人工智能来了——无论好坏——改变能源格局

去年,人们开始讨论人工智能对天然气需求的影响。今年,它终于开始显现。


技术人员称人工智能的部署是革命性的。

工作变得更轻松。医疗保健变得更有效。交通变得更快捷。一切都在改变,而且一切都应该从现在开始改变。

特别是对于生产和运输天然气的人们来说。

据CNBC今年2月报道,科技巨头谷歌、亚马逊、微软和谷歌母公司Alphabet计划在2025年向人工智能领域投资高达3200亿美元。去年,这一数字接近1000亿美元。

亚马逊首席执行官安迪·贾西 (Andy Jassy) 在二月份的公司财报电话会议上表示:“我认为,从中长期来看,我们的企业、客户和股东都会对我们正在追求人工智能领域的资本机会和商业机会感到高兴。”

亚马逊计划今年在人工智能领域投资约 1000 亿美元,贾西称之为“千载难逢的商业机会”。

有人持怀疑态度。那些承诺的改变生活的进步至今尚未到来,甚至尚未被定义,而且制作猫咪装扮图片到底能赚多少钱?

但随着今年的开始,科技公司对芯片、仓库以及(对能源行业来说最重要的)天然气获取渠道的竞争愈演愈烈。

由于供应过剩,价格接近2美元/百万英热单位,天然气业务在2024年前三个季度举步维艰。然而,随着时间的推移,中游公司高管开始讨论电力供应新市场的诞生

人工智能服务器中的计算机芯片能耗是早期芯片的10倍。它们在服务器机箱内封装得更密集,需要更强大的冷却能力。随着电力需求的增加,以及许多燃煤发电厂的淘汰,迄今为止显而易见的解决方案是增加天然气发电。

2024 年全年供应协议谈判都处于保密状态,但现在中游、天然气生产商和科技公司之间的交易正在公开。

今年2月,Energy Transfer宣布了一项协议,将向德克萨斯州中部的一个数据中心园区供应4.5亿立方英尺/天的天然气,这是同类交易中首批公开交易之一。金德摩根和南方能源宣布建立合作伙伴关系,为东南部的几个正在建设中的数据中心提供天然气。

雪佛龙提出合作建设天然气发电厂,为数据中心园区提供现场发电。

Meta首席执行官马克·扎克伯格告诉投资者,2025 年将是“人工智能的决定性一年”。他的公司已拨出 600 亿美元的资本支出用于人工智能开发。

关注能源行业的播客节目《查克·耶茨需要工作》主持人查克·耶茨表示,天然气行业已经被重新定义。

“告诉所有人,你们不是从事天然气行业的,”耶茨在 NAPE 2025 的一次会议上告诉观众。

“你不卖天然气”。你卖电子产品,”他说。 “所以弄清楚吧,因为如果你今天看看我们的电网,它刚刚被淘汰。”

增长和其他选择

国际能源署(IEA)在 2024 年预测,到 2026 年,全球人工智能所需的能源将相当于日本的用电量。OpenAI 的大型语言模型 GPT-3 的训练需要近 1300 兆瓦时的电力,相当于约 130 个美国家庭一年的用电量。

国际能源署表示,美国五个州的数据中心已经消耗了该州 10% 或更多的电力。

由于建设数据中心的公司不断变化计划和预测,对未来几年电力需求的预测也有所不同。

去年,摩根士丹利估计,从2024年到2025年,全球数据中心的用电量将增长一倍以上,到2026年将增长两倍以上,达到约160太瓦时(TWh)。富国银行预测,人工智能电力需求将增长550%,从2024年的8太瓦时增至2026年的52太瓦时,到2030年将再增长1150%,达到652太瓦时。在全球范围内,这意味着人工智能电力需求将比2024年的水平增长8050%。

在美国,国际能源署预测2030年人工智能将达到606TWh至652TWh,占该国电力需求的11.7%至16%。

高盛的一份报告预测,天然气将满足人工智能和数据中心电力需求增长的 60%,其余 40% 将由风能和太阳能提供。

尽管科技公司普遍对可再生能源持积极态度,但天然气更有可能胜出。

美国需要在未来十年内大规模扩建其电网,以满足未来的需求预测,这不仅是针对人工智能,而且对于一个正在转向电气化并正在经历“外包”崛起的国家来说也是如此,许多公司正在将工业生产从海外转移到美国。

