Meta Forecasts Record $145 Billion in AI Data Center Spend
Tech giant elevates capital expenditure forecast to build out high-density compute infrastructure
Meta has dramatically raised its 2026 capital expenditure guidance, forecasting a range between $130 billion and $145 billion to fund massive AI-optimized data center campuses and grid-scale energy assets. The elevated target reflects the accelerating infrastructure costs of training next-generation large language models and operating gigawatt-scale AI factories.
Key details
According to the company's latest quarterly financial report, Meta is raising the lower bound of its full-year capital expenditure forecast by $15 billion, moving from an initial estimate of $115–$130 billion to a record $130–$145 billion.
The massive spending increase is directly driven by the high costs of building high-density data centers, purchasing thousands of advanced liquid-cooled GPUs, and securing long-term power generation. Key metrics highlight the scale of the expansion:
- Capital expenditure increase: An upward revision to $130–$145 billion, representing a significant year-over-year increase in hardware and physical facilities investment.
- Compute scaling: Funds will support the deployment of hundreds of thousands of next-generation accelerators to train future iterations of Llama models.
- Energy integration: The company is aggressively investing in dedicated grid connections and behind-the-meter generation, moving toward gigawatt-scale campus footprints.
As a result of this capital-intensive strategy, Meta’s quarterly free cash flow has faced downward pressure, contributing to a short-term drop in share price as investors adjust to the financial reality of the AI scaling race.
Why this matters
This development highlights the astronomical costs of scaling modern AI infrastructure. While software optimization has improved, the physical requirements for training frontier models remain locked in a capital-intensive trajectory. The $130–$145 billion expenditure demonstrates that compute power is the primary bottleneck in the AI race, forcing tech giants to operate at financial and physical scales comparable to national utility grids or public works programs.
Context
Meta's guidance update comes amid a industry-wide surge in AI-related infrastructure capital expenditure. Just days prior, Amazon reported that its annual capital expenditure is on track to hit $220 billion, driven in large part by AI-related hardware and data center buildouts. As tech companies compete to deploy the largest AI clusters, they are shifting away from traditional public grid models toward dedicated energy partnerships and custom-built, water-efficient liquid-cooling designs.
What happens next
Meta plans to continue ramping up infrastructure investments through 2027 to ensure sufficient capacity for its next-generation multimodal models. Analysts expect the company to increasingly focus on securing long-duration energy storage and behind-the-meter power generation to protect its facilities from localized grid shortages and municipal water limits. Investors will closely monitor Meta's subsequent quarterly reports to determine if the massive infrastructure buildout translates into commensurate revenue growth.
Source: Data Center Dynamics Published on AI Usage Global, author: AUG Bot
