Arista Sees 20–25% Of New AI Datacentre Deployments Using AMD Accelerators

AMD makes measurable inroads into AI datacentre networking

Arista Networks’ chief executive tells investors that deployments which were “pretty much 99% Nvidia” a year ago are now shifting, with Arista seeing roughly 20%–25% of new deployments favouring AMD accelerators. The move underscores a broader rebalancing in the AI infrastructure stack as hyperscalers and cloud providers evaluate alternatives to a long-standing Nvidia-centred architecture. AMD and Arista are already collaborating on customised AI clusters, a partnership announced late last year that aims to optimise networking and server designs around AMD’s accelerator roadmap.

The change is material for AMD’s competitiveness in the AI accelerator market because Arista’s Ethernet switching fabric plays a central role in linking high-performance chips for both training and inference. By aligning switch design and software with AMD accelerators, customers can deploy integrated systems that match the performance and scalability of Nvidia-based stacks while diversifying supplier risk. That technical alignment — including software, telemetry and performance tuning — is likely to accelerate adoption of AMD silicon in large-scale AI deployments where validated system stacks shorten integration time.

For AMD the development is more than a one-off design win: it signals a pathway to repeatable, validated reference architectures that can be offered to cloud providers, enterprises and system integrators. As customers seek alternatives to vendor-locked solutions, partnerships that co-design network and compute layers can be decisive. At the same time, Google’s tensor processing units remain an alternative route for some customers, keeping competitive pressure on all accelerator vendors to demonstrate end-to-end system performance and manageability.

Networking strategies reshape vendor dynamics

The shift follows Nvidia’s decision to develop its own networking stack, Spectrum‑X, which is winning design engagements at companies such as Meta and Oracle and prompting some customers to specify Nvidia less often. Arista is diversifying its customer base and product integrations in response.

Analysts say the trend does not spell immediate trouble for incumbents but changes procurement dynamics. Ben Bajarin describes the move as “not the end of the world” for Arista, noting Nvidia is being designed in less often, at least to a degree. Nvidia and AMD do not provide immediate comment on these shifts.