Is Axle AI the DeepSeek of the video industry?

January 28, 2025
Written by Sam Bogoch, CEO of Axle AI

This week, Chinese startup DeepSeek made worldwide headlines with its revolutionary approach to AI. DeepSeek’s R1 is a Large Language Model that has been engineered to run on mainstream GPU hardware – versions are even available for consumer-grade GPUs like Nvidia’s 4090. DeepSeek is able to achieve results comparable to vastly larger cloud GPU farms run by hyperscalers like Google, Meta, Amazon and OpenAI; it proves that you don’t need multibillion-dollar investments to achieve amazing results, and has benefitted from a lot of optimization and fine-tuning to fit within the constraints of a desktop computer.

For years, the video industry has been heralding the arrival of AI from those same cloud hyperscalers, but at a huge cost: videos uploaded to engines like Amazon’s Recognition or Adobe’s Sensei are kept by the vendors for future training of LLMs and Generative AI models. Axle AI’s new Tags 2.0 engine, like DeepSeek R1, can be run locally, on affordable consumer hardware costing as little as $1,500 with an Nvidia 4060 GPU. It includes advanced scene understanding, semantic search and trainable face recognition capabilities. This means that a video creation team can now set up its own local server to analyze, search and manage hundreds of terabytes of content, without sending that valuable content to cloud providers who would use it for their own purposes. In addition, cloud video engines like Rekognition charge $10-$25 dollars per hour of analysis, making their approach uneconomical for the large amounts of footage being captured today. Axle AI's Tags 2.0 engine is priced at an affordable monthly rate, and is integrated with a wide range of workflows and tools.

Like DeepSeek, Axle AI pursues an open-platform approach, supporting the development and integration of third party and open source models into the Axle AI Platform framework. Learn more at www.axle.ai and here and follow Axle AI on Linkedin.