On-site hydrogen power for AI data centers is generation built inside the facility's own fence line, so the compute does not wait on a utility interconnection. Our design runs fuel-flexible generators from day one on pipeline gas, then makes its own hydrogen on site from aluminum in a proprietary reactor. That one reaction produces three things: hydrogen, alumina and heat. The generators' waste heat is routed back to drive it.Key takeaways
You can get the compute. You can find the land. The grid connection is what holds the schedule back.
The IEA estimates global data center electricity use at around 415 TWh in 2024 and projects around 945 TWh by 2030 in its Base Case (IEA, Energy and AI, 2025). The grid can't be expanded at that speed. Berkeley Lab found that US projects built in 2025 took a median of over five years from interconnection request to commercial operation, and about 1,312 GW of generation was still waiting in US queues at the end of 2025 (Queued Up, 2026 edition).
A facility that waits in that queue pays for land, capital and a design that sits idle, and its hardware roadmap moves on without it. So much of the market has moved to generation behind the meter. That solves the timeline, but it leaves a second problem: a facility running on bought-in fuel is fully exposed to that fuel's price for its whole life. Our design is built to solve both.
For the per-rack and per-megawatt numbers behind this, see how much power an AI data center actually needs.
Two fuel paths feed one bus, and a closed heat loop joins them. Nothing in the loop depends on a utility being ready.
Fuel-flexible linear generators start on pipeline natural gas, so the facility powers its first compute without waiting for the reactor or the utility. These are the same generators that later run on hydrogen, so nothing is thrown away when the fuel changes.
Aluminum goes into our proprietary reactor. One reaction releases hydrogen, which fuels the generators on site, and leaves alumina, which is sold as an industrial co-product. Aluminum is a dense, stable solid that ships and stores like any other commodity. That avoids the compression, liquefaction and delivery logistics that make bought-in hydrogen expensive.
Generator exhaust heat goes back to drive the reactor, so the energy system's waste heat becomes one of its inputs. Heat from the chips leaves through a sealed two-phase direct-to-chip loop and is rejected to air, with no evaporative cooling water. Process water in the reactor is recovered and recirculated.
Every generator feeds one electrical bus with N+1 inside the array, and power reaches the racks over Class 4 fault-managed distribution. Capacity is added in repeatable 10 MW self-powered pods. On our model, the first pod takes 18 to 24 months and each pod after it 6 to 8 months, and the racks are designed for roughly 250 kW each.
| Layer | What It Does | Design Basis |
|---|---|---|
| Hydrogen reactor | Turns aluminum into hydrogen fuel, alumina co-product and heat, inside the fence line | Proprietary AJ Projects design. A pilot is running. |
| Generation array | Fuel-flexible linear generators that run on natural gas, hydrogen or a blend | N+1 inside the array. 70 dB(A) at 3 ft per unit. |
| Heat recovery | Routes generator exhaust heat back into the reactor | Closed loop, heat out, fuel in |
| Power distribution | Delivers power from one bus to the racks | Class 4 fault-managed power |
| Compute cooling | Removes chip heat through a sealed dielectric refrigerant loop | Two-phase direct-to-chip, ~250 kW racks, no evaporative water |
| Water recovery | Recovers and recirculates reactor process water | Instrumented recovery train, in pilot |
| Operations layer | Orchestrates generation, fuel, heat and load as one system | Licensable to other operators |
This is a standalone line. You do not have to build a data center with us to use our power, and you do not have to use our power to build with us.
An energy asset whose value doesn't rest on a single commodity. It has compute revenue, a co-product revenue line and a licensable operations layer, and it is built in repeatable 10 MW increments you can underwrite pod by pod.
Drop self-powered pods into your own site to reach power before your interconnection clears, or to stop depending on it. Keep your own architect, GC and tenants. We supply and integrate the energy system.
Discuss an integration →Aluminum supply, alumina offtake, gas supply and heat users all sit on the system's inputs and outputs. It is designed around industrial partners, not a single site.
How aluminum becomes hydrogen →We will walk you through the design, the modeled economics and what the pilot has and hasn't proven yet, in a session built for investment committees and integration teams.
