Tom Nudell, CEO and Co-Founder of Piq Energy – Interview Series

Tom Nudell, CEO and Co-Founder of Piq Energy, is a power systems engineer and technology executive with deep experience in grid analytics, transmission planning, and energy infrastructure. Before founding Piq Energy, he spent nearly seven years at Smart Wires, where he progressed through several analytics leadership roles and ultimately served as Director of Product and Solution Analytics, leading interdisciplinary teams of transmission planning engineers, software engineers, and analysts working on advanced grid planning studies and automated analytical systems. Earlier, Nudell was a Postdoctoral Research Associate at MIT, where his work focused on resilience in interconnected energy systems, microgrid energy management, and transactive control. He also conducted research at North Carolina State University on locating disturbances in large-scale power networks using real-time measurements, physics-informed machine learning, control theory, and optimization.

Piq Energy is a grid planning software company focused on accelerating the connection of new energy projects, data centers, and other large electricity loads to the power grid. Its platform enables developers and utilities to bring complex grid studies in-house, automate engineering workflows, evaluate potential points of interconnection, and test numerous project and grid scenarios more rapidly. Piq is developing an agentic grid planning platform that combines power-system models, data, engineering tools, and AI-driven workflows in a unified environment, with applications spanning solar, wind, battery energy storage, and large-load development. The company says its technology manages more than 1,000 transmission grid models and has been used to complete more than 10,000 engineering workflows, as it seeks to reduce one of the biggest bottlenecks facing new energy and computing infrastructure: lengthy grid interconnection processes.

Your career has spanned physics-informed machine learning research, grid resilience work at MIT, and seven years building transmission-planning analytics at Smart Wires. How did those experiences lead you to identify the specific problem that became Piq Energy?

My PhD was in computational power systems where I focused on real-time wide-area monitoring and control. From there I pursued a postdoc at MIT on resilient power grids and dynamic market mechanisms. Following my postdoc, I joined Smart Wires, where I spent seven years founding and scaling the analytics team. My team had global power system engineering coverage across four continents, with expertise in power flow, market simulation, dynamics, EMT, and hardware-in-the-loop.

Within three months of joining Smart Wires, I realized that grid planning is essentially a massive and complex data science problem. I started dreaming about how to streamline and automate the most expensive parts of the process: the definition of analysis, coordination between different subject matter experts, and the building and validation of models. To improve this grid planning process, we developed a few key ideas. First was reproducibility – if your analysis is 100% reproducible then there should be zero marginal cost for another person/expert to re-run the analysis. This reproducibility is crucial to enable seamless collaboration between developers, utilities and other third parties on grid studies. Second was applying the concept that “all models are wrong, some are useful” and focusing on verification that we have a useful model for the problem at hand, rather than trying to build and manage the most “accurate” model for everything. (When we refer to “models” here we mean the grid network models, dynamic models of devices and plants, etc.).

These initial insights led me and Dio, my co-founder, to start dreaming of a sophisticated software platform to automate the work we were doing. At the same time, we were observing how the interconnection queue was becoming an enormous bottleneck for the grid. We were also seeing the increasing challenges to interconnect newer grid technologies including inverter-based resources (IBRs, including all new renewable projects) and grid enhancing technologies such as the SmartValve hardware that Smart Wires was selling. Dio and I realized that we knew how to solve these problems, but we needed to go start a software company to do so.

There are three pieces: First, my background in computational power systems and the experience running a global analytics team was the foundation. Second, the work we’d already done at Smart Wires was the proof that there was real tangible value to create. And finally, the macro climate – the scale of the challenge, which is also the scale of the opportunity – was the catalyst to make the jump to start Piq Energy.

Piq describes itself as an “agentic grid planning platform.” What makes the system agentic rather than simply an automation or analytics platform, and what tasks can its AI agents independently plan and execute?

Piq is accelerating interconnection. Technical analysis is currently the biggest bottleneck to interconnection, so we’re building an autonomous grid planning platform to accelerate these technical analyses.

There are levels of automation. Today, these levels of automation have not been explicitly defined for grid planning and interconnection, but you can imagine something analogous to SAE’s levels of automation for driving. I alluded to this idea in my 2018 paper published at the CIGRE World Congress [T. Nudell, et al. “On Developing Automated Tools for Reliability Planning”, CIGRE Paris C1-201, 2018.].

