Computer Vision, Grid Inspection, and the AI-Ready Utility

In this episode of our Insights from Tech Leaders series, Marin Butori of Dedale Intelligence sits down with Vikhyat Chaudhry, co-founder, Chief Technology Officer and Chief Operating Officer of Buzz Solutions, to talk about what it takes to put artificial intelligence to work on the power grid.

Chaudhry has spent 13 or 14 plus years at the intersection of artificial intelligence and energy, and led machine learning and AI teams at Cisco Systems before founding Buzz Solutions out of Stanford University in 2017. The conversation covers how a computer vision platform for utility inspection got built, sold and scaled, and what the next decade looks like for a grid being asked to carry more load than it was designed for.

🎥 Watch the full interview

‍

🎧 Listen to the podcast

Three Years of Data Before a Business

Most software companies build a product and then go looking for data. Buzz Solutions did it the other way around, and Chaudhry is clear that there was no alternative.

"During the company's first three years, we focused heavily on data because the right data is the main asset for an AI company."

Only after that came the models, and only after the models came the work of persuading an industry with genuine reasons for caution to put artificial intelligence anywhere near the power grid. It is a useful conversation for anyone looking at energy and utilities software, because it deals with the two things that decide outcomes in this segment and rarely appear in a pitch deck: where the training data comes from, and how long the buyer takes to say yes.

From Wind Turbines to Power Lines, in Two Weeks

Buzz Solutions launched in 2017 out of Stanford University. Chaudhry is an energy engineer by trade and studied power systems and energy engineering at Stanford, and the company started from a project in which he was flying drones to inspect vertical-axis wind turbines. Through Stanford's Launchpad course, which focuses on entrepreneurship for real-economy and societal change, he met his co-founder Kaitlyn, and the two began discussing how to apply AI and machine learning to the energy sector.

The original plan was wind-turbine inspection. It did not survive contact with customers. After speaking with around 35 utilities, the founders were told repeatedly that the technology belonged somewhere else.

"We pivoted within the first two weeks after the company's incorporation to the much larger power-line infrastructure market."

The timing matters. This was before the major California wildfires of 2018 that were sparked by grid failures. At the point Buzz Solutions chose it, inspection, monitoring and maintenance of power infrastructure was, in Chaudhry's words, a niche problem that subsequently became a major utility priority. The market found the company as much as the company found the market, which is worth noting for anyone sourcing in infrastructure software: the segment that looks small today can be reclassified by a single catastrophic event.

What the Platform Actually Does

Buzz Solutions is a software platform company selling to power companies and utilities. The core technology is visual intelligence, specifically computer-vision AI.

Utilities collect enormous volumes of imagery and video from their infrastructure using drones, helicopters, tablets, phones and fixed cameras. The platform ingests that material and answers three questions: what problems are occurring, where they are occurring, and what the utility can do about them. Coverage spans high-voltage transmission infrastructure, low-voltage distribution infrastructure, substations and, more recently, solar farms.

In practice that means detecting overheated electrical components, vegetation on a pole or wire that could create a fire hazard, and damaged components on poles or towers. The platform then prioritizes those findings so maintenance crews can be targeted efficiently and repairs completed before small problems become large ones.

The return on investment, as Chaudhry frames it, comes from saving time, costs and resources while helping prevent wildfires and power outages. That second half is why this category exists at all. An inspection platform that only saved money would be a nice-to-have. One that reduces the probability of a utility igniting a wildfire is a different conversation with a different budget holder.

The Bottleneck Was Never the Drones

Understanding the value requires understanding what the work looked like before. Some utilities, Chaudhry notes, were still walking along power lines with binoculars and looking up at poles and towers.

Over the last ten years, adoption of drones has grown substantially, with helicopters carrying cameras still used for long-range inspections along transmission corridors and drones becoming the norm for concentrated, granular inspections. That solved the collection problem and created a new one.

"The bottleneck was the data."

Utilities could collect millions of images annually, but lineworkers, field technicians and field engineers still had to review every image manually. At that volume the task becomes practically impossible, and it slows the path to the field for proactive or condition-based maintenance. Prolonged manual review also contributes to eye strain and increased human error, which is a quality problem layered on top of a throughput problem.

This is the structural point that generalizes well beyond utilities, and it appears repeatedly across the vertical software stacks Dedale Intelligence covers. Hardware adoption outruns the software needed to make the output usable, and the gap between the two is where the durable business sits.

Why Utilities Buy Slowly, and What Shortens the Cycle

Chaudhry is unusually even-handed about his customers' reputation for slow decision-making.

"Utilities are slow adopters for a valid reason. If they adopt technology that has not been fully vetted and it fails, the consequences can affect society, including hospitals and people's lives during a major power outage."

Add a regulated industry with rigorous validation and vendor-vetting processes, and long cycles stop looking like inertia and start looking like appropriate risk management.