Request an Investor Briefing →No pitch · No obligation · 30 minutes of actual strategy
| Dimension | Grid-Tied | Gas-Only Behind the Meter | Delivered Hydrogen | AJ Projects System |
|---|---|---|---|---|
| Time to power | Set by the interconnection queue | Fast | Fast, but set by hydrogen supply | Fast. The gas bridge runs from day one. |
| Fuel price exposure | Tariff and capacity charges | Full gas price exposure | Delivered hydrogen price, today well above gas | Hedged across gas and self-made hydrogen |
| Fuel logistics | None | Pipeline | Compressed or liquid hydrogen deliveries and storage | A stored solid (aluminum), turned into hydrogen on site |
| Waste heat | Rejected | Rejected | Rejected | Drives the reactor |
| Revenue beyond compute | None | None | None | Alumina co-product and a licensable operations layer |
| Main risk | Schedule | Fuel price and emissions | Fuel cost and supply | Scale-up of a proprietary reactor. See below. |
Investors should hear this from us first, not find it in diligence.
| Item | Status |
|---|---|
| Reaction and water recovery | A pilot plant is running with instrumented reaction and water-recovery trains. |
| Alumina grade | Not yet independently assayed. We do not quote a grade until an independent lab has confirmed it. |
| Campus scale | Not yet proven. A 73 MW self-generating campus in North Texas is in development, before construction. |
| Cost per kWh | $0.044 is a modeled operating cost (fuel and O&M, running mainly on pipeline gas), excluding generator capital, against a Texas commercial grid rate of roughly $0.070. It is not a metered result. |
| Uptime | Five nines is a design target, carried by N+1 inside the generator array. |
Why say this on a sales page? A system that overstates its own completion is worse than one that states its gaps, because it gives people a reason to stop looking. We would rather you underwrite the real risk. That risk is scaling a proprietary reactor, not whether the physics works.
It is generation built inside the facility's own fence line that runs on hydrogen, so compute does not wait on a utility interconnection. In the AJ Projects design, fuel-flexible linear generators feed one electrical bus. They start on pipeline natural gas from day one and move to hydrogen made on site from aluminum in a proprietary reactor. That reaction produces hydrogen, alumina and heat, and the generators' exhaust heat goes back to drive it.
Aluminum reacts with water to release hydrogen and heat and leave aluminum oxide, which is alumina. This is textbook chemistry. What has historically stopped it being useful is the thin oxide film that forms on aluminum and stops the reaction. Our reactor is a proprietary design that makes the reaction practical at facility scale. We do not publish its internals.
No. The system is designed to run without the grid. A grid connection can still be useful for backup or other commercial reasons, but the compute does not depend on one to start or to run.
The compute cooling uses a sealed two-phase direct-to-chip loop that rejects heat to air, with no evaporative cooling water. The reactor process does use water, which is recovered and recirculated through an instrumented recovery train. We say that plainly rather than claim the facility uses no water.
Yes. The energy system is a standalone line and it is not exclusive to AJ Projects facilities. Developers, operators and industrial sites can integrate self-powered pods and keep their own architect, general contractor and tenants. We supply and integrate the energy system.
A pilot plant is running with instrumented reaction and water-recovery trains. The alumina grade has not yet been independently assayed, and campus scale has not yet been demonstrated. The $0.044 per kWh operating cost (fuel and O&M, excluding generator capital), the pod timelines and five-nines uptime are design and model figures, not metered results. A 73 MW self-generating campus in North Texas is in development, before construction.
It has three lines: the compute or colocation capacity the power serves, the alumina co-product the reaction leaves behind, and the operations layer that runs generation, fuel, heat and load together, which can be licensed to other operators. Waste heat is used inside the system to drive the reactor, not sold.
The energy system is one layer. If you are developing the whole facility, retrofitting a legacy site or converting an industrial building, our data center solution architects design the whole facility and direct a network of 165 partners and businesses who build it. For the facility-level model, see the self-powered AI data center.
Whether you plan to invest in the system, integrate it at your site or build a campus around it, the first conversation is the same: your load, your timeline and your fuel position.
Request an Investor Briefing →No pitch · No obligation · 30 minutes of actual strategy
Hello, we identify our visitors so we can improve your experience and ensure you are never left out. Please select Allow Visibility or see our Privacy Policy to continue.