Wafr Technologies, a Vancouver startup founded in March 2025, says it has raised $100 million from undisclosed private investors, part of a wider $300 million fundraising goal, to commercialise a cooling system for AI data centres and build an AI research lab in Canada. The company claims its thermal-battery system can cut a data centre's water use by up to 95% and cooling energy use by up to 80% compared with conventional evaporative cooling. Those numbers should be read with two caveats attached from the outset: they are Wafr's own figures, not independently validated performance results, and the company has not announced a commercial deployment. The raise itself is modest by AI-infrastructure standards. The water and cooling problem it targets is not.

Why cooling became a bottleneck

For most of the cloud era, cooling was treated as an engineering detail. In the AI era, it has become a constraint on where and whether a data centre can be built at all — sitting alongside power availability, GPU supply, land and grid interconnection as one of the hard limits on expansion. The reason is density. AI training and inference pack advanced chips tightly together, and those chips generate heat that has to be removed continuously; the larger facilities run by the major cloud providers now operate at 100 megawatts and above, where cooling can account for a large share of a facility's electricity use.

The water figures make the constraint concrete. The International Energy Agency estimated in 2025 that a typical 100-megawatt hyperscale data centre in the United States consumes around two million litres of water a day — roughly the daily water use of 6,500 households. That does not mean every AI facility everywhere has the same water profile; cooling design, climate, workload and local water systems all matter. But it shows why water is becoming part of the siting and permitting equation alongside power, land and grid interconnection. In some water-stressed regions, water availability can become a permitting constraint in its own right: an operator can have the power, the land and the chips lined up and still struggle to build because the local water supply cannot support conventional cooling. That is the pressure point a wave of cooling startups, Wafr among them, is trying to relieve.

What Wafr is actually selling

Wafr's system is built around what it calls a thermal battery — a thermal-energy-storage approach that stores cooling capacity when electricity is cheaper or more available, then releases it during peak demand hours. In principle, a closed-loop design can reduce the evaporative water loss of conventional cooling towers and shift part of the cooling load away from expensive peak grid periods. The appeal, if it performs as described, is twofold: less water consumed, and a cheaper, flatter electricity profile for cooling. The caution is that thermal storage as a concept is not new — chilled-water thermal storage has been studied and used for load-shifting before — while Wafr's claimed performance at commercial AI data-centre scale has not yet been independently demonstrated.

The rest of the picture reinforces how early this is. Wafr says the technology has been shown in India and Dubai, but it has not disclosed the scale, duration or results of those demonstrations, and no independent validation of the 95% and 80% figures has been published. The company has signed letters of intent with international partners building AI data centres, and is targeting the US and Germany, but it has not announced a single commercial deployment or a named operator customer. It is also pursuing an unusually broad model — selling cooling systems to third-party operators while simultaneously building and operating its own data centres — which doubles the execution challenge for a company barely a year old. The $100 million raised is the first tranche toward a $300 million target; the remaining $200 million, which Wafr wants from a mix of government and private sources, is what would actually fund commercialisation.

A crowded market that suddenly matters

Wafr is entering a category that has gone from unglamorous to strategic very quickly. The clearest sign of that shift came on 20 March 2026, when Ecolab agreed to buy the liquid-cooling firm CoolIT Systems for US$4.75 billion — one of the larger deals the sector has seen. Established thermal-management players such as Vertiv, Schneider Electric and Munters supply much of the conventional cooling stack, while liquid-cooling specialists including CoolIT, GRC and Asetek are scaling direct-to-chip and immersion systems for GPU-dense racks. Wafr sits in a different niche: thermal energy storage aimed at shifting the cooling load and cutting water use. The distinction matters for buyers — direct-to-chip liquid cooling attacks the heat at the chip, while thermal storage mainly changes when and how the cooling load is served. They are complementary approaches to the same crunch, not substitutes, and an operator may well need both.

The regional read

The water constraint is not an abstract North American concern. Southeast Asia is both a fast-growing data-centre region and, in places, water- and power-constrained. Singapore paused new data-centre approvals in 2019 over exactly these limits, and reopened under a sustainability-led allocation framework — its IMDA Green Data Centre Roadmap now emphasises energy- and water-efficient designs and liquid cooling. For operators expanding across the region, cooling technology that genuinely reduces water and peak-power dependence would ease a real bottleneck — which is precisely why claims like Wafr's need independent validation before anyone builds a regional deployment plan around them.

Key Takeaways

  • Wafr Technologies (Vancouver, founded March 2025) says it raised $100 million from undisclosed private investors on 3 July 2026, toward a $300 million goal, for a thermal-battery cooling system and a planned Canadian AI research lab; it is seeking a further $200 million from government and private sources.

  • Wafr claims its closed-loop system cuts data-centre water use by up to 95% and cooling energy by up to 80% versus evaporative cooling — but those figures are the company's own, have not been independently validated, and Wafr has no commercial deployments (only letters of intent and undisclosed demonstrations in India and Dubai).

  • The real story is the bottleneck: the IEA estimated in 2025 that a typical 100-megawatt US hyperscale data centre uses roughly two million litres of water a day (about 6,500 households), which is turning water availability into a siting-and-permitting factor in some water-stressed regions, alongside power, land and grid access.

  • Cooling is now strategic infrastructure: Ecolab agreed to buy CoolIT Systems for US$4.75 billion in March 2026, and incumbents (Vertiv, Schneider, Munters) and liquid-cooling specialists (CoolIT, GRC, Asetek) are scaling fast. Thermal-storage cooling like Wafr's is a complementary but, in its specific form, unproven niche. The signal is the bottleneck's importance, not proof any one startup has solved it — and for water-constrained regions such as parts of Southeast Asia, the claims need validation first.