Antares Raises $470M for 1 MW Military Microreactors as Orange Backs 400 MW of French Data Centers
Two energy financings closed within hours of each other, aimed at the same shortage from opposite ends. Antares, developing one-megawatt reactors for United States military bases, raised $370 million in equity and $100 million in debt in a round co-led by Paradigm and Caffeinated. French carrier Orange and infrastructure investor Morrison agreed to build a data center platform in France targeting 400 megawatts, backed by a €3 billion investment program.
One is small, distributed, and defense-anchored. The other is large, centralized, and telecom-anchored. Both exist because compute demand has outrun the ability of existing grids to deliver firm power on the timeline buyers require.
The megawatt is now the scarce unit
Data center development used to be constrained by capital, land, and fiber. Those constraints have been replaced by interconnection queues. In most developed grids, the wait for a large new connection is measured in years, and the queue does not clear faster because the applicant is willing to pay more.
That reality reshapes every financing in the sector. A project with secured power ahead of the queue carries a premium over an identical project without it, and the entire structure of these two deals reflects the same insight arrived at independently.
Why a telecom is building the French platform
Orange’s participation is not incidental. Telecom operators hold assets that are unusually well matched to the constraint: existing sites with established grid connections, rights of way, backhaul fiber already terminated, and decades of regulatory relationships with grid operators and municipalities.
Central offices and exchange buildings were built when copper switching required distributed real estate near population centers. Much of that estate has been rendered surplus by fiber consolidation, and it comes with electrical service already provisioned. Converting a site with an existing connection is a fundamentally different regulatory exercise from applying for a new one, which is the arbitrage a 400-megawatt target implies.
The French siting choice adds a second advantage. France operates one of the most nuclear-heavy grids in Europe, which means a load of this scale can be served with low-carbon firm power without the additionality argument that complicates renewable procurement elsewhere. A data center platform can make a credible carbon claim on grid power alone.
Why the microreactor round is defense-anchored
Antares is targeting one megawatt per unit, which is small by any conventional power generation standard and roughly the scale of a single data hall’s requirement rather than a campus.
The military base application is the more important detail than the capacity. Defense installations offer four conditions that a commercial site does not: an existing security perimeter satisfying physical protection requirements, a customer with a mission requirement for power that survives grid failure, a federal procurement and permitting pathway that differs from the commercial nuclear route, and an operator willing to pay a premium for resilience rather than the lowest levelized cost.
That combination is the standard route to market for a novel reactor design. The first units are not competing against grid electricity on price. They are competing against diesel generators and fuel convoys on availability, and against nothing at all on the requirement to keep operating when the grid does not.
The commercial path runs through the demonstration
If small modular and micro reactors reach data center applications, it will be by way of installations like these. The technical case is straightforward: firm, dispatchable, carbon-free power sited at the load, bypassing interconnection entirely.
The obstacles are regulatory and financial rather than technical. Licensing frameworks were built for gigawatt-scale plants and adapt awkwardly to units three orders of magnitude smaller. Fuel supply chains for advanced designs are thin. Manufacturing learning curves require volume that does not exist until the first units prove out. Each of those resolves through deployment, which is why an anchor customer with non-price requirements matters more at this stage than any cost projection.
The third case is the one that got scrutiny
The same week produced reporting on the process Meta used to secure its Hyperion data center in Louisiana, including tax concessions and limited public input.
That is the third financing model for the same shortage, and the least sustainable of the three. Where Orange is monetizing existing infrastructure and Antares is selling resilience to a customer that values it, the Louisiana approach transfers cost to a public that was not meaningfully consulted. The megawatts get built either way. The difference is who absorbs the price, and that difference determines how long the political license for the next round of buildout lasts.