Catalyst with Shayle Kann

Episode: The Rise of Flexible Data Centers

Date: April 9, 2026

Host: Shayle Kann
Guest: Varun Sivaram, CEO of Emerald AI
Produced by Latitude Media

Episode Overview

This episode explores the rapidly evolving landscape of data center flexibility amid surging AI workloads and grid pressures. Shayle Kann reunites with Varun Sivaram—the CEO of Emerald AI, and now a business partner following Kann’s recent investment—to discuss recent breakthroughs, the economics and mechanics of flexible data centers, and how AI “factories” could help address pressing grid challenges without sacrificing growth or affordability.

Key Discussion Points & Insights

1. Data Center Impact on Grids: Recent Shifts

Timestamps: [03:22]–[07:47]

• Surge in Energy Demand: Since their last conversation in August 2025, data centers’ grid impact has come into stark focus.
  • “Data centers now account for 94% of PJM's projected peak load growth. And by 2030, EPRI forecasts that data centers could use up to 17% of America's power. All of these are incredible, insane statistics and they're reshaping the landscape of energy as we know it.” — Varun Sivaram [04:57]
• Consequences of Unchecked Growth: Building out the power system exclusively to meet this growth could trigger higher rates and slow AI industry expansion.
  • "224 gigawatts of peak load growth is between a quarter and a third of peak demand ... if we try and build our way out of this, ... we risk higher rates and slower AI growth. And that's good for nobody." — Varun Sivaram [05:39]
• Affordability Front and Center: The conversation about grid and data centers now pivots on affordability more than ever before [06:20].

2. The Promise and Challenge of Data Center Flexibility

Timestamps: [07:47]–[11:09]

• Schism Between Compute and Utility Flexibility: There’s a key disconnect between the nuanced service tiers on the compute side (offering different levels of reliability and latency) and the historically rigid service in electric utilities.
  • “There is this deep divide ... between the level of flexibility, service levels and tiers that the compute industry offers and the level ... that the utility and grid operator ecosystem offers ... And that schism is one of the reasons that this is such a hard problem to solve.” — Varun Sivaram [08:12]
• Flexibility as a Grid Solution: Data centers can tap into vast “stranded” power capacity if they can be flexible during rare grid peaks—turning a challenge into an opportunity.
  • “If these new AI factories … can be flexible just a little bit of the time, they can utilize this vast amount of unused capacity on the grid.” — Varun Sivaram [08:38]

3. Evolving Service Tiers and Demand Response

Timestamps: [11:09]–[16:47]

• Service Levels Emerging in AI: Companies like Google now offer priority and flex tiers for AI inference, hinting at service-level flexibility that could translate to power use.
  • “Earlier this week, Google announced its Flex and Priority inference tiers. … It's not just the price component, it is the service component. The service literally changes between those tiers.” — Varun Sivaram [09:55]
• On the Grid Side—Lagging: The electric grid remains stuck providing only “firm” (always-on) service, complicating efforts to unlock this latent flexibility.
• Demand Response Evolution: Traditional demand response mechanisms aren’t enough; what’s needed is connection acceleration or increased interconnect size in exchange for flexibility, which currently doesn’t widely exist [13:08].
• Economic Rationale (and limitations): Flexing for minor cost savings doesn’t pencil out, given the massive revenue per watt of AI compute. But being able to connect or expand faster would significantly outweigh potential lost compute revenue [16:47].
  • “If the benefit of [flexing] is purely a cost savings on your electricity bill… those numbers don't pencil… The economics of getting a data center connected larger or faster are orders of magnitude different.” — Shayle Kann [16:47]

4. Economics and Workloads: Where Flexibility Works

Timestamps: [17:52]–[25:22]

• The Logic Shifts with Incentives: As inference tokens become less lucrative relative to electricity costs, curtailing for price and grid events will make increasing sense—the dynamic is changing long-term [19:44].
• Workload Categorization: Real-world demonstrations show many AI workloads (training, inference, batch processing, background updates) are inherently flexible.
  • “We've now done these five demonstrations… And in each of these, we've tried to reenact real production grade actual workloads… we could achieve performance levels that real customers find acceptable, making sure not to throttle workloads that the customer labels as mission critical, while at the same time precisely meeting grid objectives.” — Varun Sivaram [21:12]
• Geo-shifting and Batching: Workloads can be shifted geographically or batched to adapt to grid events with little/no user impact.
  • “We showcased migrating AI workloads from one location, Virginia, to another, Chicago… inference queries got rerouted … you were able to precisely meet the Dominion grid's power constraint while utilizing capacity far away.” — Varun Sivaram [22:43]
• Opening the System: The crucial next step is to treat the grid itself as a part of the data center optimization system.

