Madrone is building hyper-efficient data center cooling that can save 30% of grid capacity — already closing first investor check

Madrone is betting that data centers are grid-constrained before they are chip-constrained, and that cooling is where capacity gets unlocked.

Founder Akshay Trikha says the problem is structural. In Texas last year, Semi Analysis reported 150 gigawatts of load requests to a grid that peaks at 60 gigawatts on a good day, with only 1.5 gigawatts approved by local utilities. The grid permit, not the GPU count, is the binding constraint. Conventional mechanical chillers consume roughly 30% of that permit — and cost $500 million per gigawatt — leaving hyperscalers paying a steep tax on capacity they could redirect to compute.

Madrone's approach is a thermodynamic cooling system engineered specifically for hot and dry climates, where most hyperscale buildout is currently concentrated. The system sits alongside a data center — physically resembling a shipping container — and outputs cold water, making it agnostic to chip or rack design inside. Power draw goes to fans and pumps circulating fluid, not to the cooling chemistry itself. Trikha says all thermodynamics, mechanical structure, electronics, and control software are designed in-house, citing Tesla's vertical integration model as the template. Competitors, he argues, outsource each engineering layer and integrate them as a product.

“We've engineered a thermodynamic process that works really well in hot and dry climates, which is where most hyperscaling is happening right now... The big problem really is that data centers are limited today by power and not by chips or land. If you can save power on cooling, that's more power for your flops... My competitors are 100-year-old companies that have been designing the same thing for decades.”

Prototype trajectory

Madrone scaled from a 1 kilowatt prototype to a 100 kilowatt version during the YC batch. A 1 megawatt version is next, and Trikha says hyperscalers have made clear they need at least that scale before real procurement conversations open up. The 100 kilowatt unit running live at Demo Day is scheduled for its first pilot in two months.

First close

Madrone closed its first investor check on Demo Day. The investor is building $20 billion worth of data centers in Texas, giving Madrone a deployment partner for that initial pilot rather than just capital.

Trikha is also in active conversations with two hyperscalers and was scheduled to meet a third the week after Demo Day.

Founder background

Trikha scaled manufacturing at QuantumScape, working in computer vision directly under CTO Tim Home, who took the company from inception through IPO. He also holds a graduate degree from Berkeley in material science, which he says underpins work at Madrone that didn't come up in the conversation.

The commercial thesis is straightforward: grid permits take years to secure, chillers eat 30% of the capacity those permits allow, and any credible alternative with a faster delivery timeline has a short queue of buyers. Madrone's first proof point is whether the 1 megawatt unit performs reliably enough to convert pilot access into purchase orders from an investor already spending at scale in Texas.