Meta is continuing to spend heavily on the infrastructure needed to support its AI plans, including datacentre expansion and supply chain agreements to secure components for future capacity. In its latest quarterly earnings filing, the company said it has entered into multi-year cloud contracts, bringing total contractual commitments to $107 billion in Q1 2026.
For the quarter ending in March 2026, Meta reported revenue of $56.3 billion, up 33% from the same period in 2025.
The company also said capital expenditures, including principal payments on finance leases, will be $10 billion higher than previously expected because of rising component prices and additional datacentre costs. That puts CapEx in the range of $125 billion to $145 billion.
Chief financial officer Susan Li said the spending is aimed at building the training capacity needed for future models and, more importantly, the inference capacity required to deliver personal and business agents to billions of users, along with other AI experiences the company is developing.
During the earnings call, CEO Mark Zuckerberg said Meta is moving toward more capable and more scalable AI models as it balances model training with product rollout. He said the research team is focused on increasingly intelligent models built for business and personal agents.
Zuckerberg added that Meta’s next generation of advanced models is already in training and that the work will continue for the foreseeable future. He also said Meta’s product teams are now better positioned to build directly on top of the company’s models because those models are significantly stronger than before.
When asked how Meta uses large language models in its advertising business, Li said the models are too expensive to use directly at the size and complexity required for ad targeting.
Instead, Meta uses large language models to transfer knowledge to smaller, more efficient models. Li said these inference models face strict latency requirements because they must identify the right ad within milliseconds, which has historically limited how much they can grow in size and complexity.
To address that limitation, Li said Meta plans to introduce an adaptive ranking model later this year, with model complexity reaching the trillion-parameter level. She said the company has made advances in model architecture and has co-designed the system with the underlying silicon to preserve the sub-second speed needed to serve ads at scale.
Industry analyst Mike Proulx said Meta’s AI ambitions are still being funded almost entirely by its legacy advertising business, and that there is not yet any meaningful AI revenue. He added that the key question is whether Meta’s core ad business can continue generating enough cash while the company cuts headcount and shifts focus toward AI.
Proulx also warned that if Meta’s ad engine slows down, investor patience could fade quickly. He noted that a slight decline in daily active users is already drawing attention, and said Q2 will show whether that drop was temporary or the start of a broader trend.
