Meta Expands Louisiana AI Supercluster to 5 GW; Brookfield Plans 1.2 GW Kentucky Campus
The global AI compute build-out surges with massive data center expansions and power deals, as hyperscalers prioritize rapid deployment despite increasing strain on power grids.
The story
The race for AI compute capacity saw significant new commitments this week, with Meta announcing a substantial expansion of its Hyperion campus in northeast Louisiana into a 5 GW AI supercluster. This initiative is expected to have broad implications for regional power systems.
Simultaneously, Brookfield and NextEra Energy unveiled plans to convert part of the former Paducah Gaseous Diffusion Plant in Kentucky into a 1.2 GW AI data center campus, featuring dedicated on-site power generation. Elsewhere, Google was identified as the force behind 'Project Tembo,' a 2.7 GW facility in Cheyenne, Wyoming, further solidifying the region as an emerging tech hub.
These developments underscore the industry's push to rapidly deploy large-scale infrastructure, even as rapid swings in AI data centers' power demands begin to strain vital equipment and rattle existing grids. Hyperscalers are increasingly focused on the speed of converting new infrastructure into revenue-generating compute, with Amazon indicating it will increase AI infrastructure spending this year, yet still expects capacity to trail customer demand.
Silicon
| Chip | Maker | What | For Whom |
|---|---|---|---|
| Blackwell B300 GPUs | Nvidia | Over 1,000 units deployed in a new 1.7 MW data center. | DeepInfra's international data center in Toronto |
| AI inference technology | AMD | Acquisition of AI chip startup Taalas to strengthen its position. | AMD's future AI inference products |
| NVIDIA MGX/DGX Spark systems, ExpertCenter Pro ET900N G3 | ASUS (built on NVIDIA architecture) | Full-stack AI infrastructure, including a 20 petaFLOPS deskside AI supercomputer. | Enterprises building on-prem AI factories |
The build-out
| Project | Who | Scale | Where |
|---|---|---|---|
| AI data center campus | Brookfield and NextEra Energy | 1.2 GW | Paducah, Kentucky, US |
| Hyperion campus expansion (AI supercluster) | Meta | 5 GW | Northeast Louisiana, US |
| AI campus ('Project Merlin') | Galaxy Digital | 74 MW (first phase) | McGregor, Texas, US |
| AI data center | Google ('Project Tembo') | 2.7 GW | Cheyenne, Wyoming, US |
| AI data center | DeepInfra | 1.7 MW with 1,000+ Nvidia Blackwell B300 GPUs | Toronto, Canada |
| AI data center campus | CyrusOne | $500 million, 460,000 square feet | Medina County, Texas, US |
| AI campus | EdgeMode, BlackBerry AIF, and the town of Mora | €3 billion ($3.4 billion) | Central Spain |
| Floating data centers | Samsung Heavy Industries (SHI) and Mousterian Corporation | 50 MW per unit | Texas and other US markets (initial deployments) |
Supply & policy signals
Samsung warned that the shortage of DRAM and high-bandwidth memory (HBM) could persist until 2028.
Implication: Continued tight supply for critical components in AI accelerators, impacting availability and cost.
Top four US hyperscalers guided capital expenditure (capex) to over $700 billion for 2026, a 75% increase from 2025.
Implication: Indicates aggressive investment in AI infrastructure, driving demand across the supply chain.
Bank of America Research estimates the total addressable market for AI accelerators will reach $1.1 trillion by 2030, up from an estimated $380 billion in 2026.
Implication: Highlights the massive projected growth in the AI chip market, suggesting sustained demand for manufacturing capacity.
New York announced a one-year moratorium on hyperscale data center facilities.
Implication: Regulatory pushback against the energy footprint of large data centers could shift investment to other regions or necessitate more sustainable power solutions.
Rapid swings in AI data centers' power demands are straining vital equipment, causing batteries, generators, and cooling systems to malfunction or wear out prematurely.
Implication: Suggests unforeseen operational costs and potential reliability problems for AI infrastructure, requiring more robust power management solutions.
SK Hynix and Samsung are evaluating chipmaking equipment from China's Advanced Micro-Fabrication Equipment (AMEC) for potential use in their Chinese factories.
Implication: Indicates a potential diversification of equipment suppliers in the memory sector, possibly influenced by geopolitical considerations or cost efficiencies.
What we'll be watching
- AMD's Jack Huynh will deliver the opening keynote at IFA 2026 on September 4, where the Ryzen AI MAX 400 portfolio is expected to launch.
- The Louisiana Public Service Commission is expected to take up a subpoena dispute related to Meta's power plant agreements on August 12.
- Elea Data Centers and AXIA Energia plan initial operations for the Amazon region's first AI-neutral data center in Belém, Brazil, in Q2 2027.
Reporting + analyst voices: grounded via Google Search at publish time.