Nvidia's Rubin CPX plans clouded as Groq gains bigger inference role - digitimes

NVIDIA's Rubin CPX inference-focused graphics processor project faces uncertainty as supply chain activities around the product appear to have stalled, according to The Elec. This development reflects... Read original →

The stalling of NVIDIA’s Rubin CPX isn’t an isolated hiccup but a symptom of strained advanced packaging capacity compounded by geopolitical friction. Technically, any delay accelerates cloud providers’ shift toward alternatives—Groq’s LPU or custom ASICs—eroding GPU’s software moat in inference. Compliance-wise, U.S. export controls force NVIDIA to bake in regulatory redundancy early in design, inflating BOM costs and extending validation cycles. Groq is capitalizing on this by targeting high-value sectors like finance and autonomous driving with deterministic latency—a key differentiator. Over the next 18 months, the inference market will fragment: hyperscalers’ in-house chips, novel architectures, and legacy GPUs will coexist. Crucially, supply chain resilience—not peak throughput—will dominate customer decisions. Allocation of advanced packaging capacity in Taiwan, China and South Korea will dictate who can ship at scale by 2027.