与风能和太阳能不同,美国已经拥有庞大的天然气输送系统,其供应量巨大,增长速度远超阵列和风力发电场。更重要的是,天然气可以作为稳定的基础负荷电力,这对于需要持续运行的数据中心至关重要。

对于科技公司和大多数公用事业公司来说,增加煤炭供应是不可能的。企业不愿开发更多能源,因为这会与遏制温室气体排放的努力相冲突。目前电网的需求很可能会耗尽所有可用的燃煤发电量。

2022 年和 2023 年,美国公用事业公司淘汰了 22.3 吉瓦 (GW) 的燃煤发电能力,但这一趋势在 2024 年有所放缓。根据美国能源信息署的数据,电力公司仅淘汰了年初运营的美国燃煤发电能力的 1.3%

预计2025年煤炭退役规模将再次增加,运营商预计将退役10.9吉瓦的煤炭发电量。然而,整体电力需求可能会压倒生产商的减排意愿。

一位分析师在 2024 年底的标准普尔商品洞察报告中表示,由于美国能源需求的“爆炸式增长”,许多现有燃煤电厂的运行时间很有可能超过预期。

核能被视为天然气更清洁、更可靠的替代品,但它面临着监管和成本问题,并且在未来几年内无法成为能源供应商领域的主导者。

格鲁吉亚沃格特尔 3 号和 4 号机组是该国最后一批核电站,建设历时 15 年,耗资 300 亿美元,是最初预计时间和成本的两倍。

小型模块化反应堆 (SMR) 被视为一种替代方案,但这些模型大部分仍处于设计阶段,并且面临着与沃格特勒反应堆相同的监管障碍,至少目前如此。

长期以来支持核电的能源部长克里斯·赖特在其任命听证会上表示,小型模块化反应堆(SMR)大约需要10年时间才能扫清进入市场所需的开发和监管障碍。赖特乐观地认为,这项任务有望在6年内完成。

科技公司不愿意等那么久。微软出资重启了宾夕法尼亚州三哩岛核电站的一个机组。预计该机组的发电成本将是燃气发电成本的10倍。其他公司面临的问题是,美国各地可供重启的核电站并不多。

关注增长

星际之门公告
耗资5000亿美元的“星际之门”项目将在德克萨斯州阿比林打造全球最大的高能耗人工智能集群之一。(来源:星际之门项目)

要了解天然气如何推动人工智能革命,请注意数据中心园区的出现地点。

美国最活跃的数据中心开发地点之一已在德克萨斯州科技中心阿比林崛起。

特朗普政府在就职后不久就启动了耗资 5000 亿美元的星际之门计划,首个园区选在了位于二叠纪盆地和沃斯堡之间的西德克萨斯州偏远小镇,该镇人口 13 万。

负责跟踪East Daley Analytics数据中心发展的 Ethan Warrick 表示:“Targate 的阿比林数据中心项目预计将从 1.2 GW 扩展到 5 GW,成为世界上最大的电力密集型 AI 集群之一。”

星际之门是第二个落户阿比林的人工智能项目。2022年,总部位于休斯顿的Lancium公司破土动工,该项目最初计划用于比特币挖矿,并由当地可再生能源提供动力。(阿比林附近有大量风力涡轮机。)

到2023年,该公司宣布与Crusoe Energy建立合作伙伴关系,并从专注于比特币挖矿转向建设数据处理园区。虽然两家公司都强调该地区可用的可再生能源,但了解中游网络的人很难不注意到穿过该地区的众多大直径管道,这些管道将二叠纪盆地的天然气输送到达拉斯/沃斯堡都市区附近的网络连接点。

能源转移公司(ET)的休·布林森(Hugh Brinson)管道是下一条从二叠纪盆地输送天然气的主要管道之一,它将沿着一条经过阿比林附近的路线运行。ET 于 12 月宣布了该项目。

沃里克在东戴利的一份报告中写道:“我们推测新的15亿立方英尺/天的项目将有助于满足这个新兴的需求中心。”

根据其功能,人工智能数据中心可能会在宽带接入充足、土地充足以及至关重要的是靠近稳定天然气供应的地区发展。数据中心主要建在弗吉尼亚州、德克萨斯州和加利福尼亚州硅谷的大都市附近。现在,由于人工智能编程的不同需求,新的设施越来越多地涌现到偏远地区。

麦肯锡 10 月份发布的一份分析报告指出:“在训练人工智能模型时,低延迟和网络冗余等典型的性能因素就不那么重要了。因此,专门用于训练人工智能模型的数据中心正在美国更偏远的地方建设,例如印第安纳州、爱荷华州和怀俄明州,这些地方电力仍然充足,电网压力较小。”

