Ask anyone what Nvidia produces and will probably say “GPU” first. For decades, the chipmaker has been defined by advanced parallel computing, and the emergence of generative artificial intelligence and the resulting surge in demand for GPUs has been a boon for the company.
However, Nvidia’s latest moves signal that it intends to attract more customers from the less computationally intensive part of the AI market – customers that don’t necessarily need the most powerful and powerful GPUs to train AI models, but are instead looking for the most capable ways to run agentic AI software. Nvidia recently spent billions to license technology from a chip startup focused on low-latency artificial intelligence processing, and has also begun selling standalone processors as part of its latest superchip system.
Yesterday Nvidia and Meta announced that the social media giant agreed to buy billions of dollars worth of Nvidia chips to provide computing power for its massive infrastructure projects – and Nvidia processors were part of the deal.
The multi-year agreement builds on a cozy, ongoing partnership between the two companies. Meta previously estimated that it would complete the purchase by the end of 2024 350,000 H100 tokens from Nvidia, to which the company will have access by the end of 2025 A total of 1.3 million GPUs (though it was unclear whether these would be all Nvidia chips).
As part of the latest announcement, Nvidia said that Meta will “build hyperscale data centers optimized for both training and inference to support the company’s long-term AI infrastructure roadmap.” This includes “large-scale deployment” of Nvidia processors and “millions of Nvidia Blackwell and Rubin GPUs.”
Notably, Meta is the first tech giant to announce it is making a large-scale purchase of Nvidia’s Grace processor as a standalone chip, which Nvidia says will be an option once it reveals the full specifications of its up-to-date Vera Rubin superchip in January. Nvidia also emphasizes that it offers technology that combines different chips in a “soup-to-nuts approach” to computing power, as one analyst put it.
Ben Bajarin, CEO and principal analyst at technology market research firm Imaginative Strategies, says the move signals that Nvidia realizes that a growing range of AI software now needs to run on processors, just like conventional cloud applications. “The reason the industry is so bullish on data center processors right now is because agent-based AI is putting new demands on general-purpose processor architectures,” he says.
AND the latest report from the Semianalytics chip bulletin emphasized this point. Analysts noted that CPU utilization is increasing to support AI training and inference, citing as an example one of Microsoft’s data centers for OpenAI where “tens of thousands of processors are now needed to process and manage petabytes of data generated by GPUs, a use case that would not otherwise be required without AI.”
Bajarin notes, however, that processors are still only one element of the most advanced AI hardware systems. The number of GPUs Meta buys from Nvidia still outweighs the number of CPUs.
“If you’re one of those hyperscalers, you’re not going to run All inference on CPUs,” says Bajarin. “You just need to make sure whatever software you’re using is fast enough on the CPU to be able to interact with the GPU architecture, which is actually the driving force behind that processing. Otherwise, the CPU becomes a bottleneck.”
Meta declined to comment on the expanded deal with Nvidia. During a recent earnings call, the social media giant said it plans to dramatically boost its AI infrastructure spending this year to $115 billion to $135 billion, up from $72.2 billion last year.