The numbers tell the story of how that changed. Buzz Solutions' average software-as-a-service sales cycle was initially around 12 to 18 months. The company went through numerous requests for proposal, winning many of them. Successes with early enterprise customers including the New York Power Authority, Dominion Energy and American Electric Power then created a network effect, helped by a structural feature of the sector that most markets do not offer: utilities do not generally compete with one another, and they recommend solutions that work at conferences, trade shows, webinars and seminars.

The partner ecosystem did the rest of the work. Esri was a major partner as a geographic information system provider, alongside POWER Engineers, now part of WSP, and World Wide Technology, NVIDIA, Amazon Web Services and, to some extent, Microsoft. Those relationships brought Buzz Solutions into utility conversations it would otherwise have had to originate cold.

The result was an average sales cycle of approximately six to eight months, roughly half of where it started. Chaudhry wants it shorter still, with the software established as a standard part of the utility inspection process.

Understand how enterprise software categories mature inside regulated industries. Explore Dedale Intelligence's research on the US energy and utilities software market.

The Data Moat, and Why It Had to Come First

Buzz Solutions' training data came from a combination of proprietary collection and work with early customers and partners, including the Electric Power Research Institute and utilities whose innovation teams saw potential in the technology early.

The deliberate part was geography. West Coast utilities face different problems from utilities on the East Coast or in the Midwest, and the company set out to build a dataset covering the full range of conditions and infrastructure issues rather than the conditions in one territory.

"That investment created a proprietary data moat that we have continued to compound."

The sequencing was explicit. First, build the proprietary dataset so the algorithms could understand the problems and reach the required accuracy. Second, educate the industry and deploy the models in production. Buzz Solutions also developed playbooks explaining effective and ineffective approaches to adopting AI at scale and shared them, with the aim of helping establish industry standards. Publishing the method rather than guarding it is a defensible choice when the asset you actually own is the data, not the technique.

That moat shows up directly in the competitive dynamics. Drone manufacturers and service providers have tried to build similar solutions, but they specialize in drone operations rather than software or AI engineering. More importantly, utilities generally own the data collected by drone providers, and contractual restrictions may prevent those providers from using it to train models. Many are also regional, which limits them to datasets from particular geographies.

Rather than compete with them, Buzz Solutions partners with drone manufacturers such as Skydio, which collects the data while Buzz Solutions provides the backend analytics.

European competitors posed a different question. Chaudhry saw several AI-native rivals emerge from Europe, which he attributes to Europe adopting drone-inspection technology roughly three or four years earlier than the United States. But European and North American grids have different components and characteristics, and a focus on North American grid data is what made the difference.

The clearest proof point came in an RFP run by the New York Power Authority around 2021 or 2022. More than 100 vendors submitted bids, including GE, Siemens, ABB and several startups. Buzz Solutions came out on top.

What Chaudhry believes he was really selling in that process was not an inspection tool.

"What we built was not just an inspection tool but an intelligence platform supporting systems of record such as geographic information systems, work-order management systems, asset-management systems, and utility workflows."

That distinction, between a point tool and a layer that feeds the systems a utility already runs on, is what separates a product from a platform in most vertical software markets.

Who Owns the Data, and Who Owns the Knowledge

Data ownership is the question that stops enterprise AI deals, and Buzz Solutions answers it with a clean split. The utility owns the raw data. Buzz Solutions owns the knowledge extracted from that data through its algorithms.

Chaudhry explains it with an analogy that does more work than most contract language.

"The principle is similar to someone reading 10 library books, returning the books, and retaining the knowledge gained from them."

Utilities retain ownership of their raw imagery, and Buzz Solutions does not take that imagery and manipulate it independently. For customers who do not want to participate in the global pool of models at all, there is an opt-out: a separate model branch, so the utility retains its own version. That version may lag slightly behind the global model pool, and some utilities accept the trade.

Offering the opt-out and being honest about its cost is a more credible position than pretending there is no trade-off, and it is a structure worth studying for any AI vendor selling into a sector where customer data is both the product input and a regulated asset.

From Power Lines to Substations to Solar

Buzz Solutions began with high-voltage transmission lines, low-voltage distribution lines and the infrastructure between them. Building the model-training pipeline for those transmission and distribution segments turned out to be the expensive part, and it shortened everything that came after.

The company built an active-learning pipeline that takes feedback from subject-matter experts through a human-in-the-loop tool and feeds it back into training. Substations followed in 2023, and training those models took less time than transmission and distribution had. Solar came in 2024, after several customers requested it and found few alternatives in the market, and it took less time again.

Solar panels have little in common with insulators and transformers, but the company already had foundation models for thermal analysis. On solar farms, Buzz Solutions detects thermal anomalies caused by cracks, dirt or contamination that reduce power output, and identifies the affected panels so customers can act.

"Overall, the time required for product expansion has continued to decrease."

