Intel unveiled new rackscale AI infrastructure at Computex 2026 in Taipei, marking a significant push into the inference market that Nvidia currently dominates. The new architecture allows enterprises to scale inference and agentic workloads using Intel Xeon processors paired with SambaNova Reconfigurable Dataflow Units.
The announcement represents Intel’s clearest challenge yet to Nvidia’s integrated AI infrastructure stack. Rather than competing solely on GPUs, Intel is positioning disaggregated compute—where different hardware components handle specific tasks—as a more flexible alternative for enterprise AI deployments.
“We’re seeing enterprises want choice,” said Rodrigo, Intel’s AI infrastructure lead, at the announcement. “Not every workload needs an integrated black box. By offering disaggregated inference, we’re giving customers flexibility to optimize for cost, performance, or both.”
The rackscale architecture supports Intel’s Xeon 6+ processors alongside SambaNova’s SN-50 RDUs. The partnership with SambaNova is strategic: the startup’s dataflow architecture has gained traction in enterprise inference workloads where efficiency matters more than raw throughput.
The competitive landscape
Nvidia’s dominance in AI infrastructure faces pressure from multiple directions. While AMD continues pushing its MI-series GPUs, Intel’s approach with disaggregated compute targets a different market segment—enterprises already invested in general-purpose compute who don’t want to rip and replace their entire stack.
The disaggregated model also appeals to cloud providers offering AI as a service. Rather than selling pre-configured GPU instances, providers can now offer more granular compute options where customers pay only for what they use.
This announcement follows Intel’s broader AI strategy refresh earlier this year, which included new Gaudi accelerators and partnerships across the software stack. The Computex announcement adds a critical infrastructure layer that positions Intel as a full-stack alternative—not just a GPU alternative—for enterprise AI.
Whether enterprises embrace disaggregated inference over integrated solutions remains to be seen. But Intel’s move signals that the AI infrastructure market is increasingly competitive, with choice becoming a selling point rather than a compromise.