VCs start the week shifting fresh checks from AI apps to data center power and networking
Sofia Reyes ·
New York and San Francisco venture firms began the week prioritizing infrastructure plays over AI applications, steering new term sheets to startups that add power, cooling, and networking capacity to data centers. Partners cite stubborn GPU bottlenecks and grid constraints that are delaying AI deployments, moving budget toward the layers that unlock capacity.
The timing matters because funds are setting fourth quarter allocations now, and founders racing to close by year end are meeting investors who want durable revenue tied to cloud and utility procurement cycles rather than consumer or workflow apps with soft retention.
Investor shift to AI infrastructure
Generalist firms that leaned into AI applications in 2023 and 2024 are redirecting early stage attention to physical and network adjacencies. Climate and industrial specialists like Breakthrough Energy Ventures, DCVC, Lux Capital, Eclipse Ventures, and Congruent Ventures have seen increased syndicate interest from multistage funds on microgrids, thermal management, and optical interconnects.
Recent financings for networking and chip adjacency companies, including optical interconnect players such as Lightmatter and Celestial AI and fabric and data movement startups like Enfabrica, established a template for multi hundred million dollar capital needs supported by cloud partnerships. On the power side, compute providers using stranded energy such as Crusoe Energy, and geothermal developers like Fervo Energy working with hyperscalers, show a path to revenue under long term offtake.
Investors say the unit economics look more defensible than many AI apps that rely on commoditized foundation models and face rising inference costs. Infrastructure companies sell into procurement driven buyers, accept lower gross margins, and trade margin for multi year contracts and switching costs that resemble classic picks and shovels.
Why it matters now
Hyperscaler capex guidance and interconnection backlogs are dictating AI deployment speed, and those signals are cascading into venture pacing. A backlog of AI software startups that depend on faster and cheaper inference is now constrained by power, cooling, and network throughput, giving infrastructure startups stronger pull from customers and clearer pricing power.
For the broader tech landscape, the next leg of AI returns is likely to accrue to industrial layers, not just software wrappers. That shifts venture’s center of gravity toward energy, manufacturing, and networking expertise, and it ties startup outcomes more tightly to utility timelines and cloud buildouts.