NVIDIA is extending the layout of the AI infrastructure from the training side to the inference side. The company has collaborated with Equinix and Together AI to provide open model inference services to enterprise customers, aiming to integrate the processes of model training, deployment, and operation into a cohesive whole.

Clear division of labor among three parties

Equinix is responsible for data center hosting, Together AI provides the inference platform, while NVIDIA supplies GPU and the software stack. The goal of this collaboration is to enable enterprises to directly run inference workloads for open models, rather than relying solely on a few leading model manufacturers.

This collaboration is also seen as NVIDIA's填补 in the inference ecosystem. NVIDIA has previously established a strong position on the training side and is now extending its reach into enterprise-level inference scenarios, expanding the coverage of its computing power ecosystem.

Inference business weight has increased.

NVIDIA CEO Jensen Huang recently emphasized the importance of open-sourcing AI. At the company's investor meeting, it was also mentioned that about 18 months ago, training and inference revenues each accounted for roughly half, but now inference revenues have surpassed training revenues, and this gap is expected to widen further.

The revenue from computing infrastructure contributed by emerging cloud service providers, which is marked as AI, has also exceeded 50%, indicating that the demand for computing power is shifting from traditional hyperscale cloud vendors to a broader ecosystem.

Additional information:NVIDIA has previously introduced the core team of Groq and their LPU technology through a technical licensing agreement, and plans to integrate these capabilities into the next generation of Vera Rubin architecture.