Prime Intellect

Prime Intellect is an American artificial-intelligence infrastructure and research company. Its services include a marketplace for accelerator capacity, managed reinforcement-learning and evaluation systems, and model inference. The company also develops open-source training software and publishes open-weight models.¹

Prime Intellect’s Compute marketplace is a cloud GPU service, while its Lab and Inference products provide managed training and inference. Physical capacity is supplied by a network of cloud companies rather than a single fleet owned by Prime Intellect.

History

Prime Intellect was founded in 2023 by Vincent Weisser and Johannes Hagemann. Weisser serves as chief executive and Hagemann as chief technology officer. The company announced a US$5.5 million seed financing in April 2024 to develop a distributed computing and model-training platform.²

In February 2025, Prime Intellect announced a further US$15 million financing led by Founders Fund, bringing its reported funding above US$20 million.³ On 8 July 2026, it announced a US$130 million Series A led by Radical Ventures. NVIDIA Ventures, Intel Capital, Dell Technologies Capital, and existing investors also participated. The company reported total financing of more than US$150 million and more than 6,000 customers.⁴

Prime Intellect employs staff in San Francisco and remotely. Prime Intellect Inc. is registered in Delaware, with a registered office in Dover.¹⁵

Compute

Prime Intellect introduced its Compute marketplace in July 2024. The service aggregates machines and clusters from several infrastructure providers through a common interface. Customers select an accelerator type, number of devices, location, and other requirements, and the platform provisions a matching offer.⁶

The marketplace includes individual machines, multi-GPU nodes, and reserved clusters. Prime Intellect’s terms describe the service as a marketplace in which offerings may be supplied by third parties. Use of a machine can also be governed by the underlying cloud provider’s terms.⁷

Compute acts as a brokerage and orchestration layer. Hardware security, networking, storage, availability, and geographic location depend in part on the selected supplier. The common interface standardizes provisioning but does not make all of the underlying infrastructure identical.

Lab and inference

Prime Intellect Lab is a managed platform for reinforcement-learning post-training and evaluation. It became generally available in May 2026 and combines hosted training jobs, environments, rollouts, evaluations, checkpoints, and adapter deployment.⁸ Environments can be created with the company’s open-source Verifiers framework or selected from a public hub.

Lab abstracts the distributed training infrastructure from the customer and can charge for work in units such as processed tokens. It differs from a Compute instance, where the customer receives an accelerator environment and configures the training stack directly.

Prime Inference serves supported models through serverless or dedicated endpoints.¹ Dedicated endpoints reserve capacity for a workload while leaving the serving environment under Prime Intellect’s management. The product does not expose the same administrative access as a Compute machine.

Research

Prime Intellect conducts research into training models across geographically distributed and heterogeneous hardware. Its INTELLECT-1 project trained a 10-billion-parameter language model using machines in five countries across three continents, reaching a reported maximum of 112 H100 GPUs. The company released checkpoints, data, and a technical report in December 2024.⁹

INTELLECT-2, released in May 2025, applied globally distributed reinforcement learning to a 32-billion-parameter reasoning model.¹⁰ Subsequent work has included synthetic-data generation, reinforcement-learning software, and larger open models.

The company uses decentralized to describe computing supplied by multiple operators, training that spans several locations, and collaborative development of models or datasets. Customer deployments still use Prime Intellect’s central marketplace, accounts, scheduling, and billing. Some workloads run on conventional cloud providers through that control plane rather than on community-owned hardware.

References

Footnotes

1. Prime Intellect, official website. ↩ ↩² ↩³

2. Prime Intellect secures $5.5M in seed funding, Prime Intellect press release, 23 April 2024. ↩

3. $15M to build the open superintelligence stack, Prime Intellect, 28 February 2025. ↩

4. $130M Series A to build the open superintelligence stack, Prime Intellect, 8 July 2026. ↩

5. Imprint, Prime Intellect. ↩

6. Introducing Prime Intellect Compute, Prime Intellect, 1 July 2024. ↩

7. Terms of service, Prime Intellect. ↩

8. Prime Intellect Lab is open, Prime Intellect, 7 May 2026. ↩

9. INTELLECT-1 release, Prime Intellect, 2 December 2024. ↩

10. INTELLECT-2 release, Prime Intellect, 12 May 2025. ↩