VCs start Q4 by steering early-stage capital to AI power and data center startups
Sofia Reyes ·
On Monday, venture firms in New York and San Francisco opened the week by prioritizing term sheets for startups tied to power and data center capacity, according to multiple investors and advisors active in early-stage deals. The timing lines up with Q4 procurement cycles at cloud providers and utilities, which are driving near term demand for capacity, cooling, and grid interconnect solutions.
The shift matters because application-layer AI companies are running into slower sales cycles and margin pressure from inference costs, while infrastructure plays can sell into budgeted RFPs with offtake or reservation contracts. That revenue visibility is attracting partners who spent the last year backing copilots and agent platforms and now want assets that clear committee on unit economics.
Investor memos and recent portfolio moves from firms that have published energy and compute theses, including a16z, Founders Fund, DCVC, and Khosla Ventures, point to three categories drawing attention. First, power development and load management tied to AI campuses, including microgrids, interconnection planning software, and transformer supply. Second, data center enablement such as immersion cooling, optical interconnects, and workload orchestration that lowers power usage effectiveness and total cost of ownership. Third, long lead time generation bets such as geothermal and advanced nuclear that can be paired with credit markets enabled by the Inflation Reduction Act.
Founders pitching this week are being pressed on contracted backlog, utility relationships, and permitting timelines rather than only technology benchmarks. Investors are discounting models that rely on wholesale power arbitrage without site control or grid queue position, and they are rewarding startups that demonstrate capacity reservations from hyperscalers or AI labs and credible EPC partners.
Application AI is not off the table, but partners are sharpening thresholds. Teams selling copilots into vertical software are being asked for net retention above 120 percent without incentive credits, gross margins that account for inference, and proof that model switching does not erode differentiation. Deals that clear are layering proprietary data rights or workflow lock-in on top of falling token costs.
Why now is straightforward. Hyperscaler capex remains elevated, power shortages are delaying campus timelines in key markets, and rate relief has been uneven, which keeps capital costs visible in underwriting. That pushes generalist funds toward projects and products with defensibility that does not depend on model performance alone.
For the broader tech landscape, this reweighting blurs the boundary between venture and infrastructure equity and pulls more software talent toward regulated, asset-heavy markets. If it holds through Q4, the winners in 2026 will include founders who can pair deep technical chops with interconnection literacy, and application AI startups that prove durable margins after compute.