Meta CEO Mark Zuckerberg Just Shared Great News for Nvidia, Broadcom, and Arista Networks Investors

In This Article:

Big tech companies are spending heavily to stay at the forefront of artificial intelligence. Budgets for building AI data centers filled with GPUs and network switches jumped higher in 2024. And a recent update from Meta Platforms (NASDAQ: META) CEO Mark Zuckerberg says 2025 could see an even bigger jump in spending.

Zuckerberg said in a Facebook post that Meta will spend between $60 billion and $65 billion in capital expenditures in 2025. That's up from $38 billion to $40 billion in 2024, a 60% jump at the midpoint of each forecast. Despite previous comments from Zuckerberg hinting at a big increase in spending for 2025, analysts had expected Meta's capital investments to come in around $51 billion this year.

Zuckerberg's comments suggest big tech isn't even close to being done with spending on AI development as the arms race continues into 2025. That's great news for a few important Meta suppliers: Nvidia (NASDAQ: NVDA), Broadcom (NASDAQ: AVGO), and Arista Networks (NYSE: ANET).

An aisle of server racks in a data center.
Image source: Getty Images.

The building blocks of a data center

At the core of data centers focused on large language model training and inference are GPUs. In Zuckerberg's post, he said Meta will have 1.3 million GPUs by the end of the year. That's an increase from the 600,000 estimate Meta gave for the end of 2024.

Nvidia is the largest supplier of GPUs to Meta, and it's likely to stay that way in 2025. Nvidia's latest chips based on its Blackwell architecture promise 30 times the performance of its previous generation, with 25 times less cost and energy consumption. No competitor currently comes close to that level of performance out of the box like Nvidia's chips. And in a market where big tech companies are racing to stay ahead of one another, paying up for Nvidia's chips is the fastest path to stay ahead.

But Nvidia isn't the only company Meta relies on for AI chips. Meta designs its own silicon, working closely with Broadcom, to handle specific tasks related to Meta's AI development and deployment. For example, the ranking and recommendation models that power Facebook and Instagram's feeds, Stories, Reels, video, and more are built on Meta's own AI accelerator chips. These custom-built chips are faster and more power-efficient than general-purpose GPUs.

Meta is working to expand the scope of its own chips to include generative AI, and it'll likely continue taking steps toward that goal in 2025. That's great news for Broadcom, which is a key technology provider for Meta in that endeavor.

All those GPUs and AI accelerators need an efficient way to communicate with one another and make sure data gets from one place to another quickly. That requires Ethernet switches. Ethernet switches are like traffic controllers for data centers. They route data from one server to another and make sure there's no congestion on the network slowing everything down.