Meta's Hyperion Louisiana data center swells to 5GW, $50B+ buildout backed by tax incentives

Meta's Hyperion data center in rural Louisiana is getting much larger and more expensive. The company announced that the site in Richland Parish will be a 5GW facility costing over $50 billion, up from the $27 billion figure revealed in October when Meta and Blue Owl Capital formed a joint venture to build and manage the site. Construction began in December 2024, and Meta reports over $1.6 billion in local contracts so far, with plans to invest over $1 billion in local roads, water, and wastewater infrastructure.

What happened

Meta's Hyperion data center, originally planned as a 2GW facility, is now set to scale to 5GW over several years. The company said the supercluster will be packed with GPUs and cutting-edge hardware tailored for AI workloads. A Meta spokesperson told CNBC that Hyperion should reach 2GW by 2030, but there is no timeline for the full 5GW completion.

Louisiana Governor Jeff Landry signed a 20-year sales tax exemption for qualifying data centers built before 2029 as part of an effort to attract Meta. Meta said it pays the full costs of energy, water, and related infrastructure so consumers are not charged. The project is part of a broader AI infrastructure rush, with Microsoft, Alphabet, and Amazon also pursuing similar state incentives.

Why AI builders should care

For AI builders, founders, and developers, Hyperion represents a massive increase in compute capacity dedicated to Meta's AI efforts. Meta Superintelligence Labs, led by Alexandr Wang, will have "industry-leading levels of compute and by far the greatest compute per researcher," according to CEO Mark Zuckerberg. This compute will likely be used to train and run Meta's open-weight models like Llama, which many builders rely on for fine-tuning and deployment.

The scale of Hyperion signals that Meta is betting heavily on AI infrastructure, which could lead to more capable models and potentially lower costs for inference if Meta passes on efficiencies. However, this is Meta's own infrastructure, not a public cloud. Builders should watch for any future API offerings or compute partnerships that might emerge from this capacity.

Practical implications

For now, the direct impact on AI builders is indirect. Hyperion will power Meta's internal AI workloads, including training next-generation models. If Meta continues its pattern of releasing open-weight models, builders may benefit from more powerful base models trained on this infrastructure. The project also highlights the growing importance of state-level tax incentives in shaping where AI compute is built, which could affect long-term energy costs and availability.

Builders should also note that Meta is covering infrastructure costs to avoid passing them to consumers, which may help keep Meta's AI services competitive. The 2GW target by 2030 gives a rough timeline for when this compute will come online at scale.

Caveats

The $50 billion+ figure is an estimate and may change as the project progresses. The tax incentives have drawn criticism, with some reports noting that Meta could receive $3.3 billion in tax breaks. Energy contracts with Entergy are largely shielded from public scrutiny, raising concerns about rate impacts on local residents. The timeline for full 5GW capacity remains undefined, and the project's ultimate cost and capacity could shift based on Meta's evolving AI strategy.