Podcast Ep.298 - Akash, while GPU demand remains stable... what does the exit of small suppliers mean?

In the third quarter of 2025, the decentralized cloud computing platform Akash entered an adjustment phase through a short-term workload rebound and major network upgrades. The network fee revenue reached 715,000 AKT (approximately $860,000), representing a rise of 11% and 4% compared to the previous quarter, but this was mainly due to gradual improvements in billing and operational tools. New leasing volume rebounded 42% to 27,000 transactions, but most were short-term inference workloads and did not convert into long-term active leases.

GPU demand has decreased slightly by 1% to 367 units, but utilization remains above 50%. This is interpreted as a result of increased developer experimentation following the launch of Starcluster and the expanded support range of the latest cutting-edge models. On the other hand, the supply side has seen a decline in capacity across all areas of CPU, memory, and storage due to the exit of multiple small providers from the network, leading to a decrease in the number of active providers from 70 in the previous quarter to 63, marking the first decline.

On the technical level, through the mainnet upgrade 14, the network has migrated to Cosmos SDK v0.53, and introduced credit card payment API and JWT authentication features, lowering the entry barrier for non-native enterprises. Usability improvements for users and developers, such as API, console, and automated hosting features have also been significantly enhanced. At the same time, by integrating major AI models such as GPT-OSS-120B, Qwen3-Next-80B-A3B, and DeepSeek-V3.1, its foundation as a Decentralization AI backend has also been strengthened.

Star Cluster is a new protocol introduced by Akash that utilizes proprietary GPU infrastructure, aimed at purchasing 7,200 Nvidia GB200 GPUs through a “Starbond” worth up to $75 million. These hardware will be managed by verified enterprise data center operators (Nodekeepers) and are expected to become a core strategy for addressing the ultra-large-scale AI demands of the future. These initiatives are strengthening Akash's market position as a “cloud AWS alternative” in the field of Decentralization AI computing.

According to a study by Messari, Akash is evaluated relatively highly for its long-term potential as a distributed AI infrastructure and data-centric infrastructure protocol. However, issues such as complexity, infrastructure costs, and provider retention rates have been pointed out as challenges that need to be actively addressed.

Akash has announced that it will launch the AkashML product starting in the fourth quarter and gradually introduce cluster GPUs. Whether it can leap to a complete enterprise-level AI infrastructure in the future is a matter of great concern. Whether the short-term inference-focused workload structure can stabilize into a long-term profitable foundation, as well as whether incentives can be introduced to prevent the loss of small providers, remains an important issue.