Data Centers Are Leaving the Grid - Fuel Cells vs Small Nuclear

Why Data Centers Are Walking Away From the Grid

Imagine you're building a new data center in the US today. The biggest bottleneck isn't land or GPUs. It's grid interconnection approval. New grid connections in the US average 5 to 10 years — permits, transmission upgrades, substation construction, all of it.

Meanwhile, AI moves in quarters. Meta wants to train a new model now, not in 2034. So Big Tech reached the obvious conclusion: skip the grid and build your own power plant.

About one-third of new data centers are now off-grid or partially off-grid. From where I sit, this isn't just a tech trend — it's a structural rewrite of the utility industry. Big Tech is becoming its own utility.

Two Paths: Fuel Cells vs Small Modular Reactors

Off-grid generation splits into two main camps: natural gas fuel cells and small modular nuclear reactors (SMRs). Let's compare the leaders in each.

Bloom Energy (BE): The Right-Now Solution

Bloom Energy makes solid oxide fuel cells that convert natural gas directly into electricity. Drop them next to a data center, connect a gas pipe, and you have power — no transmission lines required.

The recent Oracle deal supplies enough power to serve 2 million homes, for a single company's data center footprint. One contract. The stock is up 105% over the last six months.

Bloom's edge is speed. The units are modular, deployable in months, and the permitting is trivial compared to a nuclear reactor.

Oklo (OKLO): The Long-Dated Nuclear Bet

Oklo builds small modular fission reactors — essentially mini nuclear plants sized for a single facility. The company recently signed a deal to supply Meta's Ohio data center.

The SMR pitch is carbon-free baseload. Runs 24/7, near-zero emissions, lets Big Tech keep its ESG commitments while securing power.

The downside is time. NRC licensing takes years, and commercial power output won't materialize until the late 2020s or 2030s. This is a long-horizon scenario bet, not a near-term revenue story.

Side-by-Side

Dimension	Bloom Energy (BE)	Oklo (OKLO)
Technology	Fuel cells (natural gas)	Small modular reactors (fission)
Deployment speed	Months	5-7 years
Carbon emissions	Moderate (gas-based)	Near zero
Near-term revenue visibility	High (Oracle, others)	Low (awaiting licensing)
Anchor customer	Oracle, AT&T	Meta (via Zoox subsidiary)
6-month stock move	+105%	High volatility, licensing-driven

Uranium: The Shared Pick-and-Shovel

If the nuclear renaissance is real, every reactor needs uranium. But almost no new reactors were built for a decade, so almost no new uranium mines opened. Supply is tight.

Cameco (CCJ) is one of the world's largest uranium suppliers. The stock is up about 30% over six months — quiet compared to BE or OKLO, which is exactly why I find it interesting here. Microsoft, Amazon, and Google are all in active negotiations on SMR deals. If those convert, uranium demand jumps like a step function.

Risks I'm Watching

Each path has its own failure mode.

Bloom risk: Natural gas price spike compresses margins. If Big Tech tightens ESG pressure on gas-based solutions, new contract growth slows.

Oklo risk: NRC delays or a botched first commercial deployment would damage the entire SMR narrative. The current price already discounts a lot of "it will work."

Cameco risk: Uranium prices have already moved meaningfully. Faster mine restarts could create oversupply concerns.

How I'd Stage the Trade

If I were building this position, I'd diversify across time horizons rather than across stocks:

• Near-term (12-24 months): BE — Oracle contract revenue starts hitting income statements
• Mid-term (24-48 months): VRT, ETN, and other data center components (covered in the next post)
• Long-term (48+ months): OKLO + CCJ — SMR licensing and uranium demand inflection

Spreading by horizon, not concentrating in one name, is the rational way to ride this cycle.