Nvidia’s GTC teases a CPU/AI chip pivot
Nvidia’s GTC (Mar 16–19) is being framed as a major hardware pivot — CEO Jensen Huang is expected to push new CPU and AI processors after a reported $20B bet on next‑gen chips argued. That shift matters for video platforms because new silicon and SDKs can change cost-performance for large-scale AI video workloads almost overnight.
Nvidia confirmed a roughly $20 billion agreement for Groq’s assets in late December [2025 reported]cnbc.com, and industry write‑ups note the deal gives Nvidia strategic access to Groq’s Language Processing Unit (LPU) inference architecture and personnel rather than a straight [merger analysis]techspot.com. GTC is scheduled for March 16–19, 2026, with Jensen Huang’s keynote set for March 16 at 11 a.m. [PT announced]blogs.nvidia.com; pre‑event previews from CNBC explicitly flag CPUs and inference accelerators as the narrative Huang is expected to push at the [show reported]cnbc.com. Groq‑derived LPU numbers cited in industry analysis put inference throughput in the 241–300 tokens/sec range on large language models, a metric proponents use to argue substantially higher inference throughput than conventional [GPUs reported]fintool.com; commentators say that throughput profile implies materially lower latency and cost per query for high‑concurrency workloads—an outcome that would change economics for large multi‑stream video inference such as real‑time captioning and scene‑level [analysis analysis]venturebeat.com. Nvidia’s own GTC materials list “chips, systems, software and ecosystem” as the show’s focus, signaling new SDKs and integrations could arrive alongside [silicon posted]blogs.nvidia.com; analysts and trade outlets warn that folding Groq tech into Nvidia’s stack could shift cloud instance price‑performance and intensify platform lock‑in, a dynamic to watch for procurement and deployment planning. venturebeat.com
