What Energy Companies and Utilities Get Wrong When Scouting AI Startups

By Andrea Course

Source: American Public Power Association

In 2017, several years before the current stampede of AI investment kicked off, Shell invested in a small startup called Innowatts. It was a smart bet: Innowatts was building an AI-powered SaaS platform that uses smart meter data to deliver sophisticated power demand forecasting, and for energy companies and utilities, better forecasting means greater efficiency. That leads to a wide range of benefits, including reduced emissions and increased margins—exactly the kind of strategic value that corporate venture capital is supposed to deliver. Innowatts has since been acquired by GridX, and its SaaS platform has been integrated successfully into their suite of technologies for the utility industry.

I happened to be working for Shell at the time, and sat on the Innowatts board, giving me a front row seat for what happened next. As an investor, Shell had not only a seat on the board but direct access to the technology and had every reason to deploy it internally. But by the end of our investment, we had zero active projects with Innowatts. We weren't even a customer. 

The disconnect here wasn't a lack of talent or technology, but something more fundamental. A VP inside the company told me directly that he couldn’t help improve Innowatts' AI models, because they could then be used by their competitors. From his perspective, this made perfect sense; from a deployment perspective, it seemed to make the entire investment nearly pointless.

This story isn't unusual. Across my years evaluating and investing in AI startups at Shell Ventures and elsewhere, I've seen this pattern repeat itself in different forms. The technology works. The people are smart. And somehow, the relationship still doesn't produce the value that both sides expected. It's not because energy companies are blind or startups are naive—it's because the two operate in fundamentally different ways, and the friction between those operating models is more destructive than most people realize.

That friction shows up in three places.

The Timeline Trap

For a major energy company like Shell, making a $1 million investment requires the same due diligence as making a $1 billion investment. It’s a cycle of background checks, compliance reviews, and board approvals that typically takes 18 months to two years in total. This isn't bureaucracy for its own sake; when you're a huge, highly visible company, you genuinely need to verify that your money isn't ending up in the wrong hands.

But for a startup, 18 months is a lifetime. In that window, they may have pivoted their product, changed out most of their team, adopted two new generations of underlying technology, or simply run out of cash. The startup that a corporate VC begins evaluating in January may bear little resemblance to the one that finally clears due diligence the following year. We missed good opportunities because of this, repeatedly.

Procurement works on a similarly glacial timeline. Getting approved as a vendor to a major energy company can take months. Once you're on the list, invoices can take six months to get paid, and for a startup with a dozen employees and no revenue, that can be the difference between making payroll and not.

Photo credit: Craig Lovelidge

The Pilot to Nowhere

Energy companies love proofs of concept. They're low-risk, low-commitment, and they give internal stakeholders something concrete to evaluate. The problem is that they often go nowhere.

I've watched startups get trapped in cycles of pilot after pilot: free trials and limited deployments that generate enthusiasm and slide decks but never convert into contracts. This happens because startups are treated like traditional vendors and run through procurement processes designed for mature companies selling proven products. The startup, meanwhile, is operating on the assumption that a successful pilot will naturally lead to wider adoption. It almost never does.

Even when a startup successfully deploys its technology within one business unit, that rarely translates to company-wide adoption. Each division, each asset, each regional operation often functions as its own procurement ecosystem. For Innowatts, selling to another asset meant starting over from scratch, with 15, 20, or 30 new stakeholders to convince. Any startup with a financial model that projects deployment across all of their investor’s operations within two years is setting itself up for a painful correction.

Source: Domaintechnik at Ledl

The Data Problem

This challenge is specific to AI, and it's the most underappreciated one I've encountered. Energy companies produce and collect enormous quantities of data. That should be good news for AI startups, whose products depend on large, high-quality datasets. In practice, it's one of the biggest barriers to deployment.

While researching carbon accounting tools for Shell, for example, I came across more than 40 different ways of doing carbon accounting within the energy company, depending on whether you were talking to finance, ventures, or some other function. The data existed, but it lived in incompatible systems, in different formats, in some cases literally on handwritten cards and tickets. Much of it was proprietary.

This is where the tension became most acute for Innowatts. To build a truly transformative demand forecasting model, they needed access to as much operational data as their investors could provide. But sharing that data risked improving a product that Shell’s competitors could use. For a company whose margins depend on forecasting advantages, that's not paranoia. It’s a real strategic concern.

The uncomfortable truth is that investing in AI necessarily means investing in data governance. If your data isn't structured, aggregated, and accessible, no startup—no matter how brilliant—can deliver value. Every energy company has to deal with this, and most haven't.

What Startups Don't Understand

None of these problems is one-sided. Startups consistently underestimate what it takes to operate inside a large energy company, and that misunderstanding creates its own damage.

The most common blind spot is around acquisition. When a major energy company approaches a startup, the founders often see it as a path to being acquired. In reality, out of more than 100 companies in Shell’s venture portfolio during my time there, they acquired exactly one. Corporate VCs invest for two primary reasons: to grow the startup’s value and exit with a return, and to get access to new technology, new markets, and operational insights that inform their core business. Neither of those requires owning the startup.

