Meta Buys Billions in Nvidia Chips, Including Grace CPUs — SkimNews
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Meta agreed to buy billions of dollars' worth of Nvidia chips under a multiyear expansion of their partnership, including a "large-scale deployment" of Nvidia's CPUs and "millions of Nvidia Blackwell and Rubin GPUs" for hyperscale data centers built for both training and inference.
- Meta is the first tech giant to announce a large-scale purchase of Nvidia's Grace CPU as a stand-alone chip, an option Nvidia revealed when it detailed the full specs of its new Vera Rubin superchip in January.
- Meta plans to dramatically increase its AI infrastructure spending this year to between $115 billion and $135 billion, up from $72.2 billion last year, and had previously estimated it would have 1.3 million GPUs total by the end of 2025.
- Nvidia spent $20 billion in December to license technology from inference-focused chip startup Groq—its largest investment to date—signaling a push into low-latency AI computing beyond its flagship GPUs.
- AI labs are diversifying chip sources as Nvidia's customers strain for compute: OpenAI signed a deal with Nvidia worth up to $100 billion, partnered with AMD for up to 6 gigawatts of chips, and announced a $10 billion deal with Cerebras for 750 MW of ultra-low-latency AI compute.
- Analyst Ben Bajarin, CEO of Creative Strategies, said the Meta deal reflects that agentic AI is driving new demand for general-purpose CPUs in data centers, though he cautioned GPUs remain the driving force—Meta's GPU purchases still outnumber its CPU orders.
Why it matters: Meta's commitment to buy Nvidia's Grace CPUs as stand-alone chips—combined with a planned $115B–$135B in AI infrastructure spend this year—validates Nvidia's "soup-to-nuts" bet that CPUs, not just GPUs, will be essential infrastructure for agentic AI, locking Meta deeper into Nvidia's ecosystem as rivals like OpenAI diversify toward AMD, Cerebras, and custom silicon.
Ask SkimNews
