▲All images in this article are sourced from the internet. Please contact us for removal if infringement occurs.

The Boss Who Paints the Best Pies

This article was first published in Shadow Memo by Mo Yingsheng.

On July 29, 2026, Meta released its Q2 2026 financial report. This report acted like a truth-revealing mirror, exposing the real embarrassment behind Zuckerberg's grand AI ambitions.

Revenue hit $60.8 billion, up 28% year-over-year (YoY), a new all-time high. However, net profit fell 14% YoY to $15.8 billion, far below market expectations of $18.5 billion. Free cash flow plummeted 91% from $8.55 billion in the same period last year to just $784 million, marking the lowest level in nearly four years.

More alarmingly, the company's Q3 revenue guidance midpoint stood at $62.5 billion, below analysts' average expectation of $63.2 billion. Following the report's release, Meta's after-hours stock price briefly crashed over 10%.

Revenue rising, profits falling, cash flow drying up, and stock price collapsing—this is the current state of Zuckerberg's claim that "AI is accelerating our core business."

On one side, the AI pie gets bigger and bigger; on the other, the funding black hole deepens. Is Meta's hundred-billion-dollar gamble betting on the future or repeating the metaverse's mistakes?

Earning Speed Can Never Catch Up with Burning Speed

Let's first examine what this financial report reveals.

The Good Side: The advertising engine still roars. Meta's ad business remains an undisputed cash cow. Q2 ad revenue reached $59.363 billion, up 27% YoY, with ad impressions growing 14% and average prices rising 12%.

Zuckerberg even stated bluntly during the earnings call: "In dollar terms, our ad business reported faster YoY revenue growth than any other company's reported ad business."

Currently, 9 million small businesses on Meta's platforms use at least one AI-powered ad creative tool, with AI-driven Advantage Plus end-to-end solutions generating an annual revenue run rate exceeding $75 billion.

The Ugly Side: Earning speed can never catch up with burning speed.

Q2 total costs and expenses reached $42.026 billion, up 55% YoY. Research and development (R&D) spending hit $21.7 billion, up 67% YoY, accounting for over one-third of revenue.

The bigger issue lies in capital expenditures (CapEx). Q2 CapEx soared to $31.08 billion, up 83% YoY.

The company raised its 2026 full-year CapEx guidance lower bound from $125 billion to $130 billion while maintaining the upper bound at $145 billion. This remains one of the most aggressive CapEx plans in the global tech industry.

This led to the shocking figure: operating cash flow of $31.86 billion minus $31.08 billion in CapEx left just $784 million in free cash flow.

In other words, nearly every dollar Meta earns gets devoured by AI infrastructure.

One analyst put it bluntly: "Nearly all cash generated is being consumed by AI infrastructure spending." Bloomberg's commentary was even more direct: "The company's Q3 revenue forecast disappointed. Investors remain skeptical about how Meta will recoup its massive AI investments."

More concerning is the trend. Some analysts point out that at current trajectories, Meta's free cash flow will likely turn negative this quarter. A tech giant with annual revenue exceeding $240 billion facing negative cash flow risks serves as a warning signal.

Meta's AI layout (strategic layout) can be summarized in four words: early starter, late arriver.

Zuckerberg doesn't underestimate AI. As early as 2013, Meta established FAIR (Facebook AI Research) lab, bringing in Yann LeCun, one of the three pioneers of deep learning. In terms of AI foundational research, Meta's accumulations hardly lag behind Google or Microsoft.

The problem lies in strategic resolve.

First Missed Opportunity: Misjudging the large model direction. When OpenAI released GPT-1 in 2018 and GPT-2 in 2019, Meta remained preoccupied with recommendation algorithm optimization. When ChatGPT exploded globally in late 2022, Meta only then woke up.

Not until 2023 did Meta hurriedly launch its Llama series models, attempting to carve out a space in the open-source large model arena.