The math illustrates why. If an energy company invests $1 million and gets a 20x return, that's a spectacular VC outcome. But for a multi-billion-dollar company like Shell, it barely registers. On the other hand, if that same $1 million investment gives access to a technology that improves their drilling speed by a factor of five, or increases production by a few percent, the strategic value runs into the billions. That's what corporate venture teams are actually optimizing for, and startups that don't understand this will misread the relationship from day one.

Startups also tend to underestimate the sheer length of enterprise sales cycles. If you're building your financial model around closing a contract in six months, and the reality is 12 to 18 months, you need to raise a larger round or you'll be scrambling for resources halfway through the engagement. Few founders, in my experience, fully grasp the implications of this. 

Source: NASA

Why Utilities are Different

Everything I've described so far was shaped by my own experience at some of the world’s largest energy companies, with their own massive in-house innovation teams. Shell alone has more than 400 of its own people working on digital innovation. And these companies essentially have their pick of startups, who benefit from an energy leader’s resources and expertise, as well as their conferred credibility: when you send an email from a Shell address, people tend to respond.

Utilities are a different story. Most don't have dedicated AI teams or corporate venture units, but they sit in what I think of as a sweet spot: large enough to have real, complex operational challenges that AI can address, but small enough to be nimble when it comes to engaging with a startup and implementing new tech. And right now, utilities are facing a convergence of pressures that makes AI adoption not just attractive but urgent.

Data center demand is easily the most pressing. If I were the CEO of a utility company, this is what would keep me up at night. The amount of computing power required by AI is growing so fast that tech companies are already beginning to invest in their own power generation, and when that happens at scale, utilities will lose market share to companies like Google and Amazon. That's not a hypothetical; it's already underway.

On top of that, the grid itself is changing. Today’s power grids are designed for the worst-case scenario—a bit like building an eight-lane highway to handle 30 minutes of rush hour traffic. AI tools can fundamentally change this, enabling utilities to model real-time demand with a precision that was previously impossible. The platform Innowatts developed, for instance, can determine which households have electric vehicles just from aggregated energy usage data. This enables granular load forecasting that makes grids far more efficient.

Other opportunities for AI in utilities are concrete and immediate: using AI to clean and structure legacy data for smarter systems; predictive maintenance that heads off transformer failures before they happen; physics-informed AI that models infrastructure behavior based on thermodynamic principles; simulation tools that can model grid impact from weather events or sudden demand spikes. Startups are building these right now, and utilities that move first will capture disproportionate value.

L: Utility innovation in action: Heimdall Power’s Neuron sensor was named one of TIME magazine’s Best Inventions of 2025. R: It takes as little as 10 seconds for a sensor to be installed by aerial drone. Source: Heimdall Power

How Utilities Can Get AI Investment Right

The companies that act on AI partnerships, instead of just researching them, share a common trait: they think in years, not quarters. Innovation doesn't deliver results immediately, and boards that evaluate AI investments on a quarterly basis will kill every promising initiative before it matures. This is one of the main reasons corporate venture units die: new leadership comes in, demands immediate returns, and pulls the plug.

For utilities serious about engaging with AI startups, the approach matters as much as the technology. Working through a well-structured accelerator program can solve several of the above problems at once. It shortens the timeline by serving as a trusted intermediary that has already vetted participating startups; it focuses the engagement on specific use cases rather than open-ended exploration; and it prepares startups for the realities of enterprise sales cycles (i.e., whatever you’re estimating for time and cost, double it).

RAVA, the Rice AI Venture Accelerator, where I currently serve as Interim Executive Director, uses this kind of technology-first model. Corporate partners collaborate directly with early-stage AI startups on their most pressing problems, and also get right-of-first-refusal investing rights into the startups they work with, along with the option to participate in an AI-focused venture fund. This approach combines the speed and focus of an accelerator with the upside potential of corporate VC, without requiring the 18-month due diligence cycle that makes direct investment so difficult to execute.

As for getting real value out of the resulting solutions, the best advice is to think of AI as a structural change, not a new tool layer. The value of AI comes when it's deployed at scale across the organization, not in a localized pilot that never learns from the broader system. If you're going to do this, go all in. Be intentional about your use cases, invest in your data infrastructure, and find partners who understand both sides of the relationship.

The window for utilities to get ahead of this is narrowing. Even when startup partnerships like these don't go exactly as planned, they almost always produce value: Innowatts went on to become one of the leading demand forecasting platforms in the energy industry, and the energy company gained industry partnerships and operational insights that justified the investment many times over. With this much potential value available, even getting it "wrong" still looks pretty good. But the utilities that get it right—and get there first—will be the ones that thrive as the energy landscape transforms around them.

--------------------------

Andrea Course spent a decade evaluating and investing in AI startups for the energy sector, including roles as Venture Principal at Shell Ventures and Schlumberger Technology Investments. She currently serves as the Interim Executive Director of Rice AI Venture Accelerator (RAVA) and is the founder and managing partner of Course Investments.