We didn't.
Data centers are eating the world. But not the ones you think. The AI giants — Meta, Amazon, Google — they’re not just buying GPUs. They’re buying batteries. High-power cylindrical batteries, the kind that keep your servers alive when the grid hiccups. And they’re running out.
A report from Serenity, the industry analysis outfit, dropped a bomb this week: the supply of high-power cylindrical cells — specifically those used in backup battery units (BBUs) for data centers — is tightening. Hard. And the only two suppliers with the right stuff? Samsung SDI and Panasonic Energy.
— Root: The "supply shortage" narrative is real, but it’s not about lithium anymore. It’s about dedicated high-rate production lines. The market is screaming for cells that can dump 10C in under a minute. The automakers are swimming in LFP surplus. The BBU world is parched.
Context: Why Now?
For years, data center backup power meant lead-acid. Heavy. Bulky. Slow. Then came lithium-ion, smaller and faster. But the real shift? AI training clusters. Nvidia’s H100 draws 700W per GPU. A rack of 8 draws 5.6kW peak. When a node faults, the power drop is instant. You need a battery that can respond in milliseconds, not seconds.
That’s the BBU. It’s not your phone battery. It’s a high-power cylindrical cell — often 18650 or 21700 format — engineered for burst discharge. The same chemistry Tesla uses in its cars, but tuned for raw power density, not range.
And here’s the kicker: data center CapEx is exploding. McKinsey estimates hyperscaler spending will hit $300B by 2027. Every new build needs BBUs. But the supply chain? It’s still calibrated for the slow growth of 2020.
Core: The Numbers Don’t Lie
Let’s get technical. A standard EV cell has a C-rate of 1-3C. A BBU cell needs 10-15C. That requires thicker electrodes, specialized electrolyte additives, and ultra-precision winding. Not every factory can switch overnight. Samsung SDI and Panasonic have been refining this for years — they supply power tools, e-bikes, and now data centers.
But the capacity is finite. Serenity’s anonymous sources — "industry insiders" — claim that Samsung and Panasonic are already allocating 80% of their high-power cylindrical output to existing contracts. New orders? Lead times stretched to 12 months.
Here’s the math: - A typical hyperscale data center (100MW IT load) needs ~50 MWh of BBU capacity. - That’s about 2.5 million 21700 cells per facility. - With 20 new hyperscale builds per year globally, demand is 50 million cells annually from this segment alone.
Now add colocation, edge, and enterprise. The gap is real.
Contrarian: The TAM Trap
Wait — don’t FOMO into buying battery stocks yet. Serenity itself warns: "Not all shortages equal a giant total addressable market." They’re right.
The total megawatt-hours for BBU is a rounding error next to EVs. By 2030, EV batteries will be 3 TWh per year. BBU might hit 50 GWh. That’s 1.6%. The opportunity is pricing power, not volume.
Samsung SDI’s battery segment generated $12B in 2023. BBU is a sliver. The shortage will lift margins, but it’s not a moonshot. It’s a cyclical spike.
And here’s the blind spot: Chinese manufacturers are coming. EVE Energy and Lishen already make high-rate cells for power tools. They’re ramping up BBU-certified lines. The certification cycle for data centers is 12-18 months. By 2026, the shortage flips to glut.
— Root: The real alpha isn’t in the cell makers. It’s in the power architecture firms — Vertiv, Eaton, Delta — that integrate BBUs with smart UPS systems. They’re the ones turning a component shortage into a system upgrade narrative.
Takeaway: The Clock is Ticking
So what do we do? The next six months are a window. If you’re a trader, buy Samsung SDI on pullbacks. If you’re an operator, lock in Panasonic supply agreements now. But don’t confuse a temporary shortfall with a structural shift.
We didn’t see the BBU bottleneck coming. But now that it’s here, the party doesn’t last forever. The next wave of capacity is already being built.
— Root: The real question — will AI’s thirst for compute outrun the battery makers’ ability to scale? Or will the grid itself collapse first?
That’s the signal to track.