RU

LG bets on sodium: a battery and a transformer for AI racks

Published: 2026-11-18 · Author: AI Release · @ai_releaseorg
LG bets on sodium: a battery and a transformer for AI racks

AI hit the wall not at the chip but at the substation. Every new GPU cluster slams the grid like a factory switching on next door — and that, not chip supply, decides how many racks ever get powered.

LG Energy Solution brought a two-part answer to RE+ 2026 in Las Vegas: a sodium-ion battery plus a solid-state transformer, SST. The booth ran November 17 to 19. Sounds like energy news, not AI news. It is AI news: without power, no model trains, no matter how many accelerators you stack.

What they actually built

SST is not the iron with windings everyone knows. Inside sit power semiconductors, a high-frequency transformer and control electronics. A conventional transformer changes voltage and stops there. SST steers the power flow: how much, where, and in which direction, across AC and DC alike.

The battery adds a buffer. Data center load does not grow smoothly — it jumps: training starts, training stalls, inference arrives in waves. The grid hates it. The operator pays for the peaks, and sometimes simply never gets the capacity it asked for.

Why this is a bet, not a footnote

Here is the uncomfortable truth: AI's bottleneck has moved from hardware to infrastructure. A rack of accelerators costs more than a year of its electricity, but it only gets switched on if a transformer, a cable and a permit are already there. Whoever sells not chips but the ability to power them sells the entry ticket to AI.

LG grew up on lithium-ion cells for cars. Sodium is an attempt to break the dependence on lithium for raw material and logistics. Paired with SST it stops being a battery and becomes a system: storage plus an actively controlled converter. For a data center that matters because the block can sit close to the load instead of behind one shared substation.

What the announcement leaves out

Capacity. Price. Customer. Timeline. No contracts, no energy figures. A booth at a trade show is not a shipment to a site. And yes, every battery maker tells an AI story today — it is the most convenient demand on earth.

So we wait, but we do not look away. Direction beats demo. If AI keeps eating electricity at this rate, the winner will be the one selling data centers not just storage but control over the flow. A transformer with electronics stops being passive iron — it gets behaviour, which means firmware, updates and vulnerabilities.

The irony: the hype around models walks right past this box. But without the box, the models never boot. While OpenAI argues about benchmarks, Korea argues about how to deliver a megawatt to a rack.

💬 What to argue about

Sodium versus lithium: cheaper raw material or worse energy density — which matters more in a data center where container space is finite?

Is SST a genuine replacement for the conventional transformer or a trade-show niche: who pays first for steerable power, the data center operator or the utility?

LG shows a system with no capacity, no price and no customer. Caution or theatre — and what would have to appear in the press release for you to believe it?

Prediction: Within a year the same booth will feature a pilot with a named data center operator — or the pairing quietly stays a showpiece.

Disclaimer: not officially confirmed: single source, trade outlet The Elec; capacity, price and timeline undisclosed.

The Elec

🤖 AI summary
  • Date: 2026-11-18
  • Source: www.thelec.kr
  • Tags: LG, Sodium, ЦОД
  • Key facts:
    • AI hit the wall not at the chip but at the substation.
    • Every new GPU cluster slams the grid like a factory switching on next door — and that, not chip supply, decides how many racks ever get powered.
    • LG Energy Solution brought a two-part answer to RE+ 2026 in Las Vegas: a sodium-ion battery plus a solid-state transformer, SST.
    • The booth ran November 17 to 19.
    • Sounds like energy news, not AI news.
    • It is AI news: without power, no model trains, no matter how many accelerators you stack.
    • SST is not the iron with windings everyone knows.
    • Inside sit power semiconductors, a high-frequency transformer and control electronics.
#LG#Sodium#ЦОД
← PreviousMemory for Long-Running AI Agents: Runtime, Not PromptNext →The Fed Warns of a Possible Bank Run — Supercharged by AI

Source: www.thelec.kr · post in Telegram