Madrone

Madrone is a very early-stage (founded 2025), two-person hardware startup based in San Francisco that is building chiller-less water cooling systems for AI/GPU data centers. It was created by CEO Akshay Trikha and cofounder Erik Meike, and went through Y Combinator's Spring 2026 batch (YC page: 'Madrone: Hyper-efficient coolers for data centers'). The company's site (madrone.cool — the '.cool' domain is a literal pun on its cooling product) and its YC launch post frame the problem as: AI data centers are increasingly power-constrained rather than chip- or land-constrained, and up to 30-40% of a facility's power can be diverted to cooling rather than GPUs, a problem made worse because many new hyperscale sites are being built in hot, dry desert regions (notably Texas) where evaporative/chiller cooling is inefficient and grid interconnection permits are backlogged for years.

Madrone's solution is a modular water-cooling system that sits outside the data center and chills facility water close to the dew point using a proprietary thermodynamic process, eliminating the need for mechanical chillers entirely. The company advertises ~20°C supply water while using about 30% less water and electricity than a traditional chiller-plus-cooling-tower system, which it estimates can save tens of millions of dollars per facility and unlock significantly more usable compute under the same grid permit. Every 1% improvement in cooling efficiency is pitched as worth an estimated $2.6M in incremental GPU revenue per 100MW per year, implying roughly $78M/100MW/year in value from Madrone's approach. The company positions its product for two deployment modes: new hyperscale builds (enabling next-generation GPUs such as NVIDIA's Vera Rubin platform to run in 45°C+ ambient conditions without thermal throttling) and retrofits/capacity expansions of existing facilities to prepare them for Blackwell- and Rubin-generation silicon.

Both founders bring relevant hands-on hardware/ML backgrounds. Akshay Trikha previously trained ML models handling millions of inferences at battery maker QuantumScape and worked on predicting material properties at UC Berkeley. Erik Meike spent about four years as a Hardware Development Engineer at Apple working on iPhone wireless power technologies, and earlier worked on early-stage hardware/robotics at JITx (automated PCB design) and PRENAV (autonomous quadcopter monitoring); he holds an engineering degree from Harvey Mudd College and says he has been exploring this cooling problem for roughly six years, well before the current AI data center boom. The founders describe themselves as full-stack, designing their own electronics, heat exchangers, and control software in-house.

Business model

Hardware/infrastructure vendor selling and installing modular water-cooling equipment (and associated controls software) to data center owners/operators; sits outside the data center as an add-on/retrofit or new-build cooling plant.

Presumed equipment sales and/or installation/service contracts with hyperscale and colocation data center operators (not publicly disclosed).