Nvidia just handed Equinix another reason to matter in the AI data center gold rush. The chipmaker announced Wednesday a new partnership with Equinix and Together AI designed to help enterprise customers run open-model inference without building their own GPU-packed data centers, a move that could reshape how mid-market companies access cutting-edge AI compute.
The AI infrastructure land grab just got another twist. Nvidia announced Wednesday it's teaming up with Equinix and Together AI on a new data center arrangement built specifically to help enterprise customers run open-model inference, the process of actually deploying and querying large language models rather than just training them. It's a small deal on paper, but it says a lot about where the AI compute market is heading next.
For months now, the AI infrastructure conversation has been dominated by hyperscalers. Amazon, Microsoft and Google have poured tens of billions into their own data centers, chasing Nvidia's GPU supply and locking enterprise customers into their respective clouds. Equinix has taken a different path. Rather than compete head-on for AI training workloads, the company has quietly built a business around being the neutral, carrier-dense colocation layer that lets enterprises plug directly into Nvidia's chips without committing to a single cloud provider.
That's exactly the niche this new partnership leans into. Together AI, known for its work optimizing open-weight models like Meta's Llama family, brings the inference stack, the software layer that actually runs a trained model efficiently. Nvidia brings the GPUs. Equinix brings the physical real estate, the fiber connectivity and the proximity to enterprise customers who don't want their AI workloads sitting three cloud regions away from their own infrastructure.
This matters because inference, not training, is quickly becoming the bigger cost center for companies deploying AI at scale. Training a model happens once, or periodically. Inference happens every single time a user sends a query, which means latency, cost per query and data locality start to matter enormously once an AI product actually ships. Enterprises running customer-facing AI features, think chatbots, fraud detection, or real-time recommendation engines, need that inference happening close to where their data already lives. That's the exact pain point Equinix is positioning itself to solve.
It also reflects a broader shift toward open-weight models. Companies that don't want to depend entirely on closed APIs from OpenAI or Google's Gemini have increasingly turned to open alternatives they can host and control themselves. Together AI has built its entire business around making that transition easier, and pairing that expertise with Nvidia's hardware and Equinix's global footprint of more than 260 data centers gives enterprises a turnkey path to running open models without standing up their own GPU clusters from scratch.
None of this happens in a vacuum. Nvidia has spent the better part of two years cementing partnerships across the entire compute stack, from hyperscalers to neoclouds to colocation providers, essentially making sure its chips are available everywhere an enterprise might want to buy compute. Equinix, for its part, has leaned into this strategy as its answer to a market where it can't out-hyperscale the hyperscalers. Instead, it's betting that neutrality itself becomes the product.
The timing also lines up with a broader capital spending surge across the AI sector. Data center investment tied to AI has been one of the largest drivers of tech capex all year, with estimates putting the total AI infrastructure buildout in the trillions of dollars over the coming years. Every player in that chain, from chipmakers to colocation providers to inference software companies, is racing to carve out a defensible position before the market consolidates.
What happens next will likely depend on how quickly enterprises actually adopt open-model inference at scale versus sticking with familiar closed APIs. If Together AI's optimization work proves the setup can match the performance and cost of hyperscaler-hosted models, Equinix's neutral colocation pitch could become a lot more attractive to companies wary of vendor lock-in. For now, this partnership is a bet that enterprises want choice in how they run AI, not just access to the chips that power it.
This deal is small in scale but big in signal. It shows Nvidia hedging its bets across every layer of the compute stack while Equinix stakes its claim as the neutral middle ground enterprises can trust with their AI workloads. Whether that bet pays off depends on how fast open-model inference becomes a mainstream enterprise choice rather than a niche alternative to the big closed AI platforms. For now, it's one more data point in a data center boom that shows no signs of slowing down.
