By KarenZ, Foresight News

In the context of the rapid development of AI technology, the centralization and high cost of computing resources are still the main bottlenecks for innovation and popularization. As a decentralized peer-to-peer computing power and intelligent protocol, Prime Intellect aims to provide a new path for open source intelligence by aggregating global distributed computing resources and promoting collaborative AI model development.

This article will introduce this project from three dimensions: investment background, team background and how the protocol works.

Prime Intellect Team Background

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

CEO Vincent Weisser has relatively rich experience in Web3, focusing mainly on DeSci and AI. He is the co-founder of DeSci projects Bio Protocol, VitaDAO, and CryoDAO, and previously served as the head of ecosystem and AI at DeSci Molecule. In addition, Vincent Weisser also helped create 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, etc. He worked as an AI research engineer at Aleph Alpha, a German AI system developer, and served as a strategic advisor to VitaDAO from May 2021 to January 2023.

Financing background: Heavyweights in the AI field have invested more

Prime Intellect In April 2024, Prime Intellect completed a $5.5 million seed round of financing, led by Distributed Global and CoinFund, with participation from Compound, Collab+Currency, and Protocol Labs founder Juan Benet. In addition, the seed round also attracted well-known angel investors such as Clem Delangue, CEO of Hugging Face, a machine learning building tool.

At the beginning of 2025, Prime Intellect made another significant financing progress and completed a new round of financing of US$15 million, led by Founders Fund and participated by Menlo Ventures. This round of financing attracted many heavyweights in the fields of AI and Web3, including Andrej Karpathy (EurekaAI, Tesla, OpenAI), Clem Delangue (Hugging Face CEO), Dylan Patel (SemiAnalysis), Tri Dao (Chief Scientist of Generative AI Cloud Platform Together.AI), Balaji Srinivasan (Network School) and Emad Mostaque (Co-founder of Stability AI), Jake Medwell (Co-founder of 8VC), Brendan McCord (Cosmos Institute), Sandeep Nailwal (Co-founder of Polygon), etc. After the completion of this round of financing, its total fundraising exceeded US$20 million.

It is worth mentioning that Prime Intellect investor Andrej Karpathy is one of the founding members of OpenAI, and has worked as a research scientist at OpenAI and the director of artificial intelligence at Tesla. Emad Mostaque was the CEO of Stability AI, but later resigned to focus on decentralized artificial intelligence. He is also a consultant to Render, a decentralized GPU rendering solution provider.

How does it work?

Officials said that Prime Intellect builds on all existing decentralized AI work to develop infrastructure and economic incentives to aggregate and coordinate global computing resources to achieve a truly sovereign open source AI ecosystem. Currently, the protocol is running on an internal testnet and is fully integrated and online in the SYNTHETIC-1 operation. Prime Intellect has delivered several key tasks or products in the past year.

Compute Exchange: Aggregates, coordinates and schedules global GPU resources, integrating computing resources from data centers, cloud service providers and even individual GPU holders into a unified resource pool. Users can choose the most cost-effective GPU resources based on requirements such as chip type, quantity and rental time. This approach not only reduces the computing cost of AI training, but also promotes the democratization of AI technology through efficient supply and demand matching and improved resource utilization.

PRIME: A decentralized training framework that supports training large-scale models on globally distributed computing resources, enables fault-tolerant training, supports dynamic turning on/off of computing resources, and optimizes communication and routing in globally distributed GPU networks.

In October 2024, Prime Intellect launched INTELLECT-1, a decentralized training of a 10 billion parameter model. Leading open source AI participants such as Hugging Face, SemiAnalysis, Arcee, Hyperbolic, Olas, Akash, and Schelling AI jointly contributed computing to this decentralized training run.

GENESYS: A synthetic data generation framework that easily extends open source libraries for synthetic data generation and verification and the invocation of crowdsourcing tasks and verifiers. Prime Intellect announced in February 2025 that it would use DeepSeek-R1 to create SYNTHETIC-1, a verified open source dataset of mathematics, coding, and scientific reasoning trajectories.

TOPLOC: Prime Intellect introduces a verifiable reasoning method, TOPLOC, which uses compact Locality Sensitive Hashing (LSH) technology to accurately detect unauthorized modifications to models, hints, or computational accuracy. When generating a response, you need to perform a pre-fill on the input token once, and then perform multiple decoding on each new token generated. When verifying, you can pass all tokens at once and perform only a pre-fill once.

The Prime Intellect Protocol forms the coordination layer that ties all of these components together and provides them with economic incentives to attract the compute and capital needed to achieve scale.

The main components of the Prime Intellect protocol include Workers (users running software to contribute computing resources), Validators, Orchestrators (servers used by computing pool creators to manage workers in the pool and coordinate the running of current workloads), and smart contracts.

The Prime Intellect smart contract is currently deployed on the Base Sepolia testnet, but Prime Intellect plans to migrate to its own chain later. Prime Intellect's contract component RewardsDistributor also describes that the rewards of nodes can be calculated based on the active time in the computing power pool and tokens can be distributed accordingly.

This means that Prime Intellect may issue tokens when it goes online on the Base mainnet or launches its own chain. Prime Intellect plans to release a public testnet in the coming months, allowing anyone to contribute computing power and make contributions without permission.

summary

Prime Intellect builds a sovereign open source AI ecosystem by aggregating large-scale computing power and providing a global GPU resource scheduling market, allowing anyone to create and contribute open models, agents, data sets, etc., and collectively obtain ownership of their contributions. This model is expected to promote global participants to jointly build open source AI models.

Prime Intellect says it is also exploring collective ownership and programmable licenses for models and agents to reward contributors and provide new revenue streams for open source AI.

As stated in its mission, Prime Intellect hopes to popularize AI development on a large scale through the power of decentralization, accelerate progress in high-impact fields such as science, agents, and code, and ultimately build an open, transparent, and intelligent ecosystem that benefits more and more people.