Meta Dropped $145 Billion on AI and Now Wants to Sell You the Leftovers — This Isn't a Strategy, It's a Confession
Meta Platforms has jolted Wall Street with the announcement of Meta Compute, a cloud services venture built on leasing out surplus GPU capacity from its AI infrastructure — a $125–$145 billion capital expenditure commitment for 2026 alone, nearly double the prior year's $72.2 billion. Bloomberg broke the story on July 1, 2026, and Meta's stock surged 9% on the day before giving back 5% two days later, a whipsaw that perfectly captured the market's conflicted feelings about whether the plan is actually executable. The venture would see Meta enter a cloud market controlled by AWS, Azure, and Google Cloud — three hyperscalers that collectively account for more than 65% of an $800 billion industry — by leasing GPU racks directly and offering API access to Meta's Muse Spark AI model. JPMorgan estimated that monetizing even 1GW of cloud capacity could generate $20 billion in annual revenue, a figure that convinced 43 of 55 covering analysts to maintain Strong Buy ratings with price targets raised to $825–$880. Whether Meta Compute proves to be a brilliant monetization of deliberate overbuilding or an inadvertent public admission that AI capital spending has spiraled beyond what internal demand can justify will be the defining investment question of the second half of 2026.