今年 2 月,Meta 宣布正在洽谈建设一个价值 2000 亿美元的人工智能项目,并在德克萨斯州、路易斯安那州和怀俄明州的农村地区设立数据中心。

对速度的需求

中游、天然气和公用事业公司一直在努力,有时合作,有时竞争,以确保他们在即将到来的数据中心市场中的地位。

随着更多交易的宣布,金德摩根已与南方公司合作,以推动美国东南部天然气市场的发展。

2024年,合作伙伴批准了南方天然气公司南线系统扩建4号项目,旨在将南方天然气公司的南线输气能力提高12亿立方英尺/天。这项耗资30亿美元的项目计划于2028年底投入使用,从而增加整个东南部地区的天然气供应。该项目将服务于一个总体增长的地区,尽管公司高管表示,很大一部分需求来自人工智能(AI)。

科技公司面临两种选择:要么走公用事业公司常见的繁琐程序,搞清楚如何管理互连流程;要么直接联系已经拥有电力供应的供应商。许多高管认为,数据中心建设者会主动联系他们。

今年2月,Energy Transfer宣布了一项可能成为科技和天然气公司之间常规交易的协议:为位于德克萨斯州圣马科斯附近(奥斯汀和圣安东尼奥之间)的CloudBurst园区提供4.5亿英热单位/天的电表后天然气供应。该供应将为预计于2026年启用的园区提供约1.2吉瓦的电力。

用户侧协议或将成为常态。随着人工智能持续飞速发展,科技公司通常需要迅速采取行动,才能保持与全球其他公司的竞争力。

Williams Cos.首席执行官艾伦·阿姆斯特朗 (Alan Armstrong) 在接受 Argus Media 采访时表示:“数据中心不会等待。他们会去那些允许使用电表的州。”

总部位于匹兹堡的天然气生产商EQT近期收购了自己的中游公司,并在此基础上更进一步。在公司第四季度财报电话会议上,高管们谈到了其作为数据中心供应商和交付商的地位。

快速上市至关重要,”EQT 首席执行官托比·赖斯 (Toby Rice) 表示,“这正是EQT 脱颖而出的关键所在。我们提供一站式服务,提供市场上最好、最清洁、最可靠、最实惠的天然气。”

其他一些公司正在利用不同细分市场的机会。

雪佛龙今年1月宣布,将为数据中心建设现场燃气发电厂。该项目将使用通用电气Vernova的天然气涡轮机,为位于美国东南部、中西部和西部地区的数据中心输送高达4吉瓦的电力,足以为约300万户家庭供电。

熊市案例

尽管企业通常会寻找机会,但一些分析师认为有理由降低对快速发展的市场的预期,因为该市场的参数似乎每隔一天就会发生变化。

今年1月,中国公司DeepSeek首次推出了其人工智能应用。据称,该产品的处理能力远低于美国版本,因此所需的电量远不及用电量。天然气价格花了两周时间才从该公告带来的冲击中恢复过来。

分析公司RBN解释了预期的天然气需求繁荣可能会如何消退。

一月中旬发表的分析报告指出:“人们太快相信了这种炒作。虽然许多数据中心确实正在建设中,但天然气将面临来自风能和太阳能的激烈竞争——数据中心运营商强烈青睐这些绿色能源,而不是化石燃料。”

尽管中游企业希望开发碳捕获和储存系统,但该领域仍欠发达,项目难以获得多个州和联邦机构的许可。另一个因素是交易。中游企业通常需要10到20年的承诺来支付所需的基础设施费用。许多科技公司在达成如此期限的交易之前可能会犹豫不决。

RBN 表示,“公用事业公司愿意冒险建设昂贵的基础设施,但随着行业找到降低人工智能功耗的方法,需求却逐渐减少。”

最后,即使各方不再争夺建造发电机所需的相同设备,天然气发电项目通常也需要数年时间才能建成。科技公司的时间表可能过于乐观,与现实不符。

不过,分析师认为,天然气开发的整体前景依然乐观。印度最大的天然气生产商Expand Energy的高管在NAPE会议上表示,液化天然气出口可能会占据更大的市场份额,这将推动市场发展。

Expand Energy 首席财务官 Mohit Singh 表示:“需要明确的是,我们认为,至少 75% 的天然气需求增长将来自液化天然气。”

