Madrone

Madrone is a Mountain View/San Francisco-based hardware startup (YC Spring 2026) that builds hyper-efficient cooling systems for data centers using novel dew-point cooling technology. Rather than relying on traditional mechanical chillers, Madrone's modular systems sit outside the data center and chill water down to the dew point - a thermodynamic approach that produces cooler supply water (around 20°C) while consuming 30% less power and water than conventional chiller-plus-cooling-tower configurations. Because they eliminate the chiller entirely, customers can save tens of millions of dollars per facility in capital expenditure, while also avoiding the multi-year grid permit backlogs that constrain data center growth.

The company's solution is particularly valuable in hot, arid climates like Texas, where the majority of new hyperscale data center builds are concentrated. As AI workloads demand ever-denser GPU clusters (including NVIDIA's Vera Rubin and Blackwell architectures), up to 40% of facility power can be diverted to cooling rather than compute. Madrone's efficiency gains translate directly into more GPU revenue - by the company's estimate, every 1% improvement in cooling efficiency saves $2.6 million per 100 MW per year, meaning Madrone's system can unlock roughly $78 million in additional GPU revenue per 100 MW annually. Their system works both for new hyperscale builds and retrofit capacity expansions in existing data center shells.

Madrone is a fully vertically-integrated hardware company: the team designs its own electronics, heat exchangers, and software in-house to maximize system performance. Co-founder and CEO Akshay Trikha previously built ML inference systems at QuantumScape and studied material science ML at UC Berkeley, while co-founder Erik Meike spent four years at Apple designing power and thermal systems for the iPhone. Backed by Y Combinator and enrolled in the NVIDIA Inception Program, Madrone represents a critical node in the US energy supply chain - enabling data centers to squeeze more compute from constrained grid capacity at a time when AI infrastructure investment is surging nationwide.