AI’s clean – A half-day Columbia event brought geothermal, green hydrogen, AI-driven demand flexibility, and clean-firm power into the same room—posing a blunt question: how do you fuel the AI boom fast enough to meet climate goals and keep the grid working? Speakers point
For a growing number of people watching the AI boom. the numbers are starting to feel personal: power demand is surging. grid operators are under pressure. and climate goals are not waiting in line. At Columbia’s half-day Columbia Climate Business and Investment Conference—organized by the Tamer Institute for Social Enterprise and Climate Change and the Columbia Climate School—the urgency landed early. and it didn’t let go.
In opening remarks. Bruce Usher—co-director of the Tamer Institute and professor of professional practice at Columbia Business School and Columbia Climate School—framed the problem as three problems braided together: environmental. economic. and strategic. “The challenge [of meeting demand] is not only an environmental one. but also an economic one and strategic one. ” Usher said. “The race to power the next generation of growth will help shape industries. capital flows. competitiveness and long-term resilience across the global economy.”.
That set the tone for the conference’s central argument: the next wave of growth depends on electricity that isn’t only clean, but also stable, reliable, and fast to build.
A geothermal bet moves closer to the mainstream
On one panel. the message was that enhanced geothermal systems—EGS—are edging from a niche idea toward serious AI infrastructure. The geothermal discussion featured Vijay Vaitheeswaran. global energy and climate innovation editor at the Economist. and James Blake. head of capital markets of Fervo Energy.
EGS. as described in the session. uses techniques similar to those in oil and gas drilling to tap underground reservoirs of steam and hot water to generate electricity. Blake said the United States has tapped only four gigawatts (GW) of geothermal power out of a potential 100 GW of clean. round-the-clock geothermal capacity.
The market signal, too, is getting louder. Blake pointed to investor interest: $2.05 billion raised to date, plus an IPO roadshow launched on May 4 of this year. An update in the conference materials added that Fervo’s initial public offering netted the Houston-based company about $1.9 billion and valued it at roughly $7.7 billion.
Green hydrogen faces a tougher clock
If geothermal felt like a straight line toward dispatchable clean power. green hydrogen was portrayed as a more complicated cousin—promising. but under sharper constraints. In the hydrogen discussion. Jacob Susman. former senior vice president of development at Electric Hydrogen. spoke with moderator Dan Esposito. associate professor of chemical engineering at Columbia.
Electric Hydrogen’s focus is on proton exchange membrane (PEM) electrolyzers designed to pair with intermittent power sources like wind and solar. In the session’s explanation. PEM electrolyzers respond quickly and are meant to avoid large batteries by using excess renewable electricity to split water molecules and convert them into storable green hydrogen.
But the conversation tightened around timing and policy. Under the One Big Beautiful Bill Act. Susman’s presentation highlighted. the construction deadline to qualify for the Inflation Reduction Act’s clean hydrogen production tax credit was shortened by five years—now January 1. 2028 instead of January 1. 2033.
Emerald AI goes after peaks, not just generation
Another path showed up on the Emerald AI demand response panel. moderated by Andrea Turner Moffitt (Columbia Business School ’07). founding partner at Future Heights Ventures. Here, the pitch wasn’t about new power plants. It was about shaving demand spikes from AI systems—so the grid can handle the load without needing everything built at once.
Ayse Coskun. co-founder and chief scientific officer at Emerald AI. explained that the company’s Conductor software platform can help dial back AI data center power demand at crunch times for the grid. The underlying idea was direct: if data centers can shift. or “flex. ” some computing load to periods of lower power demand. they can help utilities reduce demand peaks.
Coskun made a case that flexibility isn’t a side benefit. She argued it may be one of the fastest ways to unlock more compute without waiting years for new wires and plants.
Emerald AI has raised $64.99 million since its founding in 2024, according to Pitchbook. The session also pointed to participation by Nvidia’s venture arm, NVentures, Salesforce Ventures, Samsung Ventures and Siemens in the latest financing round.
Utilities, clouds, and “clean, firm” power
The conference’s clean energy discussions brought the tension back to the people who must keep electricity flowing as demand climbs. On the “Scaling Clean Energy for AI” panel. Gernot Wagner. senior lecturer at Columbia Business School and faculty director of the Climate Knowledge Initiative at the Tamer Institute for Social Enterprise and Climate Change. moderated.
Chelle Izzi (Columbia Business School ’04). chief commercial officer at PG&E—one of the largest utility companies in the United States—spoke alongside Heather McGeory (SIPA ’05). vice president of energy and sustainability at CoreWeave. which rents out cloud capacity specialized for AI computing. Nestor Andres Sepulveda, commercial lead for advanced energy investments and partnerships at Google, completed the panel.
Their shared point was that utilities are under pressure to serve rapidly growing large energy loads while also managing affordability, community expectations and decarbonization goals. CoreWeave and Google, the discussion emphasized, are increasingly looking for “clean, firm” power.
Compact nuclear enters with a concrete deadline
The panel also turned toward nuclear in the form of mass-manufactured compact systems built for data centers. Jon Guidroz, senior vice president of commercialization and strategy at Aalo Atomics, described a company effort aimed at speeding up deployment.
Aalo specializes in mass-manufactured compact nuclear power systems purpose-built for data centers. The conference material said the company is valued at just under $700 million dollars by Pitchbook. and that in early May it announced it is moving into the final safety check before testing its reactor technology with the Department of Energy. The goal, Guidroz said, is a first controlled nuclear reaction by July 4, 2026.
The throughline: urgency, but momentum
What tied the conference together wasn’t a denial of difficulty. It was the belief—expressed directly by Usher in his closing remarks—that the tools now exist to move faster than many people assume.
“What makes this moment especially exciting is that innovation is rapidly expanding,” Usher said. “Advances in clean energy, storage, green monetization and next-generation infrastructure are creating new pathways for growth.”
Across geothermal’s scale-up. green hydrogen’s policy deadline crunch. AI demand flexibility’s focus on peak reduction. and the push for clean. firm power from utilities. clouds. and data-center-focused nuclear. the same tension stayed visible: the problem is huge and the timelines are tight. But the panels’ consistent message was that the next steps—investment, deployment, and operational adaptation—are already in the room.
Columbia Climate Business and Investment Conference Tamer Institute AI power demand geothermal enhanced geothermal systems Fervo Energy green hydrogen Electric Hydrogen PEM electrolyzers Inflation Reduction Act One Big Beautiful Bill Act Emerald AI demand response data centers PG&E CoreWeave Google clean firm power Aalo Atomics compact nuclear