Decentralized AI inference platforms are capturing a larger share of onchain requests by offering lower costs than centralized providers. Chutes and Surplus Intelligence are leading this trend, focusing on making AI model execution more affordable for users.
Chutes, built on Bittensor's Subnet 64, has processed more than 34 trillion tokens in total, with daily peaks reaching over 120 billion tokens. Launched in late January 2025 as a serverless platform, it utilizes independent GPU miners who are ranked by speed, capacity, and availability.
The platform uses Trusted Execution Environments (TEEs) and GraVal technology for hardware verification and data privacy. Chutes charges single-digit cents per million tokens for specific models, with all settlements conducted onchain in USDC.
Surplus Intelligence, launched in late May 2026, operates as a marketplace for unused AI inference capacity. In late May 2026, it handled 47,000 requests and 1.7 billion input tokens, which grew to 2.5 million requests and 107 billion tokens by late July 2026.
Surplus reports that users save 40-60% compared to traditional providers. The platform uses a dynamic price discovery model based on supply and demand, and like Chutes, it settles all transactions in USDC onchain.