META’S AI BET IS NOW A BALANCE-SHEET BET Meta’s revenue grew 28% in Q2, driven almost entirely by advertising, yet the stock still dropped 10% after hours. That reaction says investors are no longer asking whether AI helps Meta—they are asking whether Meta can finance the infrastructure required to deliver it. A different company Meta used to be a cash machine that funded speculative projects entirely from operating profits. Now it increasingly resembles a leveraged infrastructure developer whose only proven repayment engine is advertising. The company expects roughly $137.5 billion of capex in 2026, enough to push free cash flow negative during the second half for the first time since its 2012 IPO. The pressure could intensify dramatically in 2027. Current projections assume approximately $174 billion of capex and more than $20 billion of cash burn, while analyst estimates stretch from $215 billion to $280 billion if Meta doubles computing capacity from 7 GW to 14 GW. This is more than an earnings problem. It changes the balance sheet. Long-term borrowings reached $83.7 billion at the end of Q2, despite Meta carrying minimal debt before 2022. That figure excludes tens of billions in off-balance-sheet obligations connected to data centers in Louisiana and Texas, where Meta has committed to leases and financial guarantees. Its 40-year bonds sold in May now yield close to 7%, suggesting the debt market is already demanding more compensation for financing this strategy. The return mismatch AI is clearly strengthening the core franchise, with better targeting and ad performance helping produce that 28% revenue growth. But the proven payoff remains incremental advertising improvement, while the spending is constructing an entirely new industrial base. Subscriptions, business chatbots, paid model access and compute leasing could create new revenue streams, but all remain early and unproven. Selling surplus capacity to major AI labs at a premium might improve utilization, but it is not the same as building an enterprise cloud platform comparable with AWS or Azure. Management says it can reassess actual computing needs from 2028 onward. By then, much of the capital, debt, leases and guarantees will already be committed. Bottom line: Meta can still win the AI race, but it is losing the luxury of being wrong cheaply. If monetization arrives late, the damage will extend beyond margins into debt costs, contractual obligations and reduced strategic flexibility for years.