In Q3 2025, Akash Network’s new leases rebounded 42% QoQ to 27,000 following two quarters of decline, though the surge was driven primarily by short-duration inference workloads rather than long-lived deployments. Despite this, revenue remained stable, with lease income rising 4% to $851,700 and network fee revenue growing 4% to $860,000, supported by improved billing stability and operational tooling. GPU usage held steady at -1% QoQ against a 7% contraction in capacity, keeping utilization above 50% as demand for inference workloads persisted. The number of active providers declined from 70 to 63, an 11% QoQ decrease and the first pullback after multiple quarters of growth, as smaller operators exited ahead of Starcluster and during Mainnet 14 preparation. This trend contributed to reductions across CPU, RAM, and storage supply, shifting the network’s workload mix toward shorter, burst-style deployments. Governance remained active with proposals funding community programs, incentive pilots, analytics infrastructure, and marketing initiatives, and Akash expanded its AI backend capabilities through new frontier model deployments and multiple Console and API upgrades. Overall, Q3 2025 reflected a repositioning phase for Akash as the network prepared for major supply expansions and product launches. While provider capacity tightened and long-running workloads moderated, underlying demand for decentralized inference compute remained steady, and revenue stabilized. With Mainnet 14 delivered and foundational work for Starcluster and AkashML underway, Akash entered Q4 with strengthened infrastructure and growing momentum toward enterprise-scale AI adoption.