Meta Commits $50 Billion to Louisiana Datacenter, Explores Compute Leasing to Anthropic
The global AI compute build-out continues at massive scale, with hyperscalers now exploring new models to monetize their vast infrastructure investments and manage power demands.
The story
Meta Platforms is significantly expanding its AI infrastructure, announcing a $50 billion investment to grow its Richland Parish, Louisiana datacenter to 5 gigawatts of compute capacity. This massive expansion reflects the ongoing demand for AI processing power within hyperscalers.
Concurrently, Meta is reportedly in preliminary discussions to lease up to $10 billion worth of its computing infrastructure to AI research lab Anthropic. This potential deal signals a strategic shift for Meta, allowing it to monetize excess capacity and transform a capital-intensive liability into a revenue stream.
The company recently appointed former Amazon Web Services executive Dave Brown as its new head of infrastructure, indicating a deeper commercial intent for its compute resources. This dual approach of aggressive build-out and external monetization highlights the evolving economics of AI infrastructure.
Silicon
RTX Spark (formerly N1X)
Maker: Nvidia
What: 20-core ARM-based CPU, engineering sample benchmarked on Cinebench 2026 with 80-95W power limits.
For Whom: Laptops, mini PCs, and compact AI workstations, including a Microsoft Surface Laptop Ultra prototype.
The build-out
| Project | Who | Scale | Where |
|---|---|---|---|
| Richland Parish Datacenter Expansion | Meta Platforms | $50 billion investment, 5 GW compute capacity | Richland Parish, Louisiana, US |
| Taiwan Chiayi Science Park Phase 2 | TSMC | Additional advanced packaging plants | Chiayi Science Park, Taiwan |
| Statewide Datacenter Moratorium | New York State | One-year pause on new datacenters 50 MW or larger | New York, US |
Supply & policy signals
China's IC exports reached $177 billion in H1 2026, up 96% year-over-year.
Implication: Indicates significant growth in China's semiconductor output and market presence, potentially driven by domestic demand and increased memory prices.
ASML plans to increase its low-NA EUV and DUV immersion capacity by approximately 30% in 2027.
Implication: Suggests anticipation of continued strong demand for advanced chip manufacturing equipment, crucial for leading-edge node production.
Richard Reyle of Questar Capital Partners advises investors to take profits in AI hardware and memory stocks, citing Meta selling excess compute as a warning sign.
Implication: Signals growing investor skepticism about the sustainability of current AI hardware valuations and potential softening in demand or oversupply of compute capacity.
What we'll be watching
- Moonshot AI's Kimi K3 model open weights release by July 27, 2026.
- Further details and deployment timelines for Nvidia's Vera Rubin AI chip platform throughout 2026.
- The outcome of Meta's preliminary discussions to lease computing capacity to Anthropic.
- New York State's assessment and potential policy changes following its one-year datacenter moratorium.
Reporting + analyst voices: grounded via Google Search at publish time.