Report says Nvidia’s next “Kyber” AI rack is pushed to 2028, potentially giving AMD a window to gain traction

Nvidia’s roadmap for its next-generation AI data-center hardware is under new pressure after a market report said the company’s “Kyber” rack design is not expected until 2028. The claim matters less for one product date than for what it implies about timing, customer procurement cycles, and the pace at which competitors can challenge Nvidia’s lead in AI infrastructure.

Kyber, as described in the recent coverage, is Nvidia’s rack-level approach to supplying AI compute in data centers, bundling accelerators and system components so customers can deploy training and inference workloads at scale. In the report circulating among investors, the Kyber rack is portrayed as Nvidia’s next major platform step, and the delay therefore changes expectations for when buyers can standardize on the next wave of performance and efficiency improvements.

The same coverage links the Kyber timing to a separate industry narrative: whether Nvidia will face a wider opportunity for rivals to grow deployments before the next Nvidia platform arrives. One external piece of reporting noted that Goldman Sachs sees AMD as narrowing the gap with Nvidia and cited the Nvidia Kyber delay as giving AMD a “critical window” to expand AI deployments.

AMD’s competitive push is framed in that commentary as part of a broader shift in customer behavior. Rather than relying on a single supplier for an entire AI stack, data-center buyers increasingly want options, including reducing concentration risk and qualifying alternative ecosystems. The implication is that even if Nvidia keeps leadership, buyers may start diversifying their hardware purchases earlier than they otherwise would.

Nvidia remains central to the AI infrastructure buildout largely because of its software ecosystem and installed base, according to the same context described in market commentary. While the report and related commentary do not offer new evidence that Nvidia is losing existing customer commitments, they suggest that customer procurement timelines could be more flexible when a next platform is pushed out.

It is also notable that, in the items reviewed here, Nvidia did not provide a public statement or updated guidance that directly confirms the 2028 timing. The Kyber delay figure therefore should be treated as an account from market news and analyst discussion, not as an official company update, until Nvidia clarifies specifics through an investor presentation, regulatory filing, or primary communications.

More broadly, the story sits within an AI hardware market that has become unusually schedule-driven. Hyperscalers and other large customers plan capex around rack qualification timelines, datacenter builds, and power and cooling constraints, so delays at the platform level can ripple into evaluation cycles and qualification roadmaps for multiple vendors.

For investors and customers, the next checkpoints are straightforward: whether Nvidia addresses the Kyber timeline in any subsequent guidance or product communications, whether AMD’s ecosystem gains keep translating into measurable design wins or revenue indicators, and how system integrators and cloud customers describe their qualification schedules for next-generation AI racks.