Susana Schwartz
Nvidia shows further confidence in CoreWeave
Susana Schwartz
Technology Editor
RCRTech
AI Infrastructure Top Stories
Higher voltages in DCs: Dell’Oro’s Alex Cordovill discusses the “next big headline,” which he thinks will be higher voltages in power distribution, up to switch gear and closer to the rack in AC power and also DC power distribution within data center.
OpenRAN myths and realities: In 2019 and 2020 the wave of 5G contracts did not include OpenRAN in RFPs, but now, “75% of contracts up for renewal include OpenRAN,” says Teral Research chief analyst Stéphane Téral.
Colocation a key to growing in APAC: Colocation is a major focus in APAC, says Cushman & Wakefield’s Pritesh Swamy, which could mean the region will overtake U.S. data centers in terms of colocation capacity by 2029.
AI Today: What You Need to Know
CoreWeave to adopt Nvidia CPU, storage: NVIDIA and CoreWeave will accelerate AI factory buildouts, with CoreWeave agreeing to build out more than 5 GW of AI factories by 2030, and NVIDIA investing $2B in CoreWeave stock.
SoftBank halts acquisition of Switch: SoftBank has halted talks about a $50 billion acquisition of U.S. data center operator Switch, which was a part of a plan to roll out energy-efficient compute for partner OpenAI and its Stargate project.
Nvidia’s Huang in Shanghai: Nvidia CEO Jensen Huang today arrived in China, meeting with Shanghai-based staff at a time when Beijing is expected to allow imports of the company’s H200 AI chips. Discussions centered on product pipeline.
Google DeepMind says ‘no bubble’: CEO Demis Hassabis said “we are seeing more usage than ever,” but warned that “overheated funding in emerging startups is unsustainable.” He acknowledged the challenge of advanced chip shortages.
Samsung next-gen AI accelerators: After passing recent quality tests, Samsung will in February begin large-scale shipments of its HBM4 chips to key clients like Nvidia and AMD. The memory chips run at speeds of up to 11.7 Gb/s.
Machine learning model failure: MIT says it’s critical to move beyond overly aggregated machine-learning metrics, with new research detecting hidden evidence of mistaken correlations and new methods to improve accuracy.