其他人仍然相信,无论速度如何,市场都会发展。

GPA Midstream政府事务副总裁 Stuart Saulters表示,在 DeepSeek 发布后,他曾与一家公用事业公司的代表进行过交谈。如果科技界能够用更少的材料完成同样的工作,那么科技界整体上可能会做得更多。

“他告诉我,即使他们不需要为芯片提供那么多的电力,这也意味着他们会做比现在计划的更多的事情,”索尔特斯说。

评论

添加新评论

本次对话将根据 Hart Energy 社区规则进行。请在加入讨论前阅读规则。如果您遇到任何技术问题,请联系我们的客服团队。

原文链接/HartEnergy

AI is Here—for Better or Worse—and Changing the Energy Landscape

Last year, people started talking about AI’s impact on natural gas demand. This year, it’s finally starting to take shape.


The techies say the deployment of AI is revolutionary.

Work becomes easier. Health care becomes more effective. Transportation gets faster. Everything changes, and everything is supposed to start changing now.

Especially for the people who produce and move natural gas.

The tech megacompanies—Meta, Amazon, Microsoft and Alphabet, the parent company of Google—plan to invest up to $320 billion on AI in 2025, CNBC reported in February. Last year, the amount was closer to $100 billion.

“I think that both our business, our customers and shareholders will be happy, medium to long-term, that we’re pursuing the capital opportunity and the business opportunity in AI,” Amazon CEO Andy Jassy said during the company’s earnings conference call in February.

Amazon plans to invest around $100 billion in AI over the year, for what Jassy called a “once-in-a-lifetime type of business opportunity.”

There are doubters. The promised life-changing advancements haven’t arrived or even been defined, and just how much money is there in making dress-up pictures of a cat?

But as the year started, tech companies ramped up an already furious competition for chips, warehouses and—most importantly for the energy industry—access to natural gas.

With an overabundant supply and prices near $2/MMBtu, the natural gas business struggled through the first three quarters of 2024. As the year went on, however, midstream company executives began to discuss the birthing of a new market in power supply.

The computer chips in AI servers use 10 times the energy of earlier chips. They are more densely packed in their server housing, and they require a greater amount of cooling. With a greater power requirement, and with many coal-powered generating plants undergoing retirements, the obvious solution so far has been adding power generation through natural gas.

Supply agreement negotiations were under wraps throughout 2024, but now those deals among midstream, natural gas producers and tech companies are becoming public.

In February, Energy Transfer announced a deal to supply 450 MMcf/d to a data center campus in Central Texas, one of the first public deals of its kind. Kinder Morgan and Southern Energy announced a partnership to supply several developing data centers in the Southeast.

Chevron presented a partnership to build natural gas-based power plants to provide on-site electrical generation to data center campuses.

Meta CEO Mark Zuckerburg, whose company has set aside $60 billion in capex for AI development, told investors that 2025 will be a “defining year for AI.”

The natural gas industry has already been redefined, said Chuck Yates, host of the “Chuck Yates Needs a Job” podcast, which focuses on the energy industry.

“I tell everybody out there, you’re not in the natural gas business,” Yates told the audience during a session at NAPE 2025.

“You don’t sell natural gas. You sell electrons,” he said. “Go figure it out because if you look at our grid today, it’s just tapped out.”

The growth and the other options

The International Energy Agency (IEA) predicted in 2024 that the global energy required to power AI by 2026 will equal the amount required to power Japan. OpenAI’s GPT-3, a large language model, takes nearly 1,300 megawatt-hours of electricity to train—the yearly consumption of about 130 U.S. homes.

Data centers in five U.S. states already consume 10% or more of those states’ power, IEA says.

Predictions have varied for power requirements over the next few years thanks to the continually changing plans and forecasts of the companies building the data centers.

Last year, Morgan Stanley estimated global data center power use would more than double from 2024 to 2025, and more than triple in 2026, to about 160 terrawatt hours (TWh). Wells Fargo projected AI power demand to increase 550%, from 8 TWh in 2024 to 52 TWh in 2026, before rising another 1,150% to 652 TWh by 2030. Worldwide, it would represent an 8,050% growth from 2024 levels.

In the U.S., the IEA projections for AI range from 606 TWh to 652 TWh by 2030, representing 11.7% to 16% of the country’s electricity demand.

A Goldman Sachs report forecasts natural gas will supply 60% of the power demand growth from AI and data centers, with the remaining 40% generated by wind and solar.

