The rise of flexible data centers

In this episode of Catalyst, host Shayle Kann revisits Varun Sivaram, founder of Emerald AI, to explore the rapidly evolving landscape of data center flexibility in response to the massive energy demands driven by AI. Since their last conversation in 2025, data centers have become an even more dominant force on the grid—projected to account for 94% of PJM’s peak load growth and up to 17% of U.S. electricity by 2030. The episode unpacks how data centers, traditionally seen as inflexible loads, can become flexible assets through compute orchestration and behind-the-meter resources. Sivaram argues that the real bottleneck isn’t on the compute side—where companies like Google and Anthropic are already offering flexible inference tiers—but on the utility side, where differentiated service tiers (like faster interconnection or larger capacity for flexible loads) are still absent. The discussion highlights how workload flexibility, when combined with on-site generation and storage, can create a dynamic 'mini dispatch curve' that optimizes grid performance, reduces costs, and accelerates AI adoption—all while maintaining reliability. Emerald AI’s upcoming 100-megawatt commercial-scale flexible AI factory, set to launch in 2026, is presented as a pivotal test case for this vision. Key takeaways include: 1) Data center flexibility is no longer theoretical—it’s essential for grid affordability, reliability, and speed of deployment. 2) The most powerful incentive for flexibility isn’t cost savings, but faster and larger grid connections. 3) Workload flexibility is the cheapest form of grid support and should be prioritized before relying on behind-the-meter generators. 4) A unified orchestration platform is needed to coordinate multiple stakeholders—from utilities and data center operators to cloud providers and end users. 5) The future lies in hybrid AI factories that integrate compute flexibility with on-site energy resources, creating true grid-responsive infrastructure. The episode ends on an optimistic note, suggesting that with the right policy and market incentives, data centers could become the hero of the energy transition rather than the villain.