Emerald AI, a startup focused on optimizing data centers for power grids, has raised US$24.5 million in seed funding.

The funding round was led by Radical Ventures, with contributions from Nvidia, AMPLO, and several individual investors, including John Kerry, John Doerr, Jeff Dean, and Fei-Fei Li.

The company, founded by physicist Varun Sivaram, aims to address the increasing energy demands of AI workloads.

Its software adjusts computational loads to align with regional grid needs, reducing stress during peak demand and limiting the need for new grid infrastructure.

Emerald AI’s solution integrates with Nvidia chips and data center controls, allowing real-time adjustments in AI workloads.

🔗 Source: Axios

🧠 Food for thought

1️⃣ AI’s growing energy footprint creates a new category of grid solutions

Data centers globally already consume approximately 200 terawatt-hours annually, equivalent to 1% of global electricity use, with AI dramatically accelerating this demand1.

The scale of this challenge is significant: AI-related electricity consumption is projected to grow by 50% annually through 2030, with data centers expected to account for over 3% of global energy demand by decade’s end2.

In North America alone, data center power requirements nearly doubled from 2,688 megawatts in 2022 to 5,341 megawatts in 2023 specifically due to AI workloads3.

This surge is creating infrastructure bottlenecks that limit AI expansion, as Emerald AI’s founder noted about interconnection waits stretching 5-10 years, making grid-responsive solutions increasingly valuable.

Emerald AI’s approach differs from traditional efficiency measures by making data centers active participants in grid management rather than just focusing on consumption optimization.

2️⃣ Grid-responsive computing builds on proven efficiency innovations

Emerald AI’s technology represents an evolution of earlier AI-based energy optimization efforts, such as DeepMind’s implementation at Google data centers that reduced cooling energy by 40% and improved overall power usage effectiveness by 15%4.

The recent Phoenix field test showing a 25% reduction in power consumption during grid stress events demonstrates how computational workload shifting can complement traditional efficiency measures.