Meta's AI Compute Breakthrough Sparks Optimism on Costs
Meta Platforms’ stock rose after a Bank of America Securities note suggested Meta could expand AI compute capacity more cheaply than Wall Street expected, citing an internal Meta memo relayed to Reuters that implies 2026 AI buildout costs could be much lower—about $22 billion per gigawatt on roughly 6.5 GW of capacity (5.5 GW in H2), versus prior estimates near $45 billion per GW. The implied lower capital outlay could ease investor concern over Meta’s massive AI investment. Separately, Reuters reported Meta plans to start manufacturing its Iris AI chip with Broadcom and TSM later this year, a move that could reinforce its long-term AI silicon roadmap even though the chip may not fully explain the 2026 cost improvements.