Now we have this amazing, general purpose technology in large language models, and with Piq we’re figuring out how to take advantage of agentic autonomy while still ensuring that the physics and engineering is done correctly, in a fully reproducible, auditable way every single time.

Today our agent configures models and scenarios, launches studies, crossreferences published reports, analyzes results, and writes new reports that are grounded back to the study data.

Grid studies rely on established engineering software, detailed network models, and strict physical constraints. How does Piq combine AI with these deterministic tools without allowing a probabilistic model to override the underlying power-system physics?

You’re right that grid studies rely on established software, detailed models, and known physics. Our platform enables agents to augment human expertise, informed with our data and context on what kinds of studies to run, how to configure scenarios, and then grounding analysis in concrete results data.

But then everything in between is scalable, deterministic automation. There is never new code written by agents and no unfamiliar physics simulators. This results in a much higher trust platform compared to other approaches because the work that an agent does is reproducible and auditable, and will be done the correct way each time. In other words, if a human engineer inspects the output, they should be able to replicate every step of the process and get the same results.

Which parts of a traditional interconnection or transmission-planning study can Piq automate today, and which decisions must still be reviewed and approved by a qualified engineer?

Today Piq gives developers a set of tools across the interconnection study process.

The first is high-level greenfield screening across large areas that we call regional screening analysis. It gives you a heat map of N-1 headroom capacity at every node across a region, so you can see where the grid actually has room before you commit to a location.

The second is deeper screening at a specific location – a POI study, or point screening. Those studies dive deep for one point of interconnection: all of the network constraints, the potential upgrades, and what those upgrades would cost for a project connecting there.

The third is queue management. This is where developers can play out the game theory of a cluster study, by  understanding the detailed cost allocation, and how that allocation shifts as the pool of generators changes. Projects resize, projects drop out, other developers take actions, and your cost allocation moves with them. That’s a dynamic problem, and today most developers are finding out about it after the fact rather than modeling it in advance.

Piq also automates the model building and validation process for dynamic plant models – the equipment and controls that represent the data centers, solar farms, and other players on the grid.

All of these models, results, and reports require a human on the loop who can guide inputs and review outputs. We are not claiming to have a full self-driving system yet, and I would not trust anyone who is claiming to do so.

Piq has completed more than 10,000 production studies. What has operating at that scale taught you about maintaining model accuracy, tracking changing grid conditions, and producing results that engineers can reproduce and audit?

The 10,000 figure is now outdated and since January (when this figure was released) we are now running 2,000 studies a month, and growing. This is a result of our users growing, and more users building internal systems that scale as well.

Operating at that rate teaches you that reproducibility isn’t a feature you add. It’s the constraint that determines whether you can operate at that rate at all. If a study can’t be reproduced, it can’t be audited, and if it can’t be audited nobody can act on it. If your studies are not actionable, you’re just generating volume, not value. Everything about how we built the platform follows from the idea of building useful models and running valuable studies.

Piq allows data center developers and other large-load customers to evaluate potential sites and connection scenarios before approaching a utility. How can these developer-led studies accelerate the process without creating conflicting assumptions or duplicating the utility’s official analysis?

The utility still needs to run their own official study, but our customers can’t wait around. Today,  it can take over a year to receive those official study results, and meanwhile our customers are sitting on hundreds of millions of dollars of investment. They have equipment to procure, land to acquire, and interconnection positions to secure across the country. We empower our customers to make these decisions ahead of the official utility study. We’ve also seen how our customers can influence the utility direction as they are equipped with insights about flexible interconnection solutions or alternative upgrades to connect additional capacity.

What we’re offering is a fast, efficient way to preview the future. Nobody has a crystal ball (and you don’t know who else you’re competing against in the queue), but if studies are affordable and fast, then exploring many scenarios helps our customers make better informed decisions with their capital. We also offer the ability for our customers to explore alternative upgrades to unlock existing capacity on their system and iterate hand in hand with the utilities as true partners.

California utilities have received requests for approximately 18.7 GW of new data center service capacity, although many proposed projects may never be built. Can AI help utilities distinguish credible projects from speculative or duplicated requests and prevent “phantom loads” from distorting long-term grid planning?

Speculation isn’t irrational. Given how long it takes to get a project studied and connected, anyone who can afford to enter speculative interconnection requests will. It’s the rational response to the interconnection process as it exists today. As we see it, if projects could get studied and connected as fast as they could get built, there’d be no incentive to speculate in the first place.

