Energy
Emerald AI Showcases How Grid-Flexible AI Factories Unlock Latent Capacity
Objective
Emerald AI, an NVentures portfolio company and NVIDIA Inception program startup based in Washington D.C., and recently named to the TIME 100 Most Influential Companies list, has developed a solution aimed to solve the energy bottleneck for next-generation AI factories. Its Emerald Conductor platform transforms AI factories into power grid assets, capable of flexibly using power during periods of peak grid load and utilizing vast capacities of untapped power that already exists on grids today.
Customer
Emerald AI
Partner
Nebius, EPRI, Oracle Cloud Infrastructure,
Salt River Project, Arizona Public Service,
Digital Realty, and PJM Interconnection
Topic
NVIDIA Data Center / Cloud
Products
Key Takeaways
200 Simulated Grid Events
- Successfully met every requested power-reduction target
Up to 40% Power Reduction
- In power demand in under a minute while maintaining critical workloads
100% Alignment
- In alignment with National Grid and EPRI’s power targets, validating that high‑performance AI infrastructure can act as a precise, grid‑responsive asset
Underutilizing the Power Grid
The power grid is constantly working to meet global energy demands, from powering large-scale data centers to keeping the lights on in residential neighborhoods. Yet, energy usage spikes — leading to the need for backup generators to activate — might not be as common as you’d think.
“The power grid is like a large-scale freeway that only faces rush hour two times a month,” said Varun Sivaram, CEO and founder of Emerald AI. “That’s how underutilized our power grid is.”
Emerald AI’s mission is to develop a solution that allows AI factories to use existing Gity more effectively, enabling AI factories to connect faster without triggering surges in demand. As a result, 100 GW of power capacity on existing U.S. power grids could be harnessed immediately to deploy AI factories. Additionally, this technology has the potential to make power bills more affordable for energy consumers.
The Conductor Platform At Play
Across five demonstrations at commercial facilities from Arizona to the United Kingdom, the Emerald Conductor platform is proving that AI factories can be power-flexible grid assets.
In June 2025, Emerald AI published the results of its first demonstration—under EPRI’s DCFlex data center flexibility initiative — with NVIDIA, Oracle Cloud Infrastructure, Salt River Project, and Arizona Public Service.
The system responded to a real power surge that occurred on May 3, a particularly hot Phoenix, Arizona day driven by high air conditioning usage. Using the Emerald Conductor platform, the data center cluster immediately reduced energy consumption by 25% and maintained the ramp-down for three consecutive hours. All AI workloads remained within service level agreement (SLA) constraints during this period. This demonstrated that flexible data center power usage can protect mission-critical workloads while relieving grid strain when power demand is high.
After Phoenix, Emerald AI and NVIDIA collaborated on trials in Chicago, Virginia, Portland, and — most recently — London.
This five-day demonstration took place at Nebius new AI factory outside of London, United Kingdom. The trial used a 96-GPU NVIDIA Blackwell Ultra AI cluster combined with the Emerald Conductor platform to showcase how a power-flexible AI factory can operate and adjust to consumption demands within seconds.
Simulating real events — from national football match tea breaks to lightning strikes — the Conductor platform was able to:
- Successfully meet every requested power-reduction target for more than 200 simulated grid events.
- Reduce power demand by up to 40% in under a minute while maintaining critical workloads.
- Support load‑reduction requests from National Grid, the electric utility, for up to 10 hours, demonstrating sustained flexibility during prolonged grid stress.
- Shed approximately 30% of the data center’s load within 30 seconds, enabling rapid “emergency curtailment” to help maintain grid resilience.
- Deliver 100% alignment with National Grid and EPRI’s power targets, validating that high‑performance AI infrastructure can act as a precise, grid‑responsive asset.
"Together with NVIDIA, we're proving that AI compute can be power-flexible, a paradigm shift with a massive prize: unlocking 100 GW of untapped power, grid capacity, and resolving AI's energy bottleneck, while promoting reliable and clean power grids."
Varun Sivaram
CEO and Founder
Emerald AI Announcements Ahead of AI Factory Research Center
Emerald AI has been gaining significant recognition for its AI factory energy grid solution.
The company was recently named on the Bloomberg list of Pioneers as well as the TIME 100 Most Influential Companies, alongside NVIDIA.
Additionally, Emerald AI recently announced an NVIDIA and Silicon Valley Power collaboration for a pilot program to flex an NVIDIA data center in Santa Clara in order to unlock additional power in the Silicon Valley region. The team visited the White House in Washington, D.C., to share this development as a showcase for how flexible AI factories could unlock massive power capacity across the nation while keeping local energy bills affordable.
The next step for this AI-driven energy intelligence layer is large-scale commercial deployment.
This year, NVIDIA will be launching the AI Factory Research Center in Virginia — the world’s first power-flexible AI factory — in partnership with Emerald AI, Digital Realty, EPRI, and PJM Interconnection.
Learn more about Aurora, the first commercially available, power-flexible AI factory launching soon.