Meta Platforms (META 2.41%) shocked investors when it announced plans to spend between $115 billion and $135 billion on capital expenditures this year, mostly on its artificial intelligence buildout. That's an 87% year-over-year increase in spending at the high end of its guidance.
Some of the biggest beneficiaries from Meta's massive budget are chipmakers. Meta recently announced two contracts with Nvidia (NVDA 1.25%) and Advanced Micro Devices (AMD 3.12%) to use their GPUs and CPUs in its data centers, each worth tens of billions of dollars.
Here are the details investors need to know.
Image source: Meta Platforms.
Different chips for different jobs
Meta's AI efforts are quite expansive. Its Llama large language model (LLM) gets a lot of headlines, but Meta's entire business is built on AI algorithms for determining exactly what content to show which users and when. With the massive scale of its operations (over 3.5 billion monthly active users across its Family of Apps), optimizing its data centers for peak efficiency is key. As such, it sees different chips as best suited for different jobs.
Its deal with Nvidia uses its entire stack of chips, including GPUs, CPUs, and its ethernet switches. The multi-year deal includes plans to co-design Meta's next-generation AI models to optimize them for Nvidia's hardware. As such, the deal appears to focus on developing data centers specifically suited for LLM training.
Meta's deal with AMD includes its MI450 GPUs and 6th-generation EPYC CPUs deployed on its Helios rack-scale architecture. Meta CEO Mark Zuckerberg said, "We're excited to form a long-term partnership with AMD to deploy efficient inference compute" in the press release announcing the deal. As such, the AMD chips will be more focused on inference.
Indeed, AMD has shown progress in AI inference performance relative to Nvidia. Its MI450 GPUs, when combined with the Helios rack system, are expected to produce better price performance versus Nvidia's Rubin platform. That's owed to its more advanced architecture and including more high-bandwidth memory in the chip package. Additionally, AMD's MI450 chips will be custom-designed for Meta's AI models, further optimizing performance.
As such, AMD is making significant progress in taking market share from Nvidia in AI inference. Analysts expect inference to account for the majority of AI compute starting this year, and AMD's multi-year deal with one of the biggest hyperscalers puts it in an excellent position to capitalize on the growing market segment.
However, AMD had to offer a significant discount to Meta to secure that long-term contract. The deal includes stock warrants to buy nearly 10% of AMD's shares, which vest as Meta deploys GPUs. The structure is similar to a previous deal between AMD and OpenAI for the same amount of stock. Under the agreement, Meta pays just $0.01 per share. If the deal proves mutually beneficial, it could work out for AMD shareholders despite the dilution caused by the deal.
Which stock is the better buy?
Meta's deals with Nvidia and AMD look beneficial for both chipmakers, but the biggest winner of them all might be Meta itself. Meta is securing access to compute capacity in an environment where it's remained constrained by its ability to procure and bring new chips online. During Meta's fourth-quarter earnings call, Meta CFO Susan Li said, "Demands for compute resources across the company have increased even faster than our supply." She also suggested that adding compute would have a positive expected impact on its core advertising business.
Additionally, working closely with both Nvidia and AMD, Meta has the opportunity to optimize its data centers for both training and inference for its LLM and machine learning models for its ad system. Given the scale of Meta's operations, every small optimization can have a significant impact on its bottom line.
NASDAQ: META
Key Data Points
Importantly, Meta has a lot to gain from improving its artificial intelligence algorithms and scaling the use of generative AI across its platform. It's already showing progress in using additional compute to improve its financial results. Management said it doubled the compute power it used to train its LLM-inspired Generative Ads Recommendation Model (GEM). It also provided more compute capacity to its sequence learning model, which tracks user behavior and information about each piece of content on the platform.
Finally, Meta is unifying its models across content formats, providing a wider aperture to understand its users' interests. The result was an increase in ad clicks and conversions across Facebook and Instagram.
The long-term potential is for generative AI to provide a step change in its advertising business. It has the power to create and test ad creatives for small businesses, lowering the barrier to entry to start advertising on Facebook and Instagram. That could bring millions of new advertisers to the platform.
Additionally, chatbots could push more businesses to use Meta's click-to-message ads to spark AI-powered sales and customer service chats on WhatsApp and Messenger. Management shared that Business AIs in Mexico and the Philippines, its early test markets, already surpassed 1 million conversations per week.
Despite securing closer relationships with two companies that could have some of the most significant impacts on its financial results, including a steep discount in the form of stock from AMD, Meta's shares have barely moved on the news of the deals. Investors can buy shares right now for less than 22 times forward earnings estimates. With the core business looking stronger than ever, it's even more attractive than buying AMD or Nvidia.
