Meta Slashes AI Infrastructure Costs to $22 Billion Per Gigawatt, Half Previous Estimate

According to internal documents reviewed by Reuters, Meta has significantly reduced its AI computing infrastructure building costs to approximately $22 billion per gigawatt—nearly half of Bank of America Securities' prior estimate of $45 billion per gigawatt, analyst Justin Post said. BofA reiterated its "buy" rating on Meta Platforms (META-US) and maintained a $835 price target. The cost efficiency breakthrough, driven by improved data center construction and infrastructure optimization rather than new chips, addresses investor concerns about the company's $145 billion 2026 capital expenditure budget.

Meta plans to add 6.5 gigawatts of AI computing capacity in 2026, significantly exceeding BofA's prior forecast of 2.6 gigawatts. The company is also advancing production of its custom AI chip, codenamed Iris, in collaboration with Broadcom and Taiwan Semiconductor Manufacturing Company, with mass production set for September.