile, “I can't believe that with all the negotiation taking place for power right now [by] Microsoft and Amazon and [others], that someone isn't considering this.”

The ideal locations for a data center are where there is trunkline broadband, a mild climate, low humidity, land and plentiful water for cooling.

Northern Virginia is a popular spot due to trunkline broadband and temperatures milder than in the western U.S. desert, which carries low humidity.

“Somewhere between Philadelphia and D.C. is a perfect place to put a data center just based on access to trunkline broadband,” Stephani said.

‘Phenomenal’

One typical Google search, such as “McDonald’s nearby” uses about 0.33 watt of power, Stephani said.

“AI is going to change that. If using an AI-enhanced Google app search, it is 25 times more intense.

“So the power demand from AI searching sounds like this little thing, but understand there are 8.5 billion global searches a day. So the demand coming from the search side is going to be phenomenal.”

Yet, “that's the smaller piece,” he added.

“The bigger piece is the demand for training the large language models, which are what create the AI.”

A consumer could ask Expedia to plan a 10-day trip to Italy. “That's a language model focusing on travel built for Expedia.

“The training of that consumes an incredible amount of power. And that's where we are: We're at the start of the training.”

Win or ‘become IBM”

He does expect some new nuclear will come online but if and when that happens, “you're six to eight years away from those nuclear plants producing power,” Stephani said.

“So in the interim, gas has to step up.”

And tech companies have to step up as well or go the way of companies that lost to them in the tech race with Microsoft, Samsung, Google and others in the past 30 years.

“Winners and losers are going to be decided,” Stephani said.

“If Microsoft is saying, ‘Hey, we're looking at this $100 billion data center and we need 5 GW of power,’ these other guys have to be looking at the same thing.

“They have no choice…. If you lose your edge, you lose your business.”

Thus “the power demand is real. The necessity to win the AI future is the most critical decision Microsoft, Apple, Amazon and [others] face: Either they [win] or they're going to become IBM and they know that.”

Comments
Comments

Add new comment

This conversation is moderated according to Hart Energy community rules. Please read the rules before joining the discussion. If you’re experiencing any technical problems, please contact our customer care team.

About text formats