Meta shares dropped more than 4% during trading on September 28, reducing Mark Zuckerberg’s equity value by an estimated nearly 11 billion USD. The direct catalyst was a Goldman Sachs report questioning whether AI spending at current scales can generate commensurate revenue. This was the market reacting to profitability questions, not to a specific operational failure.
Goldman puts $300 billion on the table
According to the cited report, Goldman Sachs estimates that the largest AI infrastructure companies will spend around 800 billion USD on chips and data centers in 2026, and around 1,100 billion USD in 2027. These are analyst estimates, not corporate guidance. The necessary revenue is correspondingly massive: about 300 billion USD in recurring annual AI service revenue just to break even on that infrastructure, and nearly 1,000 billion USD to generate significant profit.
Meta’s own figures illustrate this strain. Capital expenditure in Q2 2026 reached approximately 31.1 billion USD, while free cash flow stood at only about 784 million USD for the same period. For the full year 2026, Meta is targeting 130 to 145 billion USD in capital spending, nearly double that of 2025.
Not a panic sell-off
Prior to this, Meta shares had climbed roughly 36% in September, bringing the company near a 2,000 billion USD valuation threshold before reversing course. That momentum was halted starting in the Friday session on September 25, when Goldman Sachs issued its report questioning the revenue required to break even on AI infrastructure. Selling pressure during the September 28 session was also driven by broader weakness across richly valued tech growth stocks, alongside elevated US Treasury yields.
Operationally, data does not indicate that AI has failed to deliver value. Meta’s Advantage+ advertising suite is generating an annualized revenue run rate of approximately 75 billion USD. Its GEM ad-ranking system reportedly lifted click-through rates by 8.3% and conversions by 15.7% on Facebook during early testing. These figures were provided by the company and have not been independently verified.
On the same day, September 28, Meta also announced the appointment of Chirantan “CJ” Desai, former CEO of MongoDB, to head its new enterprise platforms division, responsible for selling Meta’s AI models, agents, coding tools, and APIs externally. MongoDB stock fell more than 18% following the news. Desai previously agreed to work with Zuckerberg when Meta acquired a stake in Scale AI in 2025.
The real question lies in the next earnings report
Goldman’s figures of 300 billion and 1,000 billion USD represent industry-wide revenue milestones required to justify current expenditures. For Meta, the calculation is more nuanced because the company already boasts a massive, stable revenue foundation. Consequently, the primary question is not “can AI make money?” but rather “will this capital expenditure remain profitable if AI revenue falls short of expectations?” Meta holds the advantage of ongoing ad growth, enabling it to absorb years of elevated spending far better than peers without an established cash engine.
What warrants attention is not the single-day dip on September 28, but the upcoming earnings report: whether Meta adjusts its capital expenditure outlook of 130 to 145 billion USD, and whether free cash flow recovers enough to prove that the second-quarter figure of 784 million USD was merely a temporary phase rather than an early indicator of eroding profitability.