AI’s next bottleneck might not be chips. It might be getting the electricity there.

$Bloom Energy Corp(BE)$ jump caught my attention, but the more interesting story is happening underneath the price move.

Data-centre developers increasingly need power faster than traditional grid connections can deliver it. That creates an opening for onsite generation — and that’s where Bloom’s fuel-cell technology comes in.

Bloom is now pushing an 800V DC-native power architecture designed specifically around next-generation AI infrastructure. The company says its approach can reduce non-compute capital costs for a 1GW AI data centre by $3.6 billion, although that’s Bloom’s own analysis and actual economics will depend on the project. 

There are other signs that this is becoming a broader infrastructure theme. Morgan Stanley has estimated a potential U.S. power shortfall through 2028, while onsite power providers are increasingly being considered as a way around grid-connection delays. 

But here’s the part I find interesting:

Bloom doesn’t need to become the “next Nvidia” for this trade to work.

It simply needs AI data-centre operators to decide that waiting years for grid infrastructure isn’t an option.

At the same time, investors need to separate the structural power shortage from the valuation of individual companies. Bloom has also faced uncertainty around major projects, showing that even a strong industry theme doesn’t guarantee every contract or deployment happens on schedule. 

So I’m watching the power trade from a slightly different angle:

Not “Which AI company wins?”

But:

“Who gets paid when AI companies run out of electricity?”

That could become one of the more interesting second-order trades of the AI buildout.