While tech companies generally take a favorable approach toward renewable energy sources, gas is much more likely to win out.

The U.S. will need to make massive additions to its power grid over the next decade to meet future demand projections, not just for AI, but for a country that is turning toward electrification in general and is experiencing a rise in “reshoring,” in which many companies are moving industrial production from overseas to the U.S.

Unlike wind and solar, the U.S. already has a massive delivery system for natural gas, and a large supply that can be ramped up far faster than arrays and windmill fields. More importantly, gas can be relied upon for steady baseload power, a necessity for data centers that will always be on.

For tech companies and most utilities, additional coal is a non-starter. Companies aren’t willing to develop further sources because it would conflict with efforts to curb greenhouse gas emissions. The current demands on the electrical grid are likely to take up whatever coal-generated power is available.

Utilities retired 22.3 gigawatts (GW) of U.S. coal-fired electric generating capacity in 2022 and 2023, but that trend slowed in 2024. Power companies retired only 1.3% of the U.S. coal fleet that was in operation at the beginning of the year, according to the U.S. Energy Information Administration.

Coal retirements are scheduled to increase again in 2025, with operators expected to retire 10.9 GW. However, overall electrical demand may overwhelm producers’ desire to cut emissions.

An analyst told S&P Commodity Insights report at the end of 2024 that there was a “strong chance” that many existing coal plants will run longer than anticipated, thanks to the “explosive growth” in energy demand in the U.S.

Nuclear, seen as cleaner and as a dependable alternative to natural gas, faces regulatory and cost problems and won’t be a lead player on the energy provider front for years.

Vogtle units 3 and 4 in Georgia, the country’s last addition to the nuclear fleet, spent 15 years under construction and cost $30 billion, double the original projections for time and cost.

Small modular reactors (SMRs) have been pitched as an alternative, but the models remain, for the most part, in the design stage and face the same regulatory obstacle course as the Vogtle reactors, at least for now.

During his confirmation hearing, Energy Secretary Chris Wright, a longtime proponent of nuclear power, said SMRs are about 10 years away from clearing the necessary development and regulatory hurdles to be placed on the market. Wright was optimistic the task could be achieved in six years.

Tech companies aren’t willing to wait that long. Microsoft paid to restart a unit of the Three Mile Island power plant in Pennsylvania. The electricity provided is expected to cost 10 times the cost of gas-fired power. The problem for other companies is that there just aren’t many nuclear sites around the U.S. that are available for recommissioning.

Follow the growth

Stargate announcement
The $500 billion Stargate Project will build one of the largest power-intensive AI clusters in the world in Abilene, Texas. (Source: Stargate Project)

To appreciate how natural gas feeds the AI revolution, note where the data center campuses are popping up.

One of the most active sites for data center development in the U.S. has sprouted up in the techno center of … Abilene, Texas.

The Trump administration premiered the $500 billion Stargate Project shortly after his inauguration, with the first campus selected for the isolated West Texas town of 130,000 between the Permian Basin and Fort Worth.

“Stargate’s Abilene data center project is expected to scale from 1.2 GW to 5 GW, making it one of the largest power-intensive AI clusters in the world,” said Ethan Warrick, who tracks data center development for East Daley Analytics.

Stargate is the second AI project to come to Abilene. In 2022, Houston-based Lancium broke ground on what was originally planned as a Bitcoin mining operation, powered by renewables in the area. (Abilene has a large number of wind turbines located nearby.)

By 2023, the company had announced a partnership with Crusoe Energy and moved from exclusively Bitcoin mining to building a data-processing campus. While both companies emphasized the area’s available renewable energy, it was difficult for people with knowledge of midstream networks not to notice the number of large-diameter pipelines that passed through the area, carrying natural gas from the Permian Basin to network connections near the Dallas/Fort Worth metro area.

One of the next major pipelines taking gas away from the Permian, Energy Transfer’s Hugh Brinson line, will follow a route that passes close to Abilene. ET announced the project in December.

“We speculate that the new 1.5 Bcf/d project will help feed this emerging demand center,” wrote Warrick in an East Daley report.

Depending on their function, AI data centers are likely to grow in areas with plenty of broadband access, available land and, crucially, near a steady supply of natural gas. Data centers have been built largely near metro areas in Virginia, Texas and Silicon Valley, California. More often now, new facilities are popping up in out-of-the-way places, thanks to the different needs of AI programming.

