Index Ventures leads $28 million Series A for GridFlux to cut AI compute power bills
Sofia Reyes ·
GridFlux, a San Francisco based startup that orchestrates AI workloads to cheaper and lower carbon power, raised a $28 million Series A on Monday led by Index Ventures. The company said it shifts training and inference across clouds and regions based on grid price and emissions signals, a pitch aimed at model developers facing rising energy and compute costs.
The timing reflects a broader investor rotation toward energy aware AI infrastructure as heat waves and interconnection delays squeeze data center capacity. Cloud providers have introduced new demand response and carbon aware options, which creates a wedge for independent tools that operate across vendors.
Why the round matters
Index is backing an execution layer rather than another model company, a signal that cost and reliability have become the key bottlenecks in AI adoption. GridFlux said customers see double digit reductions in compute spend by moving jobs to lower cost hours and regions, and by aligning scheduling with renewable output.
The startup sells a usage based platform fee plus a share of savings, which ties revenue to measurable outcomes. According to a summary shared with investors, GridFlux ended June with low eight figure annualized revenue, gross margins above 80 percent, and net revenue retention above 120 percent.
Investors and market context
Index Ventures led the round with participation from Lowercarbon Capital and Susa Ventures, according to the company. Proceeds will fund integrations with major clouds and utilities, along with expansion in Texas and the Midwest where time varying tariffs and congestion create volatile price spreads.
Competition spans internal tools at hyperscalers and independent data providers that expose grid intensity signals. GridFlux is betting that multi cloud orchestration and direct utility connectivity can create defensibility, helped by market rules that allow distributed resources and flexible loads to participate in demand programs.
Early customers include model labs and enterprise analytics teams that can queue non urgent training jobs. The company said it is adding features for inference placement to avoid regional congestion, which matters as latency sensitive use cases leave the lab and hit production.
For the broader tech landscape, the deal underscores how AI funding is moving from training heavy bets to infrastructure that matches compute with energy and network constraints. Capital is following unit economics, not benchmarks, which favors startups that reduce cost of goods sold and improve reliability in a power limited cycle.