For years, Nvidia has been synonymous with high-end graphics processing units (GPUs) powering advancements in gaming, data science, and now, artificial intelligence. But the company’s recent moves reveal a growing focus on meeting the needs of a broader AI market – one where not every application requires the most cutting-edge, expensive hardware.
Beyond GPUs: Diversifying Compute Solutions
Nvidia isn’t just selling GPUs anymore. Recent investments in low-latency chip technology and the introduction of standalone CPUs signal a strategy to capture customers who prioritize efficiency over raw power. This is especially relevant for “agentic AI” – software requiring real-time responsiveness rather than massive training datasets.
This shift was solidified with yesterday’s announcement of a multi-billion dollar deal between Nvidia and Meta. The social media giant will purchase a mix of Nvidia chips, including CPUs, to power its expanding AI infrastructure. This partnership isn’t new; Meta previously planned to acquire 350,000 H100 chips by late 2024, with a total of 1.3 million GPUs in its arsenal by 2025.
The latest expansion will see Meta deploy Nvidia’s CPUs at scale alongside Blackwell and Rubin GPUs, optimizing its data centers for both AI training and inference.
The Rise of CPUs in AI Workloads
What’s driving this demand for CPUs? The trend is agentic AI. As AI applications become more integrated into real-time systems, CPUs play a crucial role in managing data flow and ensuring responsiveness. One data center for OpenAI now requires “tens of thousands of CPUs” to handle the massive data generated by GPUs, a need that didn’t exist before AI’s explosive growth.
However, GPUs remain dominant. Meta’s CPU purchase is significant, but it’s still dwarfed by its GPU acquisitions. The CPU serves as a supporting component, ensuring the GPU-driven architecture doesn’t become bottlenecked.
Competition and Diversification
Nvidia’s moves come as other AI giants diversify their compute sources. OpenAI, Anthropic, and Google are all exploring custom chips or partnerships with AMD and Cerebras to reduce reliance on a single supplier. OpenAI, for instance, has deals with both Nvidia (potentially worth $100 billion) and AMD (up to 6 gigawatts of chips), as well as a $10 billion agreement with Cerebras.
The underlying issue remains supply. The demand for GPUs still outstrips availability, pushing companies to explore alternatives wherever possible. Nvidia’s recent acquisition of Groq, a chip startup specializing in low-cost inference, reflects this pressure.
Meta’s Massive Investment in AI Infrastructure
Meta plans to spend between $115 billion and $135 billion on AI infrastructure this year, up from $72.2 billion last year, underscoring the company’s commitment to AI-driven growth. Nvidia has long maintained that its hardware supports inference alongside training, with its business split roughly 40% inference and 60% training two years ago.
In conclusion, Nvidia’s partnership with Meta signifies a strategic shift toward meeting the diverse demands of the AI market. The company is no longer solely focused on high-end GPUs; it’s actively expanding its CPU offerings and solidifying its position as a comprehensive compute provider in an increasingly competitive landscape.
