RDMA

Oracle’s layoffs matter to network engineers because they show exactly where hyperscaler money is moving: away from headcount and toward AI data center capacity built on ultra-low-latency GPU fabrics. According to CNBC (2026), Oracle is cutting thousands of jobs while TD Cowen estimated the move could free $8 billion to $10 billion in cash flow, and according to Oracle (2026), OCI Supercluster is now positioned for up to 131,072 GPUs with cluster latency as low as 2.5 microseconds using RoCEv2-based networking. ...

NVIDIA Spectrum-X is the platform that proved Ethernet can compete with InfiniBand for AI training workloads — and it’s winning. By tightly coupling Spectrum-4 switch ASICs with BlueField-3 SuperNICs, Spectrum-X achieves 1.6x better AI workload performance than off-the-shelf Ethernet while maintaining the cost, ecosystem, and operational advantages that made Ethernet the standard for everything else in the data center. Key Takeaway: Spectrum-X is not faster Ethernet — it’s a fundamentally different architecture that ports three InfiniBand innovations (lossless transport, adaptive routing, in-network telemetry) to Ethernet, and network engineers who understand these mechanisms will design the AI fabrics of the next decade. ...

RoCEv2 (RDMA over Converged Ethernet version 2) has emerged as the dominant networking technology for AI data centers that don’t need absolute peak performance at any cost. For most GPU cluster deployments in 2026, properly configured Ethernet with RoCEv2 delivers 85-95% of InfiniBand’s training throughput according to industry benchmarks — at significantly lower cost and with skills that network engineers already have. InfiniBand still wins for the largest training clusters, but Ethernet is closing the gap fast. ...