Prime Intellect raises $130M at $1B valuation for its AI training platform

Artificial intelligence training startup Prime Intellect Inc. has raised $130 million in funding from a group of prominent investors. 

The consortium included Nvidia Corp.’s NVentures, Intel Capital and Dell Technologies Capital. They were joined by more than a dozen others, including Cloudflare Inc. Chief Executive Matthew Prince. Prime Intellect stated in its Tuesday funding announcement that the round values it at $1 billion.

Developers often don’t train AI models from scratch but rather customize an existing open-source algorithm. Prime Intellect operates a cloud platform specifically optimized for that task. The platform customizes AI models using two open-source toolkits, Verifiers and Prime-RL, that the company released last year.

Developers train AI models in virtual environments that simulate the tasks they will be expected to automate for end-users. For example, an online research agent might be placed in a sandbox that contains a browser. AI training environments can also simulate code repositories, support ticket processing tools and other systems.

Prime Intellect’s Verifiers toolkit reduces the amount of time required to create training environments. Additionally, it provides access to a catalog of user-created AI sandboxes that remove the need to build everything from scratch.

It runs customers’ training environments on managed cloud-based graphics card cluster. The company optimizes the infrastructure using Prime-RL, the open-source project that it released alongside Verifiers last year. Prime-RL can parallelize an AI training run across thousands of graphics cards to speed up processing.

The toolkit distributes training tasks across chips using an open-source technology called FSDP2. It’s a component of PyTorch, one of the industry’s most popular AI development frameworks. FSDP2 uses less memory than competing technologies and offers more customization options.

Prime Intellect supports multiple training approaches. Customers can fine-tune all of a model’s parameters or use a faster, more hardware-efficient method called LoRA. The latter approach extends AI models with a small number of custom artificial neurons optimized for a specific task.

After developers train an AI model using Prime Intellect’s platform, they can deploy it on inference infrastructure managed by the company. The company says its graphics card clusters are optimized for LoRA-powered models. Developers can purchase inference capacity from more than 50 data centers through an auction-like system.

Last year, Prime Intellect used its platform to train an AI model called Intellect-3 that features more than 100 billion parameters. The company says that it outperformed similarly sized models across multiple reasoning benchmarks. Prime Intellect client Ramp Inc., a provider of corporate credit cards, built an AI model that can complete some spreadsheet search tasks more accurately than Claude Opus 4.6.

Prime Intellect says its platform is also used by about 6,000 other customers. The company’s annualized revenue topped $100 million ahead of this week’s funding round.

Image: Prime Intellect