A Meta internal memo reviewed by Reuters shows that on July 10, Bank of America Securities analyst Justin Post maintained a Buy rating on Meta Platforms (META) and set a target price of $835, saying the cost of building AI infrastructure has fallen significantly. Bank of America initially expected Meta’s AI computing capacity cost to be $45 billion per GW, but based on Meta’s disclosed deployment plan for 2026 and a $145 billion capital expenditure budget, the actual cost has dropped to about $22 billion per GW—roughly half of the initial estimate. Meta plans to add 6.5GW of computing capacity in 2026, far exceeding the 2.6GW that Bank of America previously expected. In addition, Meta will begin mass production of its in-house AI chip Iris in the fall of 2026, and will work with Broadcom and TSMC to reduce reliance on GPUs and further improve infrastructure efficiency.