Betting That AI Can Ease AI Infrastructure Pains

Right now, money and attention is focused on building infrastructure for AI. What’s coming next, though, is AI for infrastructure: using the very same technology to make the power and industrial systems we’re building cheaper, better and faster.

Since its founding in 2016, venture investor Blue Bear Capital has allocated about $350 million across 40 startups that boost grid efficiency, increase performance of renewables and industrial assets, and better analyze energy data. The firm is having a moment today, as electricity prices rise and demand for power surges. 

So I traveled to Jackson Hole, Wyo., late last month—hardship duty—for Blue Bear’s CEO Summit, to hear its founders talk about technologies that could ease the pressure on the grid and prove the case for AI itself. 

Partner and founder Ernst Sack believes the gigawatts of power, tons of concrete and steel, and miles of fiber being laid at great expense today will ultimately “unlock decades of capital-light technology and innovation that will capitalize on that infrastructure.” 

Another way I’ve heard entrepreneurs describe the AI-for-infrastructure niche, or infratech, is that we can actually use AI to heal some of the environmental harms the resource-hungry AI boom is unleashing. It starts with making systems we’ve already built more efficient.

At a ski lodge in the nation’s largest coal-producing state, Blue Bear’s portfolio companies and investors expressed relief that their corner of clean-tech investing avoided last year’s “solar coaster,” where builders of renewables found themselves on the wrong side of President Donald Trump’s policies. It proved much less dangerous to provide the tech that eked out better returns on power assets of all kinds.

Jane Woodward, who manages a fund of early-stage climate funds, WovenEarth Ventures, told the audience that digital startups that improve the payoff on infrastructure have the opportunity to “deliver such a leap of value it does not matter who is president.”

Blue Bear portfolio companies tend to be an unusual combination of capital-light software makers whose founders are experts in hard assets such as the power grid and nuclear reactors. They use AI to parse data and imagery on power flows, the weather and physical materials like battery cells to make scarce resources such as water, labor and energy go further.

Corporate venture arms of power and tech companies such as NextEra Energy and energy giant Chevron both compete with and coinvest with Blue Bear. Their partners showed up in Jackson, too, to discuss why all forms of energy are in today and the buzz now is about “energy addition,” not “energy transition.”

The U.S. power grid is surprisingly inefficient, often running at 50% utilization because of the risk of wind or heat causing sagging power lines to hit trees or other objects, potentially tripping safety mechanisms or starting fires. Splight, which Blue Bear has invested in, uses AI to automate decisions within milliseconds to safely expand electricity transmission from, say, a wind, solar or battery farm, or a data center.

The extra power flowing onto the grid can help keep rates low and prevent blackouts. In a recent deployment, Splight used real-time signals to safely inject 412,448 megawatt-hours onto the grid that would have been unnecessarily curtailed if the utility had arbitrarily set cutoffs based on average conditions. That’s enough to power a small city for a year, and Splight expects to keep adding similar increases in energy production on an ongoing basis. 

Before they started using Splight, “these generators were told they could only connect to the grid at a fraction of their capacity,” or just 60%, co-founder Fernando Llaver, a former grid operator in Argentina, told us. Splight tools have a significant presence on power systems in Argentina and other grids across South America and Europe. U.S. utilities have been slower to adopt such grid-enhancing technologies, but recently regulators have started encouraging their adoption, and Splight is expanding in the U.S.

Utilities are also resurrecting shuttered nuclear plants to generate clean energy. The control rooms of the 50-year-old plants look like dated sci-fi films, with analog dials and big warning lights. 

To bring these plants into the modern age, Bradley Fox and Jerrold Vincent, 30-something engineers and former data scientists at the Palo Verde Generating Station in Arizona, founded Nuclearn and built the first nuclear-specific generative AI chatbot. It ingests and makes sense of records created on software from as long ago as the ’80s and ’90s. 

Nuclearn’s large language model is compliant with nuclear security protocols and is used to address compliance and other issues. Its software has now been adopted by over 80 nuclear reactors in the U.S., Canada and the Middle East, including the Tennessee Valley Authority, where Nuclearn installed the first AI graphics processing unit behind a firewall within a nuclear environment.

Of course, large companies are picking off these opportunities too—Tapestry, a Google X Moonshot initiative, has partnered with Chile’s national grid operator to model and plan power flows using AI, and it is now helping the busiest grid in the U.S. streamline the connection of new power. Siemens is an AI leader in automated grids and factories. Such companies could snap up promising infratech startups; smart-metering giant Itron bought a Blue Bear portfolio company, Urbint, last fall. 

An intriguing theme that kept surfacing at the summit is that AI-enabled infrastructure might be the new and improved version of the Internet of Things. In the old version of IoT, you needed human labor to install sensors everywhere, and those sensors might fail or fall out of date.

Today, autonomous robots or drones might diagnose signs of failure—say, at a refinery, pipeline or solar farm—just from collecting and processing images or detecting abnormal heat. This early warning could eliminate expenses for parts and labor when maintenance workers are turning over at a rate of 30% a year and “you’ve got operations and maintenance companies that have low single-digit margins,” said Nikhil Vadhavkar, co-founder and CEO of Blue Bear’s solar-monitoring portfolio company, Raptor Maps.

One of Blue Bear’s first investments, in 2017, was in solar data analytics platform Omnidian, which then had annual revenue of $300,000. Founder and CEO Mark Liffmann said sales neared $50 million last year and could reach $75 million this year, putting it on track for a future initial public offering or sale.

The U.S. is still going to be short of power for years, though the blame lies with lack of investment in infrastructure as much as with power demand for AI. My experience in Jackson makes me think the energy catch-up won’t be linear. Instead, there will be tipping points where capacity leaps forward because we’re using what we built more intelligently.

“AI is really the best chance we have to bridge that gap,” Kai-Philipp Kairies, co-founder of battery intelligence startup ACCURE, told us at dusk, before we walked through the snow for drinks at the Mangy Moose.