5. Complexity of the Stack: Who Needs to Coordinate?

Timestamps: [26:50]–[30:15]

• Multilateral Coordination Problem: Delivering flexibility at scale involves grid operators, data center owners/operators, cloud providers, customers, and sometimes additional third parties.
  • "It's a wickedly complicated multi-party problem. … the data center is not one monolithic entity. It comprises a lot of players." — Varun Sivaram [27:42]
• Incentives as the Linchpin: If the grid offers meaningful incentives for flexibility (accelerated connection, bigger capacity), other actors will coordinate.
  • “Everybody becomes much more willing to work together when there’s a real economic incentive. And it’s the grid that sets that incentive.” — Varun Sivaram [29:42]
• Emerald’s Approach: They aim to be the "easy button," building modules for every stack layer—from utility down to end user and on-site resources.

6. Behind-the-Meter Resources and Hybrid Flexibility

Timestamps: [30:15]–[39:16]

• Emerging Trend: More data centers are deploying “bridge power”—gas turbines, batteries, on-site renewables—to meet grid signals. But full islanding is not the optimal long-term solution.
  • “As data centers become, by the end of this decade, up to 17% of America's load ... it would be a catastrophe if data centers were entirely decoupled from the electricity system.” — Varun Sivaram [32:01]
• Hybrid AI Factories: The ideal model involves integrating workload flexibility with behind-the-meter resources for grid services and rapid scaling.
  • “With Nvidia … we made a major announcement … Nvidia has a reference architecture ... One element of it is DSX Flex, the capability to be flexible. And Emerald is a software partner that helps to operationalize that.” — Varun Sivaram [33:00]
• Unified Dispatch Curve Concept:
  • “From the grid's perspective, it's kind of like a little mini dispatch curve. ... If you take [workload flex] to be true, then workload flex is the cheapest thing you can do and you should do as much of it as you can ... Then if you need more ... you should then dispatch things that cost more money ... your generator ... your battery or fuel cells ... all those things come at a significantly higher cost.” — Shayle Kann [34:56]
  • But unlike the static grid dispatch curve, a data center’s is "complicated, dynamic, constantly changing," governed by both resource constraints and workload flexibility [37:07].

7. The Four “Birds” of Flexibility and The Path Forward

Timestamps: [40:01]–[42:23]

• The Four Birds with One Stone:
  1. Faster, higher-capacity data center grid connections
  2. Lower, more stable rates by avoiding overbuilds
  3. Enhanced grid reliability (helping respond to events)
  4. (New) Ability to charge and dispatch on-site storage more intelligently [40:01]
• Challenges Remain: Differentiated service tiers for power, complex stakeholder coordination, and the temptation for data centers to go off-grid.
• Proof Point Ahead: The partners are preparing the world’s first 100MW flexible commercial AI factory, designed to prove the concept at real scale by late 2026.
  • “We will put together the world’s first 100 megawatt commercial scale AI factory that is truly power flexible... it's going to be able to respond precisely to all of these grid needs, but at a commercial scale." — Varun Sivaram [41:33]

Notable Quotes & Memorable Moments

• On the scale of the issue:
  “224 gigawatts of peak load growth ... that’s a massive increase that data centers are going to be driving.” — Varun Sivaram [05:39]
• On the underlying opportunity:
  “The farsighted way of thinking about this is ... it would be a catastrophe if data centers were entirely decoupled from the electricity system because the system loses their biggest source of anchor tenant revenue and the most exciting engine of American economic growth.” — Varun Sivaram [32:17]
• On the potential for innovation:
  “If electric power utilities and grid operators offered a range of different service tiers... innovation would solve this problem.” — Varun Sivaram [15:30]
• On multilayered complexity:
  “The data center is not one monolithic entity. … it comprises a lot of players.” — Varun Sivaram [27:42]
• On the big breakthrough to come:
  “By late 2026 and 2027, this really takes off and it kind of solves all three of those problems.” — Varun Sivaram [42:16]

Episode Structure & Timestamps

| Segment | Timestamps | |------------------------------------|-------------------| | Market Overview & Recent Changes | 03:22 – 07:47 | | The Flexibility Opportunity | 07:47 – 11:09 | | Service Tiers: AI vs. Utilities | 11:09 – 16:47 | | Workloads & Real-World Demos | 17:52 – 25:22 | | Stack Complexity/Stakeholders | 26:50 – 30:15 | | Behind-the-Meter & Hybrids | 30:15 – 39:16 | | Synthesis & Future Outlook | 40:01 – 42:23 |

Final Takeaways

• Data center-driven electricity demand is rapidly redefining grid planning and affordability discussions.
• Unlocking flexibility—by leveraging compute orchestration and integrating on-site assets—could fundamentally solve grid strains and accelerate AI deployment.
• True progress will require new electric service tiers, innovative partnerships, and sophisticated orchestration platforms bridging all stakeholders.
• A commercial-scale demonstration of a flexible AI factory is on the horizon, aiming to validate these concepts at scale and pave the way for broader adoption.

For deeper dives, check out the full episode and related resources at Latitude Media.