Ocean, the peer-to-peer computing project that withdrew last year from a highly publicized token alliance with Fetch.ai and SingularityNET, is once again sailing into a new port.

Last week, Ocean announced that users can now use its chip-rental system to process AI queries, in an expansion beyond the initial focus of letting developers fine-tune models and run data jobs.

In a blog post, Ocean said its practice of renting out computing hardware for dedicated time blocks will appeal to heavy AI users looking to rein in unpredictable bills.

For long and predictable jobs, paying by the hour is a better deal than paying by the word, Ocean said in the post, adding that the math "gets considerably easier when you see the number before the session starts rather than after it ends."

It is the latest turn for a decentralized AI project that has tacked repeatedly. The Ocean Protocol Foundation launched in 2017 as a decentralized exchange for data, and by 2021 the project's OCEAN token had a market capitalization of more than $700 million.

In March 2024, the project announced plans to form the Artificial Superintelligence Alliance with Fetch.ai and SingularityNET, only to walk out in October 2025, effective immediately. The token's market value now stands around $37 million, according to CoinGecko.

Inference, which refers to the computing power needed to process AI requests, as opposed to the capacity needed to train the models, should now be seen as Ocean's "flagship capability," according to the project documentation.

The Ocean team described the inference offering as "the capability we have been building toward since day one."

In March, Ocean launched a beta version of an "orchestration layer" that allowed developers to fine-tune AI models, but said it would initially provide the hardware through a deal with another decentralized AI project, Aethir, rather than from organic supply sourced via the peer-to-peer network.

A key marketing angle is cost savings. Ocean says it can provide an Nvidia H200 high-performance GPU for $2.16 an hour, versus $5.99 at the centralized provider Together AI, whose cheapest H200, at $2.99, can be reclaimed mid-job.

HOW AI WAS USED IN THE PRODUCTION OF THIS PIECE: I conducted an initial round of research for this story on Gemini, then created the initial draft on Claude using a skill designed to key off blog posts. I then used our Distro Publisher MCP connector to file directly to DistroVerse, our web app. I used the DistroVerse built-in story editor to almost completely rewrite the piece.