DEPLOYDatabase

Company

Akash Network

Akash Network is a decentralized cloud computing marketplace, sometimes described as an "Airbnb for data centers," where independent providers list spare GPU…

Employees

50

Overview

Akash Network is a decentralized cloud computing marketplace, sometimes described as an "Airbnb for data centers," where independent providers list spare GPU and CPU capacity and buyers bid for it through a reverse-auction settled on the Cosmos-based AKT blockchain. The protocol is developed by Overclock Labs, founded in 2015 by Greg Osuri and Adam Bozanich, and positions itself as a "Supercloud" alternative to AWS, GCP, and Azure for AI training and inference workloads. Rather than owning datacenters, Akash aggregates capacity from roughly 60 third-party providers, ranging from small colocation operators to individual RTX 4090/5090 owners under its 2026 "Homenode" program.

Its economic pitch is price: the network advertises H100 GPU rentals near $1.33/hour versus roughly $3.93/hour on AWS, and its managed inference layer AkashML (launched November 2025) claims 70 to 85 percent lower cost than AWS SageMaker. On the demand side, Akash reported crossing $5 million in all-time compute spend by Q1 2026 and processing about 1.7 billion inference tokens per day through an OpenRouter integration. Named production users include Venice.ai, ElizaOS, Razer's AVA Mini campaign, and Nous Research's Hermes agent, plus community-built agent frameworks.

Unlike centralized neocloud peers such as CoreWeave or Lambda, Akash has no owned campuses, no disclosed megawatt footprint, and no traditional venture-round funding history: the project raised via a 2020 token sale and is governed by the AKT community treasury, which returned 3.37 million AKT to itself in 2025. That makes Akash a long-tail, price-driven inference marketplace rather than a hyperscale training substitute. Its GPU pool, roughly 1,000-plus units at end-2025, is orders of magnitude smaller than the H100/H200 fleets at CoreWeave, Nebius, or Crusoe.

Reality check

AkashML (company-claimed)

Managed inference launched November 2025, claiming 70-85% lower cost than AWS SageMaker; about 1.7 billion tokens/day via OpenRouter (Q1 2026).

Verified record

Verified deployments
None on file
Active incidents
None on file

Key facts

Business model

A decentralized marketplace where third-party providers list GPU/CPU capacity and buyers bid by reverse auction, settled in AKT on a Cosmos-SDK Layer 1.

GPU inventory (end-2025)

About 1,000+ GPUs across about 60 third-party providers, no owned fleet or datacenters; 2026 Homenode adds consumer RTX 4090/5090 owners.

Hourly pricing (company-advertised)

H100 rentals advertised near $1.33/hour vs about $3.93/hour on AWS. Prices are set by reverse auction, not a fixed rate card.

Positioning

Inference-focused, price-driven long-tail marketplace. Not a hyperscale training substitute; its GPU pool is far smaller than CoreWeave or Nebius.

Network compute spend

Cumulative spend crossed $5 million in Q1 2026; 2025 spend was $3.15M, up 128% year over year.

Named users

Venice.ai, ElizaOS, Envision Labs, EaveAI, Codex Storage, Razer (AVA Mini campaign) and Nous Research (Hermes agent).

Funding and token

No VC round disclosed after seed. Funded by a 2020 AKT token sale on CoinList and the AKT community treasury; AKT trades on crypto exchanges.

Burn-Mint Equilibrium

Went live March 23, 2026: compute demand burns AKT to mint the ACT stablecoin used for settlement.

Confidential compute

Launched July 28, 2026, letting workloads run on provider machines whose owners cannot inspect them.

Data & sources

Web sources

6

6 sources backing this record.View all →

Current leadership (2)

  • Greg Osuri Founder and CEO, Overclock Labs (Akash Network)reported, not verified
  • Adam Bozanich Co-founder and CTO, Overclock Labsreported, not verified

Safety record

No incidents on record for Akash Network.

Only active incidents are counted. Retracted incidents are excluded from this summary but remain reachable at their canonical URLs.

Full safety record: incidents, sourcing, and exposure data →