VCs open week shifting focus to power and cooling startups as AI faces grid limits

Several prominent venture firms began the week redirecting deal flow to startups that expand power and cooling capacity for AI data centers. The shift, discussed in Monday partner meetings in New York and Silicon Valley, reflects a view that energy and interconnect constraints, not model performance, will cap AI growth through at least 2027.

Partners say they are reserving more capital for companies that can add megawatts and move heat, including microgrids, advanced power electronics, liquid cooling, and load planning software. The timing matters because transformer shortages, slow interconnect queues, and permitting timelines have become binding bottlenecks as hyperscalers race to add compute.

Why power now

Public deals from hyperscalers have set the tone. Microsoft has inked agreements tied to nuclear and fusion projects, Google backed enhanced geothermal with Fervo, and Amazon continues to secure long term power across North America and Europe. Those moves signal durable offtake and create room for venture backed vendors that can plug into data center buildouts without taking balance sheet risk.

Investors with deep science and industrial DNA, including DCVC, Lux Capital, Khosla Ventures, Breakthrough Energy Ventures, Lowercarbon Capital, and Eclipse, have been active across grid software, thermal storage, and manufacturing of cooling equipment. Growth firms and crossovers that had pulled back from late stage software are also revisiting power adjacent businesses where contracts resemble infrastructure, not discretionary SaaS.

What gets funded

At seed and Series A, partners are sourcing teams that shorten interconnect timelines, monitor substation capacity in real time, or deliver modular substation in a box products that ease utility coordination. In data center operations, liquid and immersion cooling, heat reuse, and high density rack power distribution are drawing interest because they map to immediate pain and can be sold alongside existing vendors.

The economics differ from classic software. Startups that tie into long duration contracts with hyperscalers or colocation providers can show higher retention and lower churn, but they must manage hardware cycles, certification, and longer sales processes. Firms are preparing to blend equity with project finance and venture debt, and to underwrite supply chain and permitting risk alongside product risk.

For the broader technology landscape, this reweighting aligns venture with a physical constraint that now governs AI deployment. Power and cooling sit on the critical path, which means technical founders who pair infrastructure know how with enterprise selling, and investors comfortable with industrial diligence, will shape the next leg of AI capacity growth.