VCs open August with power-first shift in AI investing as energy costs hit deal terms

Venture firms kicked off Monday by tightening how they underwrite AI infrastructure startups, prioritizing power access and energy efficiency before valuation. Partners in New York and Silicon Valley said new term sheets now require proof of contracted megawatts, detailed cooling plans, and clear cost of compute models. The shift follows months of strained grid capacity around major data center hubs.

The most immediate change shows up in diligence lists and covenants. Investors are asking founders to present power usage effectiveness targets, water sourcing strategies, and interconnect timelines alongside gross margin and cohort data. Several funds are conditioning milestone-based tranches on securing utility commitments or long-term colocation agreements.

What investors are changing Early-stage partners describe a rotation from model-centric bets to what they call power-aware infrastructure. That includes schedulers that route training to off-peak windows, immersion and direct-to-chip cooling, alternative compute memory stacks, and software that markets flexible loads back to the grid. Pricing discipline is part of the reset, with lower step-ups for companies that rely on spot cloud instead of contracted energy.

Growth investors are also pushing project finance structures alongside equity. Funds are pairing checks with debt for substations, heat reuse, and on-site renewables to keep unit costs predictable. Specialist climate funds that backed transmission, storage, and advanced cooling are syndicating more frequently with traditional tech funds that want AI exposure without open-ended compute risk.

Why the timing matters Power has emerged as the gating factor for AI scale. Utilities report long interconnect queues in Northern Virginia, Central Ohio, and parts of Texas, and European markets continue to restrict high-load builds in dense corridors. That backdrop has made power procurement a defensibility question, not an operations footnote, for any AI infrastructure startup raising in August.

For founders, the bar has moved from synthetic benchmarks to durable economics. Investors want proof that every dollar of compute spend expands gross margin and that retention holds when workloads move from discounted credits to contracted energy. The firms that translate power constraints into moats will look more like infrastructure businesses, which pulls venture closer to private equity on structure and timelines.