The AI trade has minted some obvious winners. Nvidia Corp. (NVDA) owns the compute layer, while Vertiv Holdings (VRT) helps keep data centers cool and powered internally. But Hyliion Holdings Corp. (HYLN) CEO Thomas Healy says there's a critical piece of the AI infrastructure puzzle that neither company is built to solve: generating the electricity in the first place.
"None of Nvidia's GPUs run without reliable power. None of Vertiv's cooling systems can operate if the facility can't get interconnects," Healy told MarketDash.
As AI companies race to build larger data centers, access to electricity is emerging as one of the industry's biggest bottlenecks. Utilities across the U.S. are struggling to keep up with demand, leaving developers facing years-long waits for grid connections.
The Problem One Layer Upstream
For years, investors viewed AI through the lens of semiconductors. More recently, attention expanded to cooling systems, power distribution equipment and data-center infrastructure. Healy says the next investment theme sits one layer further upstream.
"We want to be a major player in the power generation layer," he said in an email.
Hyliion's KARNO generator technology is designed to provide onsite electricity generation for customers that cannot get enough power from the grid. The system can run on multiple fuel sources and is aimed at applications ranging from data centers to industrial facilities.
The thesis is simple: AI data centers cannot operate without electricity, and securing that electricity is becoming increasingly difficult.
"The grid doesn't have capacity for this," Healy said when discussing the rapid growth of AI infrastructure. "These facilities can consume gigawatts of power, and in most markets the interconnect queue is measured in years, not months."
Why AI Is Changing The Equation
The power challenge has become impossible for hyperscalers to ignore. According to Healy, every major AI company is actively focused on securing power capacity as demand for computing continues to surge.
"The constraint on AI deployment right now isn't chips, it isn't talent, it isn't data, it's power," he said.
That dynamic is creating an opportunity for companies offering alternative power solutions. Rather than waiting years for new grid connections, some operators are increasingly exploring onsite generation to accelerate deployments. The trend has become large enough that Healy believes distributed generation could eventually become standard infrastructure for AI facilities.
"The companies that figure out onsite power generation are going to have a structural advantage in deployment speed," he said.
A New AI Infrastructure Category?
Hyliion is still a fraction of the size of Nvidia or Vertiv. But Healy sees the company's opportunity through a similar infrastructure lens.
"Nvidia owns the compute layer. Vertiv owns a lot of the thermal and power distribution inside the data center," he said. "We're solving the problem one layer upstream."
As investors search for the next picks-and-shovels beneficiaries of the AI boom, power generation is increasingly joining chips, cooling and networking as a key piece of the conversation. For Hyliion, the bet is that the biggest AI opportunity may not be inside the data center at all — it may be the power plant sitting next to it.
