Summary of Key Points
Meta's advertising business continues to perform well (revenue increased by 28% in the second quarter), but almost all the profits are being invested in AI. The company had operating cash flow of $31.8 billion, yet capital expenditures (for building data centers and purchasing servers) amounted to $31 billion, leaving only $780 million in free cash flow—barely enough to cover dividends. A year ago, Zuckerberg bet on 28-year-old Wang Tao to restructure Meta's AI team, hoping to use the flagship model Watermelon to compete with OpenAI. However, the model has not been released yet, and the costs of the restructuring ( layoffs and high salaries for hiring experts) have increased R&D expenses by 67%. As a result, Meta's stock price has dropped by nearly 8% due to the lack of visible returns. Meta is borrowing money and seeking partners to share the costs, while also trying to improve its advertising services using existing AI technology and considering selling computing power as a backup option. However, Wall Street fears that AI could become another unprofitable venture, similar to the metaverse project.
1. The Advertising Business Is Thriving, but Profits Are Being Drained by AI
Meta's advertising business remains its main source of revenue: second-quarter revenue was $60.8 billion (up 28%), with ad revenue at $59.3 billion (up 27%). Both the number of impressions and the price per impression increased, exceeding Wall Street's expectations. However, these profits did not stay in Meta's pockets—operating cash flow of $31.8 billion was mostly invested in AI infrastructure: $31 billion went towards building data centers and purchasing servers, leaving only $780 million in free cash flow. This amount is not even enough to cover the company's $1.35 billion in dividends for the quarter. Even more strikingly, Meta has raised its capital expenditure forecast for 2026 from $125 billion to $130 billion, with a potential maximum of $145 billion, meaning it will invest over $100 billion in AI hardware in the next three years. The profits from advertising are clearly not enough to cover these costs.
2. Zuckerberg's Bet on Wang Tao: The Missing Flagship Model and Rising Sunk Costs
A year ago, when Llama 4 underperformed, Zuckerberg invested $14.3 billion in Scale AI and hired 28-year-old Wang Tao to reorganize the AI team. Wang Tao adopted a "war-time" approach, laying off 600 positions from the existing AI team (including renowned expert Yang Likun) and hiring high-priced talents from OpenAI and Google, while streamlining management layers. The consequences were immediate: R&D expenses soared by 67% to $21.6 billion, mainly due to the high salaries of these new hires. Total costs increased by 42%, far outpacing revenue growth of 28%.
The most critical issue is that the flagship model Watermelon has yet to be released. Although internal estimates suggest it has significantly higher computing power than Muse Spark and could compete with GPT-5.5, there have been no public evaluations or timelines for its launch. Meanwhile, competitors OpenAI and Google are continuing to advance rapidly. Every day Meta waits, more money is being spent (on salaries and data centers), increasing the risk of not getting a return on investment. Wang Tao has become Zuckerberg's biggest sunk cost—replacing him would mean admitting that previous investments were wasted, while continuing to wait means there is no clear outcome in sight.
3. AI Investments: Seeking Alternatives to Make Up for Losses
Meta is not putting all its eggs in one basket:
- Improving Advertising Efficiency: By using existing AI to enhance ad recommendations and sorting, Facebook's ad clicks increased by 8.3%, and conversion rates rose by 15.7%. Over 9 million small businesses are using AI to generate ad content, which helps keep the advertising business profitable while waiting for Watermelon to be ready.
- Selling Computing Power: Meta’s advanced computing centers can be sold to other companies at a profit. Additionally, millions of companies use its enterprise intelligence solutions weekly. However, Zuckerberg has not specified which of these initiatives will generate significant profits first. The CFO stated that even if the models are not cutting-edge, they can still improve existing products; if necessary, the company can simply sell computing power as a backup option. This approach provides a safety net but does not offer investors a clear timeline for returns.
4. Wall Street's Concerns: Could AI Become the Next “Metaverse”?
Meta’s metaverse division, Reality Labs, is still incurring substantial losses (losses of $4.6 billion in the second quarter). Wall Street fears that AI could lead to similar failures:
- Stock Price Decline: Despite better-than-expected financial results, the stock price dropped by nearly 8% because investors cannot see the endgame for Meta's AI investments. More money is being spent without any visible progress, and debt is increasing (the company issued $24.9 billion in long-term bonds, bringing total debt to $83.6 billion), with stock repurchases also halted.
- Ebbing Patience: Zuckerberg is patient with long-term projects (such as the metaverse), but Wall Street expects tangible returns. If Watermelon fails to be released on time or performs below expectations, AI could become another failed venture, leading to questions about Zuckerberg's decision-making ability.
5. Meta’s Delay in AI Development: A Late Start
Meta was actually early in its AI efforts (establishing the FAIR lab in 2013 and developing technologies like PyTorch and Llama). However, with the emergence of ChatGPT, Meta's AI resources were spread across multiple departments, and it wasn't until 2025 that AI was elevated to a company-wide priority—two years behind OpenAI and Google. Wang Tao’s restructuring aims to consolidate these scattered efforts and develop cutting-edge models. By the time Meta took action, its competitors had already established their models and gained a talent advantage, forcing Meta to invest even more to catch up.
The outcome of this major gamble is still uncertain: If Watermelon succeeds, Meta could return to the top ranks in AI; otherwise, the high costs associated with Wang Tao may exhaust Zuckerberg's patience and impact the company’s future. For now, Meta’s advertising business is holding its own, giving Zuckerberg time to wait. However, Wall Street's concerns are clearly reflected in the stock price.