What is Prime Intellect—the AI project backed by key figures such as OpenAI’s founding members, Tesla’s former AI director, and the ex-CEO of Stability AI?

As AI technology advances at an unprecedented pace, the centralization of computing power and its high costs remain major obstacles to innovation and accessibility. Prime Intellect is a decentralized peer-to-peer computing and intelligence protocol designed to harness globally distributed computing resources, facilitating collaborative AI model development and offering a new path for open-source AI.

This article explores Prime Intellect from three key angles: its investment background, team, and how the protocol operates.

Prime Intellect Team Background

Prime Intellect was founded in January 2024 by two co-founders, Vincent Weisser and Johannes Hagemann.

CEO Vincent Weisser has relatively extensive experience in Web3, primarily focusing on the DeSci and AI sectors. He is a co-founder of DeSci projects such as Bio Protocol, VitaDAO, and CryoDAO. Previously, he also served as the Head of Ecosystem and AI at DeSci Molecule. In addition, Vincent Weisser helped establish the experimental community Zuzalu.city in 2023.

CTO Johannes Hagemann focuses on decentralized AI, semi-automation, energy optimization, explicit technological optimism, brain-computer interfaces (BCIs), longevity, and other fields. He previously worked as an AI Research Engineer at the German AI systems developer Aleph Alpha. From May 2021 to January 2023, he also served as a strategic advisor at VitaDAO.

Funding Background: Major AI Figures Increasing Their Investment

In April 2024, Prime Intellect completed a $5.5 million seed round led by Distributed Global and CoinFund, with participation from Compound, Collab+Currency, and Juan Benet, the founder of Protocol Labs. Additionally, this round attracted prominent angel investors, including Clem Delangue, the CEO of Hugging Face, a leading machine learning tool provider.

At the beginning of 2025, Prime Intellect achieved another major funding milestone, securing $15 million in a new round led by Peter Thiel’s Founders Fund, with Menlo Ventures among the investors. This round attracted several influential figures from the AI and Web3 industries, including:

• Andrej Karpathy (EurekaAI, Tesla, OpenAI)
• Clem Delangue (CEO of Hugging Face)
• Dylan Patel (SemiAnalysis)
• Tri Dao (Chief Scientist at generative AI cloud platform Together.AI)
• Balaji Srinivasan (Network School)
• Emad Mostaque (Co-founder of Stability AI)
• Jake Medwell (Co-founder of 8VC)
• Brendan McCord (Cosmos Institute)
• Sandeep Nailwal (Co-founder of Polygon)

With this round, Prime Intellect’s total funding surpassed $20 million.

Notably, Andrej Karpathy, one of Prime Intellect’s investors, was a founding member of OpenAI, where he served as a research scientist, and later became Tesla’s Director of AI. Emad Mostaque, formerly the CEO of Stability AI, resigned to focus on decentralized AI and also serves as an advisor for Render, a decentralized GPU rendering solution provider. Additionally, Menlo Ventures, a participating investor, has an impressive portfolio that includes Uber, Poshmark, and Anthropic.

How Does It Work?

According to its official statement, Prime Intellect builds on all existing decentralized AI efforts, developing infrastructure and economic incentives to aggregate and coordinate global computing resources, aiming to create a truly sovereign open-source AI ecosystem. The protocol is currently running on its internal testnet and has been fully integrated and launched under SYNTHETIC-1. Over the past year, Prime Intellect has delivered several key projects and products.

Compute Exchange：Prime Intellect aggregates, coordinates, and schedules global GPU resources, integrating computing power from data centers, cloud service providers, and even individual GPU owners into a unified resource pool. Users can select the most cost-effective GPU resources based on factors such as chip type, quantity, and rental duration. This approach not only lowers the cost of AI training but also enhances the democratization of AI technology by optimizing supply-demand matching and increasing resource utilization.

PRIME: Decentralized Training Framework: PRIME enables large-scale model training on a globally distributed computing network, offering fault-tolerant training, dynamic resource activation/deactivation, and optimized communication and routing across distributed GPU networks.

In October 2024, Prime Intellect launched a decentralized training run for a 10-billion-parameter model, INTELLECT-1. Leading open-source AI players, including Hugging Face, SemiAnalysis, Arcee, Hyperbolic, Olas, Akash, and Schelling AI, contributed computing resources to this initiative.

GENESYS: Synthetic Data Generation Framework: GENESYS is designed for scalable synthetic data generation, facilitating dataset expansion, verification, and crowdsourced validation tasks. In February 2025, Prime Intellect announced the use of DeepSeek-R1 to create SYNTHETIC-1, an open-source dataset with verified mathematical, coding, and scientific reasoning traces.

TOPLOC: Verifiable Inference Method: Prime Intellect introduced TOPLOC, a method for verifiable AI inference. It utilizes Locality Sensitive Hashing (LSH) to accurately detect unauthorized modifications to models, prompts, or computational accuracy. When generating responses, pre-filling is required for input tokens, followed by multiple decoding steps for each generated token. During verification, all tokens can be processed at once, requiring only a single pre-filling step.

The Prime Intellect protocol acts as a coordination layer, linking all its components together while providing economic incentives to attract the necessary computing power and capital for scaling.

The main components of the Prime Intellect protocol include Workers (users running software to contribute computing resources), Validators, Orchestrators (compute pool creators who manage the workers in the pool and coordinate the servers running the current workload), and Smart Contracts.

Currently, Prime Intellect’s smart contracts are deployed on the Base Sepolia testnet, but the project plans to migrate to its own blockchain in the future. The RewardsDistributor contract within Prime Intellect describes how node rewards are calculated based on active time within a computing pool and how tokens are distributed accordingly.

This suggests that Prime Intellect may issue a token when it goes live on Base mainnet or launches its own blockchain. Prime Intellect plans to release a public testnet in the coming months, allowing anyone to contribute computing power and participate permissionlessly.

Summary

Prime Intellect aggregates large-scale computing power and provides a global GPU resource scheduling marketplace, allowing anyone to create and contribute open models, agents, and datasets, while collectively owning the results of their contributions. Through this approach, Prime Intellect aims to build a sovereign open-source AI ecosystem, encouraging global participation in the development of open-source AI models.

Prime Intellect has also stated that it is exploring collective ownership of models and agents, as well as programmable licenses, to reward contributors and create new revenue streams for open-source AI.

As part of its mission, Prime Intellect seeks to democratize AI development through decentralization, accelerating progress in high-impact fields such as science, AI agents, and coding, ultimately building an open, transparent, and inclusive intelligent ecosystem.

Disclaimer：

1. This article is reprinted from [ForesightNews]. The copyright belongs to the original author [KarenZ, Foresight News]. If you have any objections to this reprint, please contact the Gate Learn team, and they will process it promptly following the relevant procedures.
2. Disclaimer: The views and opinions expressed in this article are solely those of the author and do not constitute investment advice.
3. The Gate Learn team is responsible for translating the article into other languages. Unless explicitly mentioned by Gate.io, copying, distributing, or plagiarizing the translated article is strictly prohibited.