Data centers powering artificial intelligence are abandoning the traditional power grid in favor of on-site ammonia engines that can be deployed in under two years, according to a breakthrough test completed in Houston. Two companies, Amogy and 2G Energy, demonstrated that liquid ammonia can be split into hydrogen and fed directly into a reciprocating engine to generate electricity, eliminating the need for hydrogen storage tanks or fuel cells. The test represents a practical solution to a growing infrastructure crisis: interconnection delays in major power markets like ERCOT and PJM now stretch multi-year timelines, forcing hyperscalers to build their own power sources behind the meter.

What Problem Are Data Centers Actually Solving?

The power grid connection bottleneck has become severe enough that it is reshaping how companies deploy AI infrastructure. Interconnection in ERCOT and PJM is now a multi-year problem, and nobody spending billions on an AI buildout wants to explain a two-year power delay to a board. This reality has made on-site generation the default strategy rather than an exception. The ammonia-to-power approach offers something different from the gas engines already common at data centers: a fuel pathway that could eventually transition to low-carbon ammonia if the market matures, while still burning natural gas today.

The Houston test, announced on August 3, proved that two very different pieces of hardware could work together reliably. Amogy supplied the ammonia reformer, a chemical reactor that cracks liquid ammonia into hydrogen gas and nitrogen. 2G Energy supplied an Agenitor 412 reciprocating engine-generator set, a machine the German company has been selling into European cogeneration for years. The hydrogen-rich stream from the reformer ran the engine successfully, demonstrating what both companies call a critical "proof point" for commercial deployment.

"An important proof point for Amogy's ammonia-to-power platform," said Seonghoon Woo, CEO of Amogy.

Seonghoon Woo, CEO at Amogy

What actually got proven is the handshake between two systems: fuel delivery, control systems, and two very different pieces of hardware agreeing on what they are doing. That sounds unglamorous until you remember it is the step that kills most integration projects. The companies also say the engine can be configured to throw off usable heat alongside the electricity, and that the architecture is modular enough to add capacity in stages rather than all at once.

Why Is Ammonia Chemistry Suddenly Relevant to Data Centers?

Ammonia is NH3, three hydrogen atoms hanging off one nitrogen atom. Push it through the right catalyst at the right temperature and it splits, or cracks, into hydrogen gas and nitrogen gas. The hydrogen is the point. The nitrogen makes up about 78 percent of the air you are breathing right now, so nobody spends much time worrying about where it ends up. Amogy has been selling this trick for years, mostly pointing the resulting hydrogen at fuel cells. The company powered a drone, a farm tractor, and a semi-truck, then in September 2024 sailed a retrofitted 1957 tugboat named the NH3 Kraken on a Hudson River tributary north of New York City.

Houston is different because the endpoint is a reciprocating internal combustion engine. Pistons, crankshaft, connecting rods, valves. The distinction matters more than it sounds: a fuel cell converts hydrogen electrochemically and a piston engine burns it, and those are two entirely separate engineering problems with separate emissions profiles. Burning a hydrogen-rich stream in air at combustion temperature makes nitrogen oxides. A fuel cell does not, and that gap is what an air permit writer in Texas gets paid to care about.

The fuel cell version of this argument is already running commercially. One Santa Clara data center has spent more than two years off the utility grid making its own electricity from hydrogen fuel cells and cooling its racks with the water those cells produce. Amogy and 2G are betting a piston engine is the cheaper, more familiar path to the same place.

How Are Reciprocating Engines Winning the Speed Race?

Amogy and 2G are not chasing a niche. They are chasing the fastest-growing corner of American power procurement. According to research from Enverus Intelligence Research published on August 11, 25.5 gigawatts of new US industrial demand will rely on behind-the-meter generation between 2026 and 2030. Data centers account for 22.5 gigawatts of that, or 88 percent, which works out to roughly 36 percent of all US data center capacity added over the period.

The speed advantage is decisive. Piston engines, fuel cells, and turbines in the small and medium frame sizes take 61 percent of the gas-fired generation total planned for data centers. These machines usually reach commercial operation in under two years, where a big combined-cycle plant can take closer to seven years. The faster-deploying options carry what analysts call "a significant timing advantage".

• Piston Engines: Expected to provide 7.1 gigawatts of the 31.6 gigawatts of generation needed for data center expansion between 2026 and 2030, deployable in under two years.
• Fuel Cells: Projected to supply 4.8 gigawatts, offering zero-emission operation but at higher capital cost than reciprocating engines.
• Gas Turbines: Estimated to deliver 6.2 gigawatts, providing flexibility but with longer deployment timelines than piston engines.

2G Energy is positioned to capture significant market share in this shift. On July 30, the company confirmed second-quarter order intake of 422.4 million euros, against 54.1 million euros a year earlier. US data centers alone accounted for 350.3 million euros of that, up from 8.3 million euros in the same quarter the previous year. First-half order intake came to 479.4 million euros, and management held full-year revenue guidance at 490 million euros with an EBIT margin of 9.5 to 10.5 percent.

What Is the Real Sales Pitch Behind Ammonia Engines?

Here is where the pitch gets clever and a little hedged at the same time. Low-carbon ammonia at hyperscale volume does not exist yet. So neither company is asking a data center operator to bet a multi-billion-dollar campus on a fuel with no mature market behind it. What they are selling is one physical machine with two modes: install it now, run methane today because that is what the pipeline delivers, and pipe in the reformer later if the ammonia market matures or the sustainability reporting starts biting.

Strip the press-release language off that and you get a pretty honest proposition: buy the engine, keep the option. 2G CEO Pablo Hofelich framed the collaboration as evidence its engine platform can handle a broad set of fuel pathways, ammonia-derived hydrogen included. Whether a hyperscaler pays a premium for optionality it may never exercise is a different question. Optionality is easier to sell to an engineering team than to a CFO.

The flexibility matters because more than 70 percent of the world's ammonia is made out of natural gas to begin with. The pathway to low-carbon ammonia exists in theory, but the infrastructure does not yet exist at scale. By selling a dual-mode engine, 2G and Amogy are hedging their bets while positioning themselves for a future where ammonia supply chains mature. Data center operators get a machine that works today and a potential upgrade path for tomorrow.

The convergence of grid delays, fast-deploying hardware, and the need for on-site power generation has created a market that did not exist five years ago. Reciprocating engines, fuel cells, and turbines are all competing for the same customers, all racing to deploy faster than the grid can connect them. Ammonia-powered engines represent one more option in that race, and the Houston test suggests the option is now technically viable.

" }