AI Investments Keep Lining Up, But Are We Sure About Returns?

The AI gold rush is in full swing. Hyperscalers like Microsoft, Amazon, Google, and Meta are pouring hundreds of billions into data centers, chips, and infrastructure to power the next wave of artificial intelligence. NVIDIA just posted record revenues driven almost entirely by AI demand. Yet one question looms large for investors and energy markets alike: Are the returns really there, or is this a classic case of hype outrunning reality?

A sharp Bloomberg Opinion piece published just days ago cuts right to the chase. Titled “AI’s Big Guns Have a Serious Inflation Problem,” the article highlights how Big Tech’s trillion-dollar AI bet is running into “chipflation”—rising prices for GPUs, memory, and other components that are inflating capex far beyond initial forecasts. It’s not just about building more data centers; the cost of the gear inside them keeps climbing, squeezing margins and raising doubts about whether hyperscalers can earn an adequate return on these massive investments. The piece warns that AI infrastructure costs “just keep on rising,” with hyperscalers on the hook for several trillion dollars over the coming years.

That skepticism matters—especially for the energy sector, which is being asked to supply the gigawatts of reliable power these AI facilities demand. While the tech giants chase returns on silicon, traditional energy players are stepping up with practical, behind-the-meter power solutions. Natural gas-fired generation, on-site plants, and long-term fuel supply deals are suddenly hot commodities. The question isn’t whether AI investments are lining up—they are. The real question is who actually makes money when the dust settles.

The Spending Spree: Trillions in Play, But Returns Under Scrutiny

Numbers don’t lie: AI infrastructure capex from the big four hyperscalers (Amazon, Microsoft, Alphabet/Google, Meta) is now projected at roughly $725 billion for 2026 alone—a 77% jump from 2025. Amazon is guiding toward $200 billion, Microsoft and Google each near $190 billion, and Meta around $145 billion. Wall Street estimates put total AI-related spending (including Oracle and others) north of $800 billion this year, with some forecasts topping $1 trillion in 2027.

NVIDIA, the clear winner on the chip side, reported fiscal 2026 full-year revenue of $215.9 billion (up 65%), with data center/AI revenue hitting $193.7 billion. Its Q4 alone was $68.1 billion. Cloud providers are seeing strong growth too—Google Cloud up 63% in Q1, Microsoft Intelligent Cloud up 29%, AWS up 28%—but the capex bills are enormous. Many analysts note that while core businesses (ads, e-commerce, Office) keep generating huge free cash flow, the AI buildout is consuming cash at a historic pace. Surveys show enterprises are seeing revenue lifts and cost savings from AI adoption, but monetization at hyperscaler scale remains the big unknown.

Bloomberg’s “chipflation” warning is real: component prices are up, supply chains are tight, and the AI boom is even crowding out conventional chips for consumer electronics. Add in exploding power needs—data centers could consume as much electricity as entire countries by 2026—and the risk of overspending becomes clear. If AI revenue growth doesn’t keep pace, these investments could look like a very expensive bubble.

Energy to the Rescue: Hyperscalers Need Power, and the Sector Is Delivering

Here’s where the energy industry shines. Data centers can’t run on hype—they need firm, dispatchable power now. Grid constraints are real, so hyperscalers are turning to private, behind-the-meter solutions powered by natural gas. This creates a direct, high-margin revenue stream for energy companies that can deliver scalable generation and fuel supply faster than new nuclear or renewables can scale.

Several traditional energy names are already moving aggressively into the AI power space:

Liberty Energy (LBRT): The clear early leader in on-site power for data centers. In January 2026, Liberty Power Innovations (a Liberty subsidiary) partnered with Vantage Data Centers to develop and operate up to 1 GW of high-efficiency power solutions across North America, with 400 MW firmly reserved for 2027. Liberty recently inked a 330 MW power reservation in Texas and signed two major equipment contracts worth $505 million with Bergen Engines for over 500 MW of generation capacity. The company now targets 3 GW of distributed power by 2029. Liberty’s model—owning and operating the power assets—gives it long-term, contracted revenue tied directly to AI demand while leveraging its oilfield expertise for rapid deployment. Management has called this pivot a “new standard for scale and reliability.”

Chevron (CVX): The supermajor is all-in on dedicated gas-fired power for AI. In January 2025, Chevron formed a joint venture with Engine No. 1 and GE Vernova to build up to 4 GW of behind-the-meter natural gas power plants co-located with data centers. Projects are targeted for the Southeast, Midwest, and West, with first facilities online 2027–2028. Chevron is also advancing a 2.5 GW (expandable to 5 GW) power complex in the Permian Basin specifically for a major data-center client. By moving from fuel supplier to integrated power provider, Chevron is locking in higher-value electrons while using its vast gas resources.

ExxonMobil (XOM): Similar playbook. Exxon has a pipeline of over 2.7 GW in data-center power projects, including natural gas-fired plants with carbon capture and storage (CCS) to meet hyperscaler sustainability goals. Partnerships with utilities and direct off-take agreements position Exxon as a reliable, low-emission power partner.

EQT Corporation: As the largest U.S. natural gas producer, EQT is the upstream fuel kingpin. It holds massive reserves (9+ years without new drilling) and is actively supplying gas to power plants serving AI data centers—including exclusive supply for a 4.4 GW redevelopment project at the former Homer City coal site. EQT’s low-emission certified gas is especially attractive to tech buyers, and management is unhedged for 2026 in anticipation of higher prices driven by AI demand.

Other midstream players like Williams are also building modular gas-to-power plants, but Liberty, Chevron, Exxon, and EQT stand out for their direct, large-scale commitments.

Which Energy Names Are Set Up Best for AI-Driven Returns?

Liberty Energy looks uniquely positioned for nimble, high-return growth. Its power subsidiary model delivers contracted, long-duration revenue with lower commodity risk than pure upstream plays. The Vantage partnership and recent equipment orders show real momentum, and the company’s oilfield DNA gives it a speed advantage in execution.

Chevron and ExxonMobil bring balance-sheet scale, CCS expertise, and integrated operations that hyperscalers crave for reliable, lower-emission power. Their projects are multi-gigawatt and multi-year, offering stable cash flows even if AI spending slows.
EQT wins on pure volume and cost structure. As the marginal gas supplier to new power generation, it benefits from any uptick in nat-gas demand without needing to build power plants itself.

All four are far better positioned today than they were two years ago. While hyperscalers wrestle with chipflation and uncertain AI ROI, these energy players are selling something tangible—electrons and molecules—that AI literally cannot run without.

The Bottom Line

AI investments are lining up faster than ever. Hyperscalers are betting trillions that the technology will transform the economy. But as Bloomberg rightly points out, the path to returns is getting narrower thanks to rising component costs, power constraints, and the sheer scale of the buildout. For the energy sector, that uncertainty is an opportunity. The power demand is real, immediate, and growing—regardless of whether the next ChatGPT killer arrives on schedule.

Energy investors aren’t betting on AI hype.

They’re betting on the infrastructure that makes it possible. And right now, Liberty Energy, Chevron, ExxonMobil, and EQT look like some of the best-positioned plays in the entire AI value chain.

One thing is certain: people are going to start looking at where the hyperscalers are operating, and if the Data Center project becomes a negative headline for water theft or eminent domain land grabs, that could hurt the project and stock prices. This should be a huge warning to executives to only pick good data center projects that would minimize the potential downside of public outcry against data centers, which is gaining traction.

The huge data center that was just caught tapping into a water system without approvals would be a major black eye, and market-damage control would need to be called in.