“When AI models are being trained, typical performance factors such as low latency and network redundancy are less important,” said a McKinsey analysis published in October. “Hence, data centers dedicated to training AI models are being built in more remote locations in the United States, such as Indiana, Iowa and Wyoming, where power is still abundant and grids are less strained.”

In February, Meta announced it was in talks to build a $200 billion AI project with data centers in rural Texas, Louisiana and Wyoming.

The need for speed

Midstream, natural gas and utility companies have been working, sometimes in partnership and sometimes competitively, to secure their positions in the oncoming data center market.

Kinder Morgan has partnered with Southern Co. to boost the Southeastern U.S. natural gas market as more deals are announced.

In 2024, the partners approved the South System Expansion 4 Project to increase the capacity of Southern Natural Gas’ South Line by 1.2 Bcf/d. The $3 billion project is also scheduled to enter service in late 2028, increasing natural gas supplies across the Southeast. The project will provide service to a generally growing area, though company executives say a large portion of demand is coming from AI.

Tech companies have two options—going through the usual red tape offered by utility companies and figuring out how to manage the interconnection process or going straight to the people who already have the supply. Many executives believe the data center builders will come to them.

In February, Energy Transfer announced what may become more of the standard deal for tech and gas companies: a behind-the-meter, 450 MMBtu/d supply of natural gas for the CloudBurst campus, to be built near San Marcos, Texas, between Austin and San Antonio. The supply will be able to generate around 1.2 GW of power for a campus expected to open in 2026.

Behind-the-meter agreements may become the norm. Tech companies generally need to move quickly to remain competitive with other companies around the globe as AI continues to develop at breakneck speed.

“The data centers are not going to wait,” said Alan Armstrong, CEO of Williams Cos., in an interview with Argus Media. “They are going to go to states that allow you to go behind the meter.”

Pittsburgh-based EQT, a natural gas producer that recently purchased its own midstream company, is taking it a step further. During the company’s fourth-quarter earnings call, executives talked about its position as a supplier and a deliverer for data centers.

Speed to market is a critical component,” said EQT CEO Toby Rice. “That is how EQT will differentiate ourselves. Simple one-stop shop, best, cleanest, most reliable, most affordable gas on the market.”

Some other companies are taking advantage of opportunities in different segments of the market.

Chevron announced in January that it will build on-site gas-fired power plants for data centers. The project will use GE Vernova’s natural gas turbines to deliver up to 4 GW—enough to power roughly 3 million homes—to data centers located in the U.S. Southeast, Midwest, and West regions.

The bear case

While companies generally look for opportunities, some analysts have seen reasons to pull back on expectations for a rapidly developing market with parameters that seem to change every other day.

In January, Chinese company DeepSeek premiered its AI application. The product, according to claims, uses far less processing power than U.S. versions and therefore won’t need near the amount of electricity. It took two weeks for natural gas prices to recover from the hit brought about by the announcement.

Analytical firm RBN explained how an expected boom in natural gas demand could fizzle.

“Don’t buy into the hype too quickly,” said the analysis, published in mid-January. “While numerous data centers are indeed in development, natural gas will face stiff competition from wind and solar—green energy sources that data center operators strongly favor over fossil fuels.”

While midstream companies hope to develop carbon capture and storage systems, that sector is still underdeveloped and projects have difficulty obtaining permits from multiple state and federal agencies. Another factor is the dealmaking. Midstream companies often need a 10- to 20-year commitment to pay for the infrastructure needed. Many tech firms are likely to hesitate before making a deal of that duration.

“No utility wants to risk building costly infrastructure only to see demand fade as the industry discovers ways to reduce AI’s power consumption,” RBN said.

Finally, natural gas power projects generally take years to build, even when everyone isn’t fighting over the same equipment needed to build generators. Tech companies may have overly optimistic timelines that don’t match up with reality.

However, the overall case for natural gas development remains bullish, according to analysts. The market is getting a boost from LNG exports that are likely to take up a larger share of the market, executives at Expand Energy, the country’s largest natural gas producer, said at NAPE.

“Just to be clear, in our view, at least 75% of the natural gas demand growth is going to come from LNG,” Expand Energy CFO Mohit Singh said.

Others remain confident that the market will develop, regardless of the pace.

Stuart Saulters, vice president of government affairs at GPA Midstream, said he had talked to a representative of a utility company following the DeepSeek announcement. If the tech world can get the same thing done with less material, the tech world was likely to do more overall.

“He told me, even if they don’t need as much power for the chips, that just means they’ll do more than they’re planning to now,” Saulters said.

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.