Nvidia’s AI‑Chip Play
Analysts say Nvidia’s recent moves — plus acquisitions and new inference products — are cementing its lead across training and inference, even as China races toward 80% AI‑GPU self‑sufficiency by 2030. The hardware landscape is tightening into an AI‑chip arms race with major geopolitical implications. ( )
NVIDIA’s Blackwell inference stack centers on the GB300 NVL72 system, which NVIDIA says delivers up to 35x lower cost-per-token compared with its previous Hopper-generation platform. (nvidia.com) Blackwell Ultra (GB300) debuted in Q2 2025 and—according to supply-chain reporting—saw a volume ramp in mid‑Q3/Q4 2025 as hyperscalers began adopting GB200/GB300 configurations, with shipments projected to accelerate through 2026. (wccftech.com) NVIDIA announced the acquisition of GPU-orchestration startup Run:ai on April 24, 2024 and completed the deal at roughly $700M by December 30, 2024 after regulatory review, positioning the company to integrate Run:ai’s workload-management software into DGX and cloud products. (blogs.nvidia.com) In parallel, NVIDIA agreed to acquire model‑optimization firm Deci for about $300M in late April 2024, adding neural‑architecture search and model‑compression tech to its software stack. (marketscreener.com) NVIDIA also closed deals for model‑inference tooling, including the September 25, 2024 acquisition of OctoML/OctoAI, bringing additional runtime and deployment automation into its ecosystem. (marketscreener.com) Wall Street price targets for NVIDIA clustered above the $250s (consensus around $275.40, with some highs near $400), while supply‑chain coverage and analysts note China’s stated plan to reach roughly 80% AI‑GPU self‑sufficiency by 2030—two market signals shaping demand, pricing, and geopolitical supply strategies. (marketbeat.com)
