Meta Platforms (NASDAQ:META) announced Monday it will expand its Hyperion data center in Louisiana to five gigawatts of computing power, pushing the total investment beyond $50 billion. The move provides concrete numbers for CEO Mark Zuckerberg's plan to sell AI computing capacity to external customers. Meta shares slipped about 1.3% to $661 in afternoon trading, following a 14.8% gain last week.
The investment equates to over $10 billion per planned gigawatt. For perspective, one gigawatt can power roughly 800,000 U.S. homes. Bank of America (NYSE:BAC) analyst Justin Post estimates that investors currently value Meta's computing at just $4 billion per gigawatt, compared to $59 billion for Amazon (NASDAQ:AMZN) and $110 billion for Alphabet (NASDAQ:GOOGL). Post's own valuation case puts Meta at $12 billion per gigawatt, noting "significant upside considering specialized AI capacity that Meta is building."
BofA's latest estimate values Hyperion at $20 billion for five gigawatts, under half of what Meta says it has invested. Post's model, however, puts it at $60 billion. For all 14 gigawatts Meta aims to build by 2027, the two estimates diverge by $112 billion. The final outcome hinges on chip utilization rates, software sales, and construction spend—not break-even math.
Zuckerberg told Bloomberg last week that Meta doesn't have spare computing power right now but still sees a compelling opportunity in launching an AI cloud business, calling offers for limited processing power "compelling." The plan remains fluid and could shift. If Meta sells capacity, it would compete directly with Amazon Web Services and Google Cloud, which already have established customer bases, billing systems, and enterprise support.
Hyperion's upfront cost represents about 34% to 40% of Meta's total 2026 capital expenditure guidance of $125 billion to $145 billion. The Louisiana campus is a multi-year build, and Meta is not funding it entirely on its own. In the original deal, Meta retained about 20% of a $27 billion joint venture, with most of the remainder going to Blue Owl Capital (NYSE:OWL) funds. Meta has not disclosed how it will split the additional expansion funding.
Cost control will be nearly as important as rental income. Meta plans to roll out seven gigawatts this year and targets 14 gigawatts next year. The company will also begin production of its Iris AI chip in September, which will back up GPUs—the chips AI workloads typically run on. "You can't become an AI titan if you are dependent on another company for chips," said Mike Gualtieri, analyst at Forrester.
Meta has yet to prove it can attract enough external cloud business to generate solid margins. JPMorgan Chase (NYSE:JPM) analyst Doug Anmuth maintained his Neutral rating and $725 target on Monday. "We're optimistic on early signs of AI monetization beyond digital advertising," he wrote, but stressed that winning over developers and corporate clients is key. If customers are slow to ramp up, rental rates drop, or internal use surges, Meta could face significant costs without matching revenue.
Spending forecasts are adding pressure to that revenue hurdle. Morgan Stanley (NYSE:MS) analyst Brian Nowak raised his Meta capex estimate to $225 billion for 2027 and $250 billion for 2028. He expects hyperscalers—the largest cloud and data center operators—to spend $1.4 trillion in 2028. For investors, the focus now shifts from capacity plans to locked-in external demand and proof that each new gigawatt generates more revenue than it consumes.
