Cleaning Up: Leadership in an Age of Climate Change Can We Stop Data Centres Breaking The Grid? Ep249: Varun Sivaram & Steve Smith
9 snips
Mar 19, 2026 Steve Smith, leader of National Grid Partners driving grid innovation, and Varun Sivaram, founder of Emerald AI focused on grid-friendly AI data centres, discuss how hyperscale compute can adapt to the electricity system. They cover a live trial showing software can throttle and shift GPU loads in seconds, the idea of flexible demand, speed-to-power vs cheap power, and plans to scale this approach commercially.
AI Snips
Chapters
Transcript
Episode notes
One Data Centre Can Be A Systemic Load
- Hyperscale AI campuses can reach gigawatt scale, a single site equating to a material fraction of national transmission peaks.
- Steve notes transmission is built for peak hours so most capacity sits idle most of the year, creating an opportunity for managed flexibility.
Sell Speed To Power Not Low Prices
- Prioritise speed to power over cheapest electricity when courting hyperscalers because they value rapid connection.
- Steve says UK should lean into fast interconnection using existing substations to attract AI investment despite higher prices.
Two Core Grid Problems Data Centres Create
- Grid challenges split into two: physical connection (substations, transformers) and time‑varying utilisation where transmission is built for peaks.
- Steve highlights transmission average utilisation ~30–35% so flexing demand during peaks unlocks spare capacity.

