Meta is preparing a much larger AI infrastructure push in 2025, with Mark Zuckerberg outlining a plan that centers on data centers, compute capacity and a sharply expanded GPU footprint.
In a Facebook post Friday, Zuckerberg said Meta expects to spend $60 billion-$80 billion on CapEx in 2025. The spending is aimed primarily at data centers and at growing the company’s AI development teams.
A bigger CapEx plan for AI
The projected $60 billion-$80 billion range marks a major step up from Meta’s CapEx last year. According to the source, Meta spent $35 billion-$40 billion on CapEx last year, making the new plan around double that level.
That increase shows how central infrastructure has become to Meta’s AI strategy. Building AI systems at scale requires more than model research. It also requires physical facilities, large amounts of computing hardware and teams able to develop and operate those systems.
Meta’s stated focus is clear: data centers and AI development teams. Data centers provide the environment where compute can run, while AI teams determine how that compute is used across development work.
What 1.3 million GPUs signals
Zuckerberg said Meta expects its data centers to pack over 1.3 million GPUs by year-end. In AI, GPUs are a core part of the infrastructure stack because they provide the computing capacity needed for large development workloads.
The number matters because it gives a concrete view of the scale Meta is trying to reach. Rather than describing AI investment only in broad terms, Zuckerberg tied the plan to physical compute capacity inside Meta’s data centers.
The company is also planning to bring around one gigawatt of compute online this year. The source describes that as roughly the amount of power consumed by 750,000 average homes.
Together, those details frame Meta’s AI investment as a capacity race. The company is not only spending more money. It is trying to add more compute, more GPUs and more data center capability within the year.
Why data centers are the battleground
Meta’s investment comes as AI rivals are putting billions of dollars into their own infrastructure projects. The source points to Microsoft and OpenAI as examples of that broader push.
Microsoft plans to spend $80 billion on AI data centers in 2025. OpenAI is contributing to a joint venture, Stargate, that could yield it hundreds of billions of dollars’ worth of data center resources.
Those comparisons help explain why Meta is raising its own spending plans. In a market where AI development depends heavily on compute, access to data center resources can shape what companies are able to build and how quickly they can build it.
The source does not describe every part of Meta’s AI roadmap. But it does show the investment logic behind the company’s next phase: more capital spending, more infrastructure and more people working on AI development.
The practical stakes for Meta
For Meta, the plan places capital expenditures at the center of its AI ambitions. A larger CapEx budget can support the physical layer behind AI systems, including the data centers that house GPUs and the compute that runs development workloads.
The scale also creates a clear benchmark for the year. By year-end, Meta expects over 1.3 million GPUs in its data centers. This gives the company a measurable infrastructure target, not just a general statement of intent.
There are three main figures that define the plan:
- $60 billion-$80 billion expected CapEx in 2025.
- Around one gigawatt of compute planned to come online this year.
- Over 1.3 million GPUs expected in Meta data centers by year-end.
Those figures make the direction of travel difficult to miss. Meta is treating AI infrastructure as a large, near-term priority, and it is describing that priority in terms of spending, power and hardware scale.
A race measured in compute
The AI competition described in the source is not only about products or research announcements. It is also about who can secure the infrastructure needed to support advanced development.
Meta’s plan sits alongside Microsoft’s $80 billion AI data center plan for 2025 and OpenAI’s role in Stargate. Each example points to the same underlying reality: AI companies and their partners are investing heavily in the data center resources behind the technology.
Zuckerberg’s update gives Meta’s side of that race a sharper outline. The company is planning a much larger CapEx year, a major increase in compute and a data center GPU base that could exceed 1.3 million by year-end.