That decreasing marginal cost of each new asset class is the quiet compounding argument in this business model, and the reason the dataset investment looks better in year eight than it did in year three.

Geographic expansion follows similar logic. The dataset was designed to include regions with different anomaly types: corrosion on the East Coast, wildfire and vegetation risks on the West Coast, storms in the Southeast, plus data from utilities in Canada, South America and Europe. Organizing the dataset around problems rather than territories lets the models recognize relevant patterns wherever they appear.

When a grid does look significantly different from anything the models have seen, there is an implementation period of approximately four to six weeks to recalibrate. Crucially, the company does not start from zero.

"We might start at approximately 65% accuracy and reach around 90% or higher within a few weeks through retraining and recalibration."

The Constraint Is Not Chips, It Is Electrons

Chaudhry has written a book on the subject, Powering Intelligence: How the Future of AI Depends on the Power Grid, and his argument in this interview is compact.

"AI's biggest challenge is not the chips; it is the electrons."

Utilities are struggling to supply power to the data centers and AI factories coming online. Electric vehicles, renewable and distributed energy resources, extreme weather and supply-chain constraints affecting equipment such as transformers and turbines all add to the pressure.

Buzz Solutions does not claim to solve this. Chaudhry is precise about the boundary: the company does not help utilities unlock additional grid capacity, it helps them manage their existing capacity. Utilities may still hold substantial capacity in existing infrastructure, but that infrastructure is under stress and requires more inspection, monitoring and maintenance to keep delivering it.

The clearest illustration is Dominion Energy, which serves Data Center Alley in Northern Virginia. Buzz Solutions uses computer-vision AI to help it understand which towers and existing infrastructure can have their capacity maintained and managed, so the utility knows the condition of what it already owns before committing to build more.

That reframing is commercially significant. Understanding existing asset condition becomes a prerequisite for responding to load growth, which moves inspection software from an operational efficiency line item toward a capital planning input.

The customer base extends beyond large investor-owned utilities. Buzz Solutions also works with municipalities and cooperatives, which are generally regional and smaller than a utility such as Dominion Energy but face the same operational-efficiency, cost and wildfire risks, often with different objectives from their larger peers. The company is increasing its footprint in that segment.

The Grid as an Organism

Asked what an AI-ready utility looks like in ten years, Chaudhry starts with the pressure the sector is under. Utilities face data-center demand, load growth, extreme heat, severe storms, wildfires and power outages at a level of complexity that did not exist 25 or 30 years ago. AI is contributing to the load growth and, in his view, is also part of how the rest gets solved.

His mental model is biological. Generation plants are the heart pumping blood, transmission and distribution lines are the arteries and capillaries, and substations are the valves controlling the flow.

"The more AI and Internet of Things sensors utilities introduce, the more self-aware and autonomous the grid can become."

His conclusion is not that autonomy is desirable so much as that full manual operation is becoming impractical given the complexity involved. The grid will need enough intelligence to regulate itself in complicated scenarios, with AI supporting use cases across the system.

That is a directional view rather than a forecast, and it is worth reading alongside how AI is reshaping adjacent categories, including industrial software, where the same tension between automation and human judgment is playing out on different timelines.

Advice for Founders Selling Into Regulated Industries

Chaudhry's closing advice is the least glamorous part of the conversation and probably the most useful.

"Patience and persistence are essential when working with utilities. Silicon Valley operates in two- to four-week sprints, while utilities operate in years."

Founders need to internalize that difference and understand why it exists: utility projects involve policies, regulations, approvals, and consequences for people's lives, electricity rates, wildfires and power outages. His practical guidance is to identify a niche utility problem, stay focused on it, develop a solution, validate it early with utilities, and find channel partners who can make introductions.

The access points are more open than they used to be. Many utilities now run internal incubation programs and innovation teams looking for technologies to bring to business units, and EPRI runs programs through which startups can conduct utility pilots. The work is connecting with the right people and partners, then staying patient.

For investors, the same facts read differently. Long cycles and heavy validation are a barrier to entry as much as a barrier to growth. A vendor that has already cleared a 100-bidder RFP, compounded a multi-geography dataset and embedded itself in utility systems of record has accumulated advantages that a well-funded newcomer cannot simply buy.

Looking at energy and utilities software, or any vertical where AI meets regulated infrastructure? Talk to Dedale Intelligence.

This interview is part of our Insights from Tech Leaders series, where Dedale Intelligence sits down with senior executives, founders and investors across the global technology landscape to explore the trends shaping software markets. Every insight we publish comes from a human source. AI supports our process, it is never the origin of what we know.

About Buzz Solutions

Buzz Solutions is a software platform company founded in 2017 out of Stanford University, providing AI-powered visual intelligence to power companies and utilities. Its computer-vision platform analyzes inspection imagery across high-voltage transmission infrastructure, low-voltage distribution infrastructure, substations and solar farms, identifying and prioritizing the problems that need maintenance attention.