Lancium data center technology is used to develop Clean Campuses, which can separate large compute requirements into critical and flexible clusters. The flexible clusters tackle non-time critical loads, often powered directly by microgrids, and providing a Controllable Load Resource, which can demand shift to smooth the volatility of renewables? Hence we reviewed Lancium’s patents.
Lancium was founded in 2017, is headquartered in the Woodlands, Texas, with c80 employees, and has developed a grid-balancing technology, for use in AI data centers and in integrating renewables.
We came across Lancium as part of our deep-dive into AI data center sizing. Most recently, the 1.2GW Stargate data center in Abilene, Texas is using Lancium’s smart response technology to adjust power consumption dynamically, while Lancium also built the “clean campus”, breaking ground in 2022.
The Abilene Clean Campus builds on a prior Clean Campus at Fort Stockton, which has qualified as a Controllable Load Resource in ERCOT, as it can flex its loads, in order to provide stability to the grid and to Texas power markets.
We wanted to understand how Lancium is achieving this feat, as it speaks to a fascinating ongoing debate about whether AI data-centers can load shift (aka demand shift) or form part of VPPs.
The key goals in Lancium’s patents are to avoid curtailment on renewable assets such as solar and wind, while also avoiding transmission and distribution costs that can make up 40-60% of total end power prices.
This is achieved by separating a data-center campus into critical loads and flexible loads. The latter can be energized directly via a microgrid, to handle non-time critical tasks, during times of heavy renewables generation, or when power prices are low. But conversely, its demand can also be scaled back when grids are strained.
Interestingly, some of the patents also integrate the flexible, microgrid-powered side of the campus with charging, of electric vehicles.
The kinds of computing tasks that are envisaged for the flexible portion of the campus are blockchain mining, Monte Carlo simulations, batch-processing of financial transactions, graphics rendering, or oil and gas field simulation models.
A fascinating futuristic vision is presented in another patent, where various data-centers bid into auctions for compute, in real-time, according to the specific costs of power at their precise locations. It might be cloudy and windless in one data center location, while it is gusty and sunny in another.
Another fascinating question for the next five years is whether older generations of GPUs, once fully depreciated at data centers, will be separated out into these kinds of flexible facilities, while newer and higher-grade GPUs command the round-the-clock power needed for cutting-edge models.
Our main question mark over the patent library is whether it truly locks up the implementation of flexible data centers, for reasons that are explored in the data-file.
Overall, we see grids becoming more flexible, as the smart eco-system advances. This may allow more deployment of renewables, especially solar, but also lower overall load growth requirements.