He’s spent the better part of his career funding water-related innovation, focusing on where new technologies stand to have the greatest impact. So I believed him when he said that while data centers don’t merit quite so much worry, water as a resource deserves a far greater role in the climate tech conversation.
“Everybody assumes that water is a dog of a market because nobody really speaks water. It’s not within their circle of competence,” Ferguson told me, explaining that many firms simply don’t have employees with industry expertise. “But it’s awfully helpful to work with people who can give you a reasonably sized check — ideally two reasonably sized checks, maybe even more — and then also be helpful on that journey to help you better diagnose reality.”
That’s the goal of Burnt Island, which just closed a $50 million fund — its second overall — dedicated to backing early-stage water innovators. Ferguson’s team may have announced the close today, but the firm has already deployed the majority of the fund’s capital into companies working on everything from advanced water treatment and filtration to infrastructure resilience and climate adaptation. At the same time, Burnt Island is also raising money for a $75 million growth fund, designed to invest in later-stage startups with more proven tech.
Ferguson is a veteran of the industry, having previously run an innovation accelerator at the water nonprofit Imagine H2O, which vets hundreds of water startups every year. He’s also solution-agnostic — Burnt Island has already backed a startup developing an underwater desalination plant, a “defrosting innovation company” pioneering a water-efficient way to thaw frozen food, and an effort to build an algae-based wastewater treatment system.
One area Ferguson is not interested in backing, however, is data center cooling systems. Most large data centers cool servers by circulating water through heat exchangers that absorb heat from the equipment. The hot water is then sent to cooling towers where a portion is evaporated. This releases heat into the air, allowing the cooled water to be recirculated. More novel and efficient — but much less proven — cooling methods include applying coolant directly to the chips themselves or submerging entire servers in a non-conductive liquid.
Those approaches are simply too risky, Ferguson told me — both for him and for the hyperscalers. Cooling, he explained, represents a relatively small fraction of a data center’s project cost, but the cost of failure is enormous. If a novel cooling system goes awry, valuable computer chips will fry and operations will grind to a halt. “Under those circumstances, why would you take that chance?” he asked. “You want to use something that has already been proven, that is totally reliable.”
Ferguson told me he’s happy to let firms with larger pocketbooks bet their money on these solutions, but he’s also assuming that hyperscalers will wind up building a lot of these systems themselves. “They’re going to develop their own stuff in house because they want to have the end-to-end control over the architecture,” he told me. “All of this adds up to a pretty tough market.”
That doesn’t mean he’s bearish on data center water efficiency in general. Many of his portfolio companies see opportunities to, say, use metering and sensing tech to track data center water use, or treat water coming into and out of the facilities. And he’s well aware of the public’s growing scrutiny of the industry’s water intensity, having followed the $3.6 billion data center project in Tucson, Arizona that was cancelled in August amidst community-led drinking water concerns.
But he thinks kerfuffles such as this are often more about perception than reality. “The water impact is slightly overblown,” he told me. Data centers “still use a lot less water than golf courses.” And while the rapid expansion of artificial intelligence infrastructure will inevitably put data centers ahead of golf courses one day, Ferguson trusts that this cash-rich industry will be able to reduce water intensity on its own, as developers have a direct incentive to expand in as many geographies as possible.
Even the canceled Arizona project, he told me, had a reliable plan to replenish the local watershed. Microsoft, Amazon, and Google have all pledged to be “water positive” by 2030, returning more water to data center communities than their facilities use by making their operations more efficient while also restoring local ecosystems and replenishing watersheds. But now that the water use narrative has gained steam, “it actually doesn’t matter what you do physically. It’s what people believe about the resource hungriness of these things,” Ferguson explained.
The more important question, he believes, is whether AI’s overall impact on the world will end up justifying the water it consumes. And as he told me, “the jury is really out” on that for now.
But when it comes to weighing water consumption against the pure economic value of data centers, Christopher Gasson, owner and publisher of the market intelligence firm Global Water Intelligence, has actual numbers.
As Gasson asserted in a presentation that Ferguson attended, in terms of the amount of fresh water used per dollar of revenue generated, data centers perform quite well compared to the world’s other leading industries. Their so-called “revenue intensity” is far lower than that of the semiconductor, power generation, food and beverage, and chemicals sectors, for example.
So for Ferguson, the AI-water intersection that feels most relevant is actually “vertical AI” — models trained specifically on water industry data to address targeted problems in the sector. Training these smaller, specialized models is not only far less resource-intensive, it also allows for much more accurate results than general purpose models, which often hallucinate when trying to address niche queries and concerns.
One of Burnt Island’s portfolio companies, SewerAI, trains its model on reams of sewer inspection data. Using video footage, the software can then perform automated sewer inspections to identify defects in pipes, eliminating the timely, costly, and often inaccurate process of manual video review. Another portfolio company, Daupler, uses its specialized model to automate how water utilities respond to service incidents, categorizing and prioritizing customer reports, dispatching crews, and tracking progress. Burnt Island led Daupler’s Series A round and has already supported it with additional capital through its growth fund.
“You have these really, really high quality, very compelling business models that are being built relatively quietly,” Ferguson said. But he expects these opportunities to gain more attention soon — because while the headlines and community uproar around the water intensity of AI may sometimes be hyperbolic, the necessity of water to human life is anything but.
“You can’t believe in water in the same way that people have chosen to believe in the impact of emissions,” Ferguson told me. “You don’t get to choose when it comes to water issues, because once they get real, they get really real.”
Dead trees in the Brazilian Amazon. Mario Tama/Getty Images
Senator Sheldon Whitehouse of Rhode Island.Andrew Harnik/Getty Images
