Performances
RD Popularity Index
RD Growth Index
Comparison
Follow Updates
Follow Lists
People
About Prime Intellect
Prime Intellect is decentralized-AI training infrastructure that aggregates globally distributed GPU compute into a collaborative training network, letting researchers and companies train large models without depending on a single cloud provider. Targeting the monopoly of frontier training by a few giants' centralized compute and budgets, its PRIME framework uses asynchronous reinforcement learning to decouple training into rollout, training and parameter-distribution stages, with SHARDCAST broadcasting weights and the TOPLOC proof system using locality-sensitive hashing to verify inference integrity across heterogeneous hardware.
Prime Intellect's trade-off is the communication and consistency overhead of decentralized training and its reliance on the stability of scattered compute supply; its fit is AI research teams pursuing open, reproducible distributed training. In June 2026 its prime-rl 0.6.0 framework completed asynchronous reinforcement-learning training of a trillion-parameter MoE model on 28 H200 nodes.
In January 2024, Vincent Weißer and Johannes Hagemann co-founded Prime Intellect to pool globally distributed GPU compute and provide open infrastructure for decentralized AI training.
In April 2024, Prime Intellect closed a $5.5M seed round co-led by Distributed Global and CoinFund, funding its compute marketplace and distributed-training framework.
In November 2024, Prime Intellect released INTELLECT-1, the first 10B-parameter model trained across three continents of distributed compute.
In February 2025, Prime Intellect secured a $15M round led by Founders Fund with angels like Karpathy, bringing total funding above $20M.
In May 2025, Prime Intellect launched INTELLECT-2, the first 32B-parameter reasoning model trained via globally distributed asynchronous reinforcement learning.
In November 2025, Prime Intellect released INTELLECT-3, a 106B-parameter mixture-of-experts model with 12B active, trained across 512 NVIDIA H200 GPUs.
In July 2026, Prime Intellect announced a $130M Series A led by Radical Ventures with NVIDIA, Intel and Dell capital, at a $1B valuation.
On July 8, 2026 Prime Intellect closed a $130 million Series A led by Radical Ventures with NVIDIA Ventures, Intel Capital and Dell Technologies Capital participating, at a $1 billion valuation and more than $150 million raised in total. The company cites over 6,000 customers and more than $100 million in annualized revenue.
Customer Ramp used Prime Intellect's Lab post-training stack to train a 35B model that beat Opus on spreadsheet search while running 27% faster and far cheaper than Haiku; Zapier and Flapping Airplanes are also users. Prime Intellect is expanding its compute marketplace and RL environments, packaging training, inference and environments for enterprises that want to own their models.
Prime Intellect's bet is that reinforcement learning spreads frontier capability beyond a few labs, so it does not sell GPU hours but the whole stack that lets an enterprise own its model: 6,000 customers and over USD 100 million in annualized revenue show task-based pricing already works. The moat is open-source distribution, where Verifiers, prime-rl and 2,500 community environments are adopted free and then monetized through hosted training, inference and evaluations; joining NVIDIA's Nemotron coalition embeds it in another model ecosystem instead of competing head-on. It faces price cuts from frontier labs and hyperscaler in-house stacks on one side and GPU-only neoclouds on the other, and must prove RLMs and continual learning support multi-day agents or the stack compresses into costlier GPU resale.
Prime Intellect was co-founded by Vincent Weisser, its CEO. Raised in Germany, he started as an entrepreneur in biotech and decentralised science, co-founding VitaDAO for longevity research funding and working on IP tokenisation at Molecule. In 2024 he assembled a team with Johannes Hagemann and took charge of hiring and research direction.
Johannes Hagemann is co-founder and CTO. He previously built training frameworks and inference optimisation at Aleph Alpha and now owns communication compression and fault-tolerant scheduling. Jimmy Z. is COO, covering partnerships and compute procurement; Will Brown leads applied research after machine learning research at Morgan Stanley; Jannik Straube heads engineering for node software and job orchestration. Research, engineering and operations report to the CTO and COO, and the founder runs a weekly all-hands.