CoreWeave bets on serverless agent builder to woo penny-pinching enterprises

Because what enterprises really love are vague consumption-based pricing models

Rent-a-GPU outfit CoreWeave continued its push into the AI services arena on Wednesday with the introduction of a platform that aims to make reinforcement learning more accessible to enterprise customers.

Reinforcement learning (RL) is an approach to machine learning in which the model essentially teaches itself through trial and error by rewarding positive outcomes and penalizing negative ones. Over the past year, the approach has gained popularity as a means of fine tuning language models. For example, DeepSeek R1's ability to "reason" was achieved using RL.

CoreWeave's Serverless RL platform, launched on Wednesday, builds on two of CoreWeave's most recent acquisitions: OpenPipe, which specialized in using RL to build custom AI agents, and Weights & Biases, which offers, among other things, a serverless platform for GPU-accelerated workloads.

Serverless in this context makes a lot of sense because workloads can be distributed across available free or underutilized GPUs, eliminating resource stranding. Additionally, many AI workloads are inherently stateless, which means they don't need to store information from previous sessions in order to work.

According to CoreWeave, this eliminates the need for customers to manually provision virtual machines or bare metal servers in order to build custom AI agents using RL. Instead, they only pay for the tokens generated as part of the fine-tuning process.

If CoreWeave is to be believed, the approach is also nearly 1.4x faster and about 40 percent cheaper than using locally hosted Nvidia H100s.

For the moment, CoreWeave is offering the service through the Weights & Biases platform. However, moving forward, the neocloud aims to extend its AI services business into new areas.

Just this week, the company announced the acquisition of Monolith AI, which specializes in different kinds of AI. Instead of generative models, Monolith uses AI to accelerate what have traditionally been high-performance computing physics and engineering simulations.

CoreWeave's push into the AI services business is part of a broader effort to diversify its customer base, which for better or worse has been dominated by larger hyperscalers, cloud providers, and model builders. When the company filed to go public earlier this year, it warned in its IPO prospectus that 77% of its 2024 revenue came from two customers, and no more than 10% came from any other single customer. The case is probably less stark now, with Google and IBM as customers, but nonetheless, diversification remains vital for the debt-saddled company. ®