The artificial intelligence industry is facing a critical infrastructure crisis: data centers need so much electricity that plugging them into existing power grids is no longer viable, forcing companies to build their own power generation facilities. Goldman Sachs expects $765 billion in AI infrastructure investments this year alone, with a significant portion directed at solving three major bottlenecks: power supply, memory chips, and networking bandwidth.

Why Can't Utilities Keep Up With AI's Power Demands?

The scale of electricity consumption is staggering. The International Energy Agency expects data centers' total power consumption to roughly double between now and 2030, reaching 945 terawatt-hours (TWh), enough electricity to support a few dozen major cities. Most utility companies, however, aren't in a strong financial position to add the required electricity production capacity in the timeframe needed.

The problem creates a vicious cycle: utilities are passing along their new buildout costs to consumers in the form of higher rates. This approach works in the short term but isn't sustainable long-term. As a result, artificial intelligence data centers are increasingly pursuing self-sufficiency by generating their own power rather than relying on the grid.

How Are Data Centers Solving the Power Crisis?

Companies are turning to multiple solutions to generate their own electricity. Natural gas turbines are emerging as a practical near-term answer. GE Vernova, the power division of the historic General Electric conglomerate, is supplying its LM2500XPRESS gas turbines to AI infrastructure operators. The company is providing 29 of these turbines to Crusoe Energy, while oil giant Chevron is testing one of its natural gas turbines as a power source for a utility service specifically designed to serve AI data centers.

Beyond natural gas, small modular reactors (SMRs) are attracting significant investor attention as a longer-term solution. Companies like Oklo and NuScale Power are developing advanced nuclear technologies that could deliver reliable, carbon-free electricity around the clock. The successful public debut of SpaceX earlier this month has sparked broader investor interest in infrastructure companies supporting AI expansion, including those focused on energy generation.

NuScale Power has a regulatory advantage: its SMR design already has U.S. Nuclear Regulatory Commission approval and uses conventional low-enriched uranium, making it more available than some advanced nuclear fuels. Oklo, meanwhile, is pursuing a broader strategy that extends beyond reactor development to include fuel fabrication, fuel recycling, and other nuclear-related services.

What Other Infrastructure Bottlenecks Are Slowing AI Growth?

Power is only one piece of the puzzle. The AI industry faces two additional critical constraints that are consuming massive capital investment:

• Memory Chip Shortage: Only three major computer memory manufacturers exist: Micron Technology, SK Hynix, and Samsung. Micron is particularly focused on high-bandwidth memory (HBM) that consumes less power and occupies less space on data center circuit boards. The company is sold out of this high-margin memory into 2027, with the HBM market expected to grow at an average annual pace of more than 25% through 2035.
• Optical Networking Capacity: Data centers need to move information at the speed of light using fiber-optic connections. Goldman Sachs believes the optical networking market will eventually grow ninefold to a $150 billion-plus business. Marvell Technology recently unveiled the industry's first-ever switch capable of processing 102.4 terabits of data every second, allowing AI data centers to achieve the next level of computing performance.
• Cooling and Thermal Management: Vertiv Holdings designs and services the power, cooling, and liquid cooling systems that keep data centers running. The company has a roughly $15 billion backlog and has made recent acquisitions like ThermoKey and PurgeRite to deepen its exposure to water-efficient cooling solutions for AI-heavy infrastructure.

How Are Companies Positioning Themselves for This Opportunity?

GE Vernova's power division is experiencing explosive growth. While its revenue grew 10% year-over-year to $5 billion in the first quarter of 2026, the company took $10 billion worth of new equipment orders during the same quarter. The company's backlog now stands at $163 billion, compared to Q1's total revenue of only $9.3 billion, indicating extraordinary demand for its power generation solutions.

Micron Technology is strengthening its position through long-term strategic customer agreements rather than traditional contracts. CEO Sanjay Mehrotra noted that the company is working with customers on agreements "that have specific commitments over a multiyear time horizon for improved visibility and stability in our business model," and highlighted that Micron signed its first five-year strategic customer agreement.

Sanjay Mehrotra

"We continue to work with customers on strategic customer agreements, or SCAs, that are different from prior LTAs and have specific commitments over a multiyear time horizon for improved visibility and stability in our business model," said Sanjay Mehrotra.

Sanjay Mehrotra, CEO at Micron Technology

Lumentum Holdings, Dell Technologies, and Vertiv Holdings are also capturing significant portions of AI infrastructure spending. Lumentum supplies the optical chips and lasers that move data inside cloud and AI data centers. Dell is tying record AI server backlogs and government contracts to its broader enterprise business. Vertiv's accelerating AI backlog and strong profitability position it at the heart of AI infrastructure, though its very high valuation multiple leaves little room for execution missteps.

Steps to Understanding AI Infrastructure Investment Opportunities

For investors and industry observers tracking this space, here are key factors to monitor:

• Backlog Growth: Track companies with expanding order backlogs relative to quarterly revenue, as this indicates sustained demand. GE Vernova's $163 billion backlog versus $9.3 billion quarterly revenue demonstrates the scale of future work.
• Strategic Customer Agreements: Long-term contracts with hyperscalers like Microsoft, Amazon, and Google provide revenue visibility and reduce market volatility. Micron's shift toward multiyear strategic agreements signals confidence in sustained demand.
• Regulatory Status: For nuclear and advanced energy solutions, regulatory approval status matters significantly. NuScale's U.S. Nuclear Regulatory Commission approval puts it ahead of earlier-stage competitors like Oklo.
• Execution Risk: Early-stage companies like Oklo and NuScale remain speculative and depend on financing, customer contracts, and actual project delivery, not just investor enthusiasm.

What Does This Mean for the Broader AI Industry?

The infrastructure constraints reveal that AI's growth is no longer limited by algorithmic innovation or computing power alone. Instead, the industry is now constrained by physical resources: electricity generation capacity, semiconductor manufacturing, optical networking equipment, and cooling systems. Companies that can solve these bottlenecks are positioned to capture enormous value as AI spending accelerates.

However, investors should remain cautious. While companies like Oklo and NuScale Power are developing promising technologies, they remain early-stage and speculative. Their long-term success depends on regulatory approvals, financing, customer contracts, and actual project delivery, not just investor enthusiasm. Similarly, established players like GE Vernova, Micron, and Vertiv face execution risks and valuation pressures that could affect returns.

The $765 billion AI capex cycle is real, but the winners will be those companies that can reliably deliver power, memory, networking, and cooling solutions at scale. The race to build AI's physical backbone is just beginning, and the stakes have never been higher.