Title: Embedding Intelligence at the Edge of the Grid
Subtitle: A Conversation With Marissa Hummon of Utilidata
Resource URL: https://www.volts.wtf/p/embedding-intelligence-at-the-edge
Publication Date: 2025-05-16
Format Type: Podcast
Reading Time: 58 minutes
Contributors: Marissa Hummon;David Roberts;
Source: Volts
Keywords: [Grid Edge Intelligence, Edge Computing, Artificial Intelligence, Data Privacy, Electric Grid Modernization]
Job Profiles: Chief Sustainability Officer (CSO);Supply Chain Analyst;Renewable Energy Manager;Product Manager;Chief Technology Officer (CTO);
Synopsis: In this podcast, host David Roberts speaks with Utilidata Chief Technology Officer Marissa Hummon about embedding AI-powered edge computing into electricity meters to deliver real-time grid visibility, enhance reliability, and enable distributed energy management.
Takeaways: [Utilities can significantly enhance real-time visibility and control at the grid edge by utilizing embedded AI modules in meters and transformers., Edge computing allows for localized analysis and rapid interventions, enabling utilities to address grid reliability issues without central system delays., Robust privacy and data governance frameworks, including encryption and granular permissions, support the responsible use of high-frequency energy data., Distributed intelligence lets utilities manage distributed energy resources and virtual power plants more efficiently, reducing unnecessary infrastructure upgrades., Similar edge AI techniques can boost compute utilization and energy efficiency in data centers by enabling precise, real-time power management.]
Summary: In this conversation between David Roberts, writer at Volts, and Marissa Hummon, Chief Technology Officer at Utilidata, the discussion centers on the collaboration between Utilidata and NVIDIA to embed AI-driven, credit-card-sized GPU computers into electricity meters. The initiative addresses the pressing need for real-time, granular data at the grid's edge, where distributed energy resources such as electric vehicles and solar panels are proliferating. The speakers highlight that traditional utility infrastructure offers limited visibility and slow response capabilities, making advanced edge computing essential for modern grid management.

Marissa Hummon explains the technical evolution from single-purpose embedded chips to fully programmable, reprogrammable modules capable of running sophisticated applications and securely processing vast quantities of waveform data—up to 32,000 measurements per second. This innovation enables not only precise monitoring of local grid conditions but also immediate intervention, such as adjusting voltage or managing EV charging loads in real-time, all while maintaining a small, energy-efficient hardware footprint. The technology’s adaptability opens the door for extensive third-party application development, further expanding potential use cases.

The podcast addresses the critical issues of data privacy and security, acknowledging public concerns about sensitive household information. Hummon stresses that utility data governance, state regulations, and the platform architecture itself provide robust controls, including encrypted data at rest and in transit, as well as granular permissioning to limit access to personally identifiable information. This approach supports both consumer trust and compliance with regulatory frameworks.

Another major theme is how this distributed intelligence enables advanced grid management strategies, such as orchestrating virtual power plants and conducting localized interventions without the need to route data through central control rooms. Hummon contrasts the emerging reality of automated, micro-event-driven grid management with the often-overestimated sophistication of today’s grid, which remains surprisingly analog and reliant on human intervention. Field pilots and deployments suggest that utilities are increasingly embracing these distributed solutions for their operational and strategic benefits.

The conversation concludes by drawing parallels between grid-edge intelligence and data center management, discussing the potential for analogous efficiencies in both domains. Hummon notes that data center operators, like utilities, often overprovision power for peak scenarios and could gain up to 30% greater compute utilization through localized AI-driven controls. The podcast closes by reflecting on regulatory, risk, and privacy challenges, with optimism that these AI advancements will accelerate digitization and efficiency throughout critical infrastructure.
Content: ## Embedding Intelligence at the Edge of the Grid

### Introduction
David Roberts, host of Volts, discusses the ongoing transformation in utility infrastructure prompted by the growing volume of distributed energy devices connecting at the grid’s distribution level. Utilities lack detailed, real-time visibility into these devices, making reliable management increasingly complex. This episode features Marissa Hummon, Chief Technology Officer at Utilidata, who details their partnership with NVIDIA to develop a compact, AI-driven computing module for integration with electricity meters and transformers.

### Technical Foundations and Innovation
Utilidata collaborated with NVIDIA to create a programmable module—smaller than a credit card—that combines a GPU, CPU, memory, and proprietary platform software. This module is specifically engineered to withstand grid environments, addressing security, durability, and low maintenance requirements. Unlike traditional single-purpose embedded chips, the new module functions as a flexible computer capable of executing and updating diverse applications over time. Hummon underscores that the platform leverages NVIDIA’s GPU technology to deliver high computational efficiency while minimizing power consumption—targeting as low as five to six watts per device—an advancement not available five years ago.

### Data Acquisition and Actionable Insights
The AI module is designed to process incoming data streams from grid-edge devices in near real-time, achieving up to 32,000 voltage and current measurements per second. This high-resolution data allows the identification of subtle grid disturbances, such as failing transformers or hazardous power quality anomalies, which would otherwise go undetected. Local data processing enables rapid event recognition and response, either autonomously or by communicating concise alerts to centralized grid operators. The system can automate responses, such as curtailing EV charging when transformer loading approaches critical limits, provided utilities enable such interventions. This creates an agile, event-driven platform for real-time grid management.

### Software, Platform Design, and Security
Utilidata’s approach emphasizes the deployment of an open platform that supports both proprietary and third-party applications. The module’s software stack includes an operating system, libraries, and service frameworks, establishing secure foundations for developers to build, customize, and manage grid applications. Hummon addresses concerns regarding AI terminology, clarifying that the technology here primarily refers to data-driven modeling, pattern recognition, and local inference—not the generalized artificial intelligence seen in large language models. Strict data encryption and permission structures are integral. The architecture supports compliance with varying state privacy regulations and enables privacy-preserving data processing, allowing sensitive insights to remain on local devices while only essential, non-personal events reach utilities.

### Grid Operations, Distributed Resources, and Data Centers
Deployment of edge intelligence fundamentally alters how utilities manage distributed resources and infrastructure. The resulting real-time visibility lets utilities identify critical hotspots, target upgrades, and coordinate resources such as virtual power plants with greater precision and effectiveness. This reduces unnecessary capital spending and supports localized energy balancing. Hummon also draws parallels with data center power provision: just as inefficient visibility leads to overprovisioning on the grid, similar dynamics affect data centers' internal distribution. Embedding AI-enabled modules at the rack level could increase compute utilization by up to 30%—making both grid and data center power delivery more efficient.

### Challenges, Privacy, and Adoption
The speakers discuss ongoing regulatory and adoption challenges, particularly utility concerns regarding reliability, risk, and data misuse. Privacy protections are robust and highly regulated, limiting data sharing and requiring customer-choice in most jurisdictions. Field deployments indicate utilities are increasingly accepting of these technologies, especially as benefits to reliability and efficiency become clearer. Hummon suggests that utilities have moved from skepticism to active engagement, and similarly, hyperscale data center operators are recognizing the value of maximizing existing capacity through intelligent management. Ultimately, the innovation represents a step toward a digitized, automated, and more resilient energy infrastructure.