Santa Clara Flexible Data Pilot
- Silicon Valley Power and Emerald AI launched a pilot in Santa Clara to demonstrate "flexible data centers" that adjust power use. - The project aims to keep computing active while shifting power loads to better utilise the local grid. - The pilot underscores that local power capacity, not lack of demand, is the primary constraint for AI infrastructure growth. (latitudemedia.com)
A data center normally pulls as much electricity as its servers need, minute by minute. In Santa Clara, Silicon Valley Power and Emerald AI are testing whether some of that demand can move around instead of forcing the grid to meet every spike instantly. (santaclaraca.gov) The pilot was announced April 21, 2026 by Silicon Valley Power, the City of Santa Clara’s municipal utility, and Emerald AI, a startup that builds software for “grid-responsive” data centers. The companies said the program will run with major Silicon Valley Power customers in Santa Clara. (siliconvalleypower.com) The basic idea is load shifting: keep the computing work going, but change when and where the biggest bursts of power happen. Emerald AI said its Conductor software can orchestrate artificial intelligence training, fine-tuning and inference workloads in real time based on grid conditions. (emeraldai.co) Silicon Valley Power said it will also use Emerald AI software to help manage and dispatch participating flexible data centers during limited periods when the local grid is tight. The companies said the test will measure whether that flexibility can improve reliability, hold down costs and make better use of existing wires and substations. (santaclaraca.gov) That matters in Santa Clara because the city already hosts one of the country’s biggest concentrations of data centers, including facilities tied to chip and cloud companies. The constraint is often not demand for more computing, but how quickly utilities can deliver enough local power capacity to new projects. (latitudemedia.com) American utilities and developers have spent the past year looking for ways to connect artificial intelligence computing faster without waiting for entirely new grid buildouts. In March, Nvidia and Emerald AI said they were working with power companies including AES, Constellation, Invenergy, NextEra Energy and Vistra on “flexible AI factories” designed to act as grid assets. (nvidianews.nvidia.com) Emerald AI publicly launched in July 2025 with a $24.5 million seed round led by Radical Ventures, with backing from NVentures, the venture arm of Nvidia, along with CRV, Neotribe and AMPLO. The company said then that it had already completed a first commercial demonstration of software that trims data-center power use during grid stress while preserving computing output. (prnewswire.com) Trade outlets covering the Santa Clara pilot said the test will examine whether flexible operations can support phased energization, meaning data centers could begin operating before every megawatt of permanent capacity is in place. That would turn software control into a tool for utility planning, not just a way to cut peak demand. (datacenterdynamics.com) If the Santa Clara test works, the pitch is straightforward: let data centers behave less like a fixed industrial load and more like a customer that can briefly ease off when the grid needs room. Silicon Valley Power and Emerald AI said the pilot is meant to show whether that trade can unlock more capacity for the next wave of artificial intelligence infrastructure. (publicpower.org)
