Infrastructure

Not all networks can handle AI traffic – and experts are sounding alarms

GPU-as-a-service providers have scaled compute capacity but lack the network infrastructure to handle secure, low-latency AI workload distribution—leaving enterprises vulnerable if they evaluate suppliers on compute metrics alone.

Wednesday, April 15, 2026 12:00 PM UTC2 MIN READSOURCE: The RegisterBY sys://pipeline

Omdia research reveals that GPU-as-a-service (neocloud) providers have scaled compute capacity for AI workloads but face critical networking bottlenecks. Many providers with origins in cryptocurrency mining or content delivery lack the advanced networking infrastructure needed for secure, low-latency data movement across distributed AI environments. Enterprise customers are advised to evaluate neocloud suppliers on network capabilities, not just raw compute metrics.

Read original at The Register

Qwen3.6-27B: Flagship-Level Coding in a 27B Dense Model

Alibaba's Qwen3.6-27B delivers flagship-level coding performance at just 27B parameters, proving dense open-source models can match much larger competitors' capabilities.