NVIDIA Partners with Equinix and Together AI to Expand Enterprise Inference
nashnova research
Nvidia announced a three-way partnership with Equinix, the world's largest data-center colocation provider, and inference platform Together AI to offer enterprise open-model inference — with inference revenue already surpassing training, this deal closes the last gap in Nvidia's commercial loop.
What does each company bring?
The split is clean: Equinix provides the data centers, Together AI provides the inference software platform, Nvidia provides the GPUs and software stack.
This means → enterprise customers get a bundled "hardware + software + hosting" package instead of stitching three vendor layers together themselves.
The focus is open models — companies can run open-source large models rather than being locked into a handful of top model providers' APIs.
Why is Nvidia moving into inference now?
Per Nvidia's investor conference: roughly eighteen months ago, training and inference revenue were split about evenly; today inference has overtaken training, and the gap is expected to keep widening.
In plain terms = training is "teaching AI," inference is "putting AI to work." More enterprises are putting AI to work, so inference spending naturally outpaces training.
Emerging cloud providers now contribute over 50% of AI infrastructure revenue. Growth is shifting from the traditional hyperscalers toward a broader compute ecosystem.
What is the logic behind Nvidia's open-model bet?
CEO Jensen Huang recently published an open letter on the importance of open-source AI; it was co-signed by nearly every major AI company.
This means → Nvidia views AI models as a complement to GPUs — the more open models flourish, the more people need GPUs, and the more cards Nvidia sells.
Put simply = Nvidia doesn't sell the gold; it sells the pickaxes. Open source brings more miners, and pickaxe demand goes up.
Where does this deal sit in Nvidia's broader chessboard?
Training side: Nvidia has already licensed Groq's core team and LPU technology, planning to integrate them into the next-gen Vera Rubin architecture.
Platform side: Nvidia plans to acquire open-source AI platform Hugging Face for $12.9 billion.
This means → training, architecture, and the open-source platform are all in place. Teaming with Equinix and Together AI fills the last piece: inference deployment.
What does Equinix get out of this?
In the AI data-center buildout boom, Equinix has found a differentiated entry point: enterprise-grade open-model inference hosting.
This reflects a competitive shift in the data-center industry — from "who has the most floor space" to "who can deliver AI-specific solutions."
The validation checkpoints ahead are clear: whether inference revenue keeps scaling, and whether the open-model ecosystem achieves mass adoption.
市场有风险，内容仅供研究参考，不构成投资建议。