Nvidia GTC, supply moves, and chip leaks
Nvidia is primed to pivot to new AI CPUs and inference chips at next week’s GTC — a strategic shift analysts say targets “agentic AI” workloads — even as rivals and global customers find creative access to its top chips. The company’s supply-side tactics and reports that ByteDance gained access to Blackwell-class hardware via rental compute in Malaysia underline how export controls and supplier pressure are reshaping the AI chip race reported and reported.
GTC runs March 16–19 in San Jose and Nvidia’s own event materials say the show will host more than 30,000 attendees. biztechmagazine.com Pre‑conference reporting identifies a standalone AI CPU and a separate inference accelerator on Nvidia’s roadmap. siliconangle.com Analysts and industry pieces link those designs to Nvidia’s reported $20 billion Groq play and contend integration of Groq tech will shape the new chips. theoutpost.ai The Wall Street Journal’s reporting—summarized by multiple outlets—says ByteDance is working with Aolani Cloud to deploy roughly 500 NVL72 Blackwell rack systems (about 36,000 B200 accelerators) in Malaysia in a buildout valued at roughly $2.5 billion. tomshardware.com The same reporting says the hardware would be supplied via Aivres and that Aolani has been leasing H100 systems to ByteDance since February 2025. tomshardware.com Nvidia tells partners it evaluates and clears cloud providers through field operations, finance and compliance before shipments, part of its formal partner vetting process. tomshardware.com U.S. trade officials are meanwhile drafting wider export‑control frameworks that could require pre‑authorization for very large GB300/GB200 shipments, a regulatory shift cited by recent reporting. tomshardware.com
