A software company filing on cryogenic-plus-liquid-metal-plus-thermal storage tells you how seriously the datacenter world takes its power problem. US11728677B2, granted to Microsoft Technology Licensing on August 15, 2023, claims an integrated storage system stacking three very different storage media.
The CPC tags are unusual and revealing: H02J 15/00 (storage of electrical energy in general), H01M 10/39 (cells operating at high temperature — the liquid-metal battery), H02J 9/06 (standby/emergency supply), and G06F 1/30 (power-supply protection for computing). That last tag is the tell — this is storage designed with datacenters in mind.
The architecture logic maps cleanly to the grid's long-duration problem. Thermal storage is cheap per kWh but slow; liquid-metal batteries offer long cycle life at grid scale; cryogenic storage adds another duration/response profile. Stacking them lets a system match different storage media to different parts of the demand curve — fast response from one, bulk energy from another.
The patent-versus-product caution is acute: this is a claim on an integrated architecture, almost certainly far ahead of any deployed Microsoft system. It fences a concept of layering storage types, not a shipping product. Reading it as "Microsoft has built this" would be exactly the error this desk warns against.
From patent to grid, the significance is who is filing. When hyperscalers start patenting layered long-duration storage, it signals that the binding constraint on datacenter growth is increasingly power and storage, not compute. The grid-storage IP landscape is being joined by the largest electricity buyers on Earth — and their bets, like this one, lean toward not betting on any single chemistry.