AI chips pull on infrastructure

Nvidia is pushing deeper into the machinery behind artificial intelligence, pairing new India data-center deals with fresh bets on the chips that feed those facilities. (cnbc.com) In India, Nvidia said on February 18 that it was working with cloud and infrastructure groups including Yotta, Larsen & Toubro, and E2E Networks, while CNBC reported the company was also continuing efforts to build domestic data centers. A New Delhi official cited by CNBC said India expects as much as $200 billion of data-center investment over the next few years. (cnbc.com) At the same New Delhi summit, Nvidia announced partnerships with Reliance New Energy, Hero MotoCorp, Larsen & Toubro Semiconductor, Tata Consultancy Services, Infosys, Persistent Systems, Tech Mahindra, and Wipro around graphics processing units, software, and industrial artificial intelligence deployments. Nvidia’s South Asia head Vishal Dhupar said India was “one of the most important markets” for the company. (livemint.com) The hardware mix is shifting inside those facilities. Nvidia said on March 16 that its new Vera central processing unit is built for “agentic” artificial intelligence systems and delivers 50% faster performance with twice the efficiency of traditional rack-scale central processing units. (investor.nvidia.com) That matters because artificial intelligence data centers do not run on graphics chips alone. CNBC reported on March 13 that Nvidia now says central processing units are “becoming the bottleneck” as newer systems move data, coordinate software agents, run code, and manage tools around the main model. (cnbc.com) Nvidia is also backing outside chip designers that could widen that infrastructure stack. TechCrunch reported on April 11 that SiFive raised $400 million at a $3.65 billion valuation, with Nvidia joining the round as the startup pushes RISC-V central processing unit designs into artificial intelligence data centers. (techcrunch.com) SiFive’s pitch is that it licenses processor blueprints rather than selling finished chips, giving cloud companies and server makers more room to customize systems. TechCrunch reported that SiFive’s designs are being built to work with Nvidia’s CUDA software and NVLink Fusion server architecture. (techcrunch.com) The India push sits inside a larger buildout race. CNBC reported that American technology companies including Amazon, Microsoft, and Google have committed more than $50 billion toward artificial intelligence infrastructure and chips in India, while Adani has announced plans to invest $100 billion in renewable energy-powered, artificial intelligence-ready data centers. (cnbc.com) Nvidia’s latest moves point to the same constraint from two directions: more demand for artificial intelligence computing, and more pressure to cut the power and cost of running it. The next phase of the boom looks less like a single-chip story than a full data-center redesign. (investor.nvidia.com)