Meta’s AI infrastructure push runs from custom chips to data centres planned at gigawatt scale. Data Centre Magazine says the company has 33 data centres supporting its apps and Llama models, with major projects under way in Louisiana, Ohio and Texas.
Meta AI Watch analysis
What happened
A Data Centre Magazine feature on Meta’s data-centre strategy, published on 2 October, reports that Meta’s planned Hyperion campus in Louisiana could reach 5GW, with more than US$50bn in planned investment. Meta says it has already contracted more than US$1.6bn with local companies. The feature also describes Prometheus, a 1GW cluster in Ohio that Meta’s chief executive has said will come online in 2026.
In Texas, Meta and BlackRock announced a venture in July to develop a 1GW campus near El Paso. BlackRock funds hold 80% of the venture and Meta 20%; capacity is expected to start coming online in 2028. The feature also says Meta expects capital expenditure of US$130bn to US$145bn in 2026, and that four new generations of its MTIA custom AI chips are due within two years, co-developed with Broadcom.
Why it matters
These plans show the scale of the physical buildout Meta is pursuing to support AI, and how it is joining chip development, data-centre construction and outside financing. The company’s reported targets are not the same as capacity already operating: Hyperion is still planned, while the El Paso project is not expected to begin coming online until 2028.
Energy is part of the story too. The feature says Meta matched its 2024 electricity use with renewable energy, while an agreement with Entergy Louisiana will fund seven natural-gas plants, three grid-scale batteries and nuclear uprates. That mix makes the question of how AI infrastructure is powered harder to wave away with a sustainability slogan.
Our read
Meta’s investment is not just a bet on more servers. It is a bet that control over chips, sites and power arrangements will help it build AI capacity at enormous scale. The useful test is delivery: how much capacity actually comes online, when it does, and what the energy mix looks like in practice. Gigawatts make a good headline; functioning infrastructure is the harder trick.
What to watch
- Whether Hyperion’s planned capacity and investment translate into construction milestones.
- Whether Prometheus comes online in 2026 as described.
- How the El Paso venture progresses towards its 2028 start date.
- Whether the next MTIA chip generations arrive on schedule and support Meta’s AI workloads.
Discussion spark: Should Meta’s investment in its own chips, data centres and power supply be seen as sensible preparation for AI demand, or as a concentration of too much infrastructure in one company’s hands?
Sources and evidence
- Inside Meta's Data Centre Strategy for AI Infrastructure – Data Centre Magazine (2 October 2026, 09:18 UTC)
not affiliated with, endorsed by, or operated by Meta