Meta confirmed on July 13 that its Richland Parish, Louisiana data center -- internally code-named "Hyperion" and the company's largest facility anywhere -- will cost more than $50 billion, a figure that keeps redefining what a single AI data-center project can cost. The 5-gigawatt facility's price tag is now up from the $27 billion Meta disclosed in October 2025, when it formed a joint venture with Blue Owl Capital to help finance the buildout, which was itself already a major jump from Meta's original plan of roughly $10 billion for a 2-gigawatt facility.
That trajectory -- $10 billion to $27 billion to $50 billion-plus in under two years -- is the clearest single data point available right now for just how fast AI infrastructure cost estimates are inflating, even at a company with Meta's balance sheet and forecasting resources. It's not an isolated Meta phenomenon either: the company raised its overall 2026 capital-expenditure forecast to $125-145 billion earlier this month, and rival hyperscalers have made similar upward revisions through the year as land, power, cooling and construction costs all climb simultaneously.
The local economics are notable in their own right. Louisiana Governor Jeff Landry signed legislation exempting data centers built before 2029 from state and local sales tax for 20 years, though Meta still pays a 1% local sales tax on equipment purchases. Since construction began in December 2024, local businesses have collected more than $1.6 billion in Meta contracts, and Meta is separately investing over $1 billion in local infrastructure improvements. Richland Parish teachers received bonuses of up to $50,000 -- a 400% increase -- funded by the project's tax revenue, though local officials caution that windfall is tied specifically to the construction phase and will likely shrink considerably once the facility is built and the tax base normalizes.
The comparison set here is other hyperscaler mega-projects: Microsoft, Google and Amazon have all announced multi-billion-dollar single-site data-center commitments over the past two years, but Hyperion's cost trajectory -- doubling roughly every year since the original announcement -- stands out even against that backdrop. It's a live example of the AI infrastructure buildout's central tension: demand for compute keeps justifying ever-larger single-project commitments, but the actual dollar cost of delivering that compute keeps outrunning even recent estimates.
For infrastructure-focused investors and LPs, Hyperion's cost trajectory is a real underwriting signal, not just a Meta-specific curiosity: any fund or company with exposure to AI-data-center-adjacent assets -- power, cooling, land, construction -- should be modeling for continued cost inflation as a base case, not a tail risk. For local and state governments courting similar projects, the Richland Parish tax-revenue windfall is a useful but time-limited case study in how construction-phase economics can diverge sharply from steady-state operations once a facility goes live.
The bear case: cost overruns of this magnitude raise real questions about capital discipline across the entire hyperscaler AI buildout, and if Meta's other planned facilities show similar multiples, the company's elevated 2026 capex guidance could prove conservative rather than aggressive. What to watch next: whether Meta discloses cost trajectories for its other major data-center projects with the same transparency, and whether Richland Parish's tax revenue actually declines as officials predict once Hyperion's construction phase winds down.