So the answer is not about using AI to identify speculation. Instead, the answer is using AI to actually improve the process – to more efficiently connect more load. This means using AI to both accelerate the interconnection study process itself and to find alternative mitigations, including grid enhancing technologies and flexible interconnections. Combined with regulatory changes, such as higher application costs or larger deposits and revised interconnection tariffs, AI can unlock a much more efficient system with much less speculation in it.

But in the absence of changes, the challenges with phantom loads will continue to compound. More phantom load means more scenarios to study, which makes long-term planning harder for utility planners, which makes the whole system slower. A slower system increases the incentive to speculate. This is exactly what we saw with the generator interconnection queues that spurred us to start Piq in 2023, and it is exactly the feedback loop we’re trying to break.

Piq says its work has helped identify nearly 1 GW of grid capacity through flexible solutions and grid-enhancing technologies. What types of operational changes or technologies are unlocking that capacity, and can you walk us through an example in which a project avoided a major grid upgrade?

Only three things can unlock capacity. They are all types of operational flexibility that were not available to the grid until recently. We call these three types of flexibility Point Flexibility, Aggregate Flexibility, and Network Flexibility.

Point flexibility is what changes at the point of interconnection, including solutions like bring your own generation (BYOG), net withdrawal limits, load curtailment/controllable loads, among others. Aggregate flexibility is distributed resources across the network like VPPs and DER aggregated to manage transmission constraints. And network flexibility is the grid parameters and configuration itself — advanced power flow control, topology optimization, and dynamic line rating. The important thing is that all three can be modeled, planned and eventually deployed to a control room.

The cleanest example is network flexibility. We studied a case where the binding constraint was a single element, and 70 MVAr of power flow control deployed directly at that constraint unlocked roughly 600 MW of additional capacity. No new transmission lines, no multi-year permitting, just a small network upgrade to unlock 600 MW of capacity.

Transmission grid capacity utilization is only around 35% under worst-case planning conditions. The grid is not full. Although the biggest challenge to connecting large loads today is network capacity, the grid is only constrained in specific places under specific conditions. Most of those constraints have a solution that doesn’t involve waiting a decade for new wires and that solution can be identified by better studying the grid.

There is a rapidly growing ecosystem for these three types of flexibility. Companies like Base Power, Voltus, Smart Wires, NewGrid and others are deploying real solutions at scale today. The only piece missing is the analytical framework to get these solutions into interconnection studies in the first place.

Engineering decisions involving the grid can affect reliability, electricity prices, and infrastructure investment for decades. What safeguards, validation processes, permissions, and audit trails are necessary before utilities can trust AI agents with increasingly complex planning workflows?

We 100% agree. In order to make sound engineering decisions every piece of automated work has to be auditable and reproducible. There needs to be a paper trail for every action, as you don’t want an AI running in a black box and going rogue.

Our audit trails are multifaceted. We keep the agent conversations, the agent actions, and the deterministic tool execution. Every substep and sub-workflow maps and manages all of its inputs and outputs, so everything is auditable end to end.

Permissions are also crucial. Our agents inherit permissions from the user they’re working with. An agent can’t see data or take actions the person it’s working for couldn’t see or take themselves.

This is related to the broader theme of data governance, and agents are changing the historical paradigms. Historically governance was about structured data, narrowly scoped inside one silo. With agents you have structured, semi-structured and unstructured data all moving across silos — which is exactly the unlock a platform like Piq promises, and exactly what makes governance harder. There’s far more data in play, and a wider range of people interacting with it. In a power systems organization that matters twice over: for competitively sensitive investment information, and for CEII. Thankfully there are smart people working on this, such as those involved in the IEEE Task Force on Data Governance for Power System Organizations.

Piq recently raised an oversubscribed $5 million seed round to expand its platform. Looking five years ahead, how do you expect AI to change grid planning, and will faster analysis be enough to keep pace with the electricity demands created by AI data centers, electrification, and industrial growth?

I believe using AI to accelerate the analysis is key to keeping up with electricity demand. The core takeaway is that automation at this scale is not just about doing the same work faster, but also enabling analysis that simply couldn’t be done if you had finite human bandwidth. Flexible interconnection, flexible transmission grids, grid-enhancing technologies — all of it unlocks latent capacity that already exists on the grid today. Combining that with accelerated interconnection studies, we believe we’ll be able to keep up with demand.

Thank you for the great interview, readers who wish to learn more should visit Piq Energy.