Journalist Alex Heath, co-founder of the Sources newsletter and a longtime Meta chronicler, argued on CNBC this morning that investors are underestimating how close Meta Platforms (NASDAQ:META | META Price Prediction) is to shipping frontier-class AI models. His reporting, drawn from a recent conversation with CEO Mark Zuckerberg, reframes the debate around Meta’s aggressive capital spending at a moment when the stock is under pressure.
Heath said Zuckerberg told him the reason he published his recent AI policy op-ed was to establish his values before Meta unveils more powerful AI models: “He told me… that the reason he wrote that is he wanted to state the values he has before Meta releases much more capable models in the near future.“
Heath interprets that framing as a strategic signal about Meta’s direction. “T hat, to me, was a pretty definitive statement that he thinks that’s going to happen in the near future, and you’ve seen them catch up pretty quickly over the last year.”
Why Investors Are Punishing Meta’s $145 Billion AI Bet #
Meta shares currently trade around $546.35 as of this morning, with the stock down 15.46% over the past month and 17.14% year to date. The pullback followed a Q2 2026 earnings report with higher-than-expected capex costs.
Q2 capital expenditures were $31.1 billion, and management raised the 2026 capex outlook to $130-$145 billion. Total 2026 expenses are guided to $165 to $169 billion. Free cash flow fell to just $784 million in the quarter, a figure that has fueled Reddit debates. Q2 revenue reached $60.8 billion, up 28%, but GAAP operating margin compressed to 31%, and diluted EPS of $6.18 came in below Street consensus. A widely shared post argued investors were looking at the wrong number, noting that** operating cash flow still grew 25%.**
What All That Capex Spending May Actually Be Buying #
What, specifically, is all this spending buying? On the Q2 call, Zuckerberg defended proprietary model development directly. “Right now, the open source models are not as strong as the frontier models,” he said, adding that
“having kind of sovereignty over building your own models is going to be an important part of that stack going forward.”
Zuckerberg also said Meta is “in the process of scaling much larger and more advanced models” beyond the early scaling ladder of MuseSpark 1.1 and related releases. He argued the real prize is a significantly higher margin on selling intelligence rather than selling compute directly.
What to Watch Next #
Meta still needs to prove that its unprecedented AI spending can produce frontier models and attractive financial returns. Zuckerberg’s comments offer no firm timeline, but they suggest the company may be closer to a breakthrough than investors realize.
With Meta trading at 19 times forward earnings after a 17% year-to-date decline, any convincing model announcement at September’s Connect event on September 23 could quickly change the narrative. Analysts have an average price target of $754.14, with 47 Buy and 8 Strong Buy ratings against 7 Hold calls.
Contact [email protected] for any questions or corrections.