Meta is funding 10 natural gas power plants in Louisiana to power its Hyperion AI data center, generating 7.5 gigawatts — rivaling South Dakota's entire grid.
Meta announced it will fund seven additional natural gas power plants on top of three already committed, all to support its $27 billion Hyperion AI data center in Louisiana. The 10 combined plants will generate ~7.5 gigawatts, roughly matching South Dakota's full electricity capacity. This contradicts Meta's longstanding sustainability narrative, including its solar, battery, and nuclear energy purchases. Meta declined to comment when contacted by TechCrunch.
The energy infrastructure decisions behind large AI models directly shape future compute costs and availability. When hyperscalers lock into decade-long gas plant commitments, it signals that cheap, abundant AI compute is the near-term bet — not constrained, expensive GPUs. For developers building on Meta's infrastructure or pricing models around inference, this signals stable (if carbon-heavy) compute capacity through the late 2020s.
If you're running high-volume inference on Meta's Llama models via third-party cloud, benchmark your current per-token cost against AWS Bedrock and Fireworks AI this week — Meta's infrastructure buildout is a signal that Llama-based inference pricing will stay competitive.
Suggested exercise · not independently tested. Access requirements and costs may vary.
Run: curl https://api.fireworks.ai/inference/v1/chat/completions -H 'Authorization: Bearer $FIREWORKS_KEY' -H 'Content-Type: application/json' -d '{"model": "accounts/fireworks/models/llama-v3p1-70b-instruct", "messages": [{"role": "user", "content": "Summarize the energy challenges facing large AI data centers in 3 bullets."}]}'
Note the response latency and compare it to your current provider's p50 latency
If latency is under 400ms and cost-per-million-tokens is lower, log it as a viable switch candidate
Tags