DataCrunch.io

DataCrunch provides cost-effective GPU services "as-a-service" to support AI processing. The company offers on-demand and long-term GPU instances on its infrastructure, featuring NVIDIA and AMD cards. Additionally, it provides custom-built GPU clusters with a specialized software stack and serverless inference options that allow autoscaling and rapid deployment with minimal setup.

As of October 2024, DataCrunch’s data centers were strategically located in Finland and Iceland, given its focus on sustainability. While DataCrunch does not aim for the lowest latency, it offers sufficiently low latency for most compute tasks, aiming for around 100 ms rather than 10. DataCrunch used shared colocation facilities but planned to build proprietary data centers from 2025 (October 2024).

The company also offers Takomo for connecting and deploying AI models. Users can build custom pipelines by linking pre-trained machine learning models and run, test, and refine pipelines. For deployment, Takomo can generate multi-model APIs on a reliable cloud infrastructure. Its ML Playground feature lets users test and compare various models, including LLMs. 

Offers integrations with Stable Diffusion, OpenAI’s GPT, and Meta’s LLama models.

The company generates revenue through on-demand pricing for its GPU cloud, ranging from USD 3.59 to USD 28.72 per hour, based on the GPU model and the number of GPUs. Enterprises can also reserve specific GPU clusters (16x NVIDIA H200 cluster, etc.).

Pricing for Takomo varies per tokens, images, and audio, ranging from ~USD 0.0045 to USD 0.002.

Key customers and partnerships

As of October 2024, the company has increased its annual revenue by 250% and serves customers, including developers at OpenAI, Sony, and various startups such as Freepik.