Second Missed Opportunity: Llama 4's " flip the car " (spectacular failure) became a textbook case. According to media reports, during Llama 3's development, Meta exhausted all technical reserves to ensure its success.

This directly led to technical discontinuities when developing Llama 4. On today's most critical fronts of reasoning and Mixture of Experts (MoE) models, Meta had to restart exploration from scratch.

Worse still, sources claim Meta's AI team manipulated some Llama 4 benchmark results by using different model versions to optimize different test items. This incident directly caused Zuckerberg to lose confidence in the existing AI team.

Ultimately, Meta announced replacing Llama 4 with its self-developed Muse Spark, effectively abandoning its most distinctive open-source brand in the large model competition.

Third Missed Opportunity: AI Agent progress fell far short of expectations. On July 2, 2026, during an internal all-hands meeting, Zuckerberg sent a rare signal by admitting his miscalculations. According to meeting recordings obtained by media, Zuckerberg confessed that Meta's AI Agent development progress had "not accelerated as we expected" over the past four months.

He further stated that the company's massive organizational restructuring around AI "could have been cleaner," with executive-level misjudgments about the transformation's pacing.

From missing the large model wave to Llama 4's failure, then to AI Agents underperforming, Meta always seems "chasing" and "half a step behind" in the AI race.

The underlying reason: Zuckerberg's investment logic has never been "think clearly before acting" but rather "all-in first, then figure it out."

From cryptocurrency Libra to metaverse, then to current AI investments, Zuckerberg's style remains consistent—seeing a trend, going all-in, pivoting when it doesn't work, then going all-in again on new trends.

This "trend-chasing" investment approach always leaves him entering late after waves have already risen, never defining the Track (race track).

"Pie-Painting" Narratives

If Meta lags technically, Zuckerberg certainly leads in narrative skills.

During the Q2 earnings call, Zuckerberg painted a dazzling vision: "Soon, we'll have AI assistants working 24/7 to help you achieve goals, improve your life, health, relationships, and finances—whatever you desire."

He further predicted: "Whether setting the timeframe at five years or otherwise, I believe the likelihood of billions not using personal agents in the future is extremely low."

"We're the only major tech company whose primary goal is to put superintelligence directly into people's hands."

In a Wall Street Journal op-ed, Zuckerberg even proposed a grand "AI philosophy"—using personal empowerment as the source of prosperity creation, invention as the primary purpose of superintelligence, and checks and balances as the foundation of safety.

Sounds wonderful, right?

The reality: About 98% of Meta's revenue still comes from traditional advertising. AI subscription services are expected to contribute up to $3 billion by 2027—a negligible increment compared to Meta's annual ad revenue exceeding $240 billion.

Regarding enterprise AI assistants, although over 1 million businesses use Meta's commercial AI assistants on WhatsApp and Messenger, their overall revenue contribution remains limited. New businesses like compute rentals and large model APIs are still in their infancy.

The pie-painting skills grow stronger, but delivery capabilities consistently lag.

Does this scene seem familiar?

Five years ago, Zuckerberg similarly believed virtual reality was the future, even renaming the entire company from Facebook to Meta. That bet cost about $80 billion.

This year, Meta significantly downsized its metaverse division, cutting 10% of the team and quietly halting new Horizon Worlds development.

Today, the Reality Labs division continues burning over $4.6 billion per quarter. Since 2021, it has accumulated about $85.6 billion in operating losses.

One online comment hit the nail on the head: "Sorry, isn't this the guy who bet $80 billion that people would wear VR goggles for meetings?"

Another more bluntly stated: "Everything Zuckerberg bet on in the past few years has failed. Not sure his 'vision' is still trustworthy."

From Libra to metaverse to AI, Zuckerberg's narrative skills never disappoint, but his delivery capabilities never inspire confidence.

Invested $130 Billion, Yet Sees No Results

Meta's AI investment intensity cannot be overstated—$130-145 billion in 2026 CapEx, representing nearly geometric growth from $28.1 billion in 2023, $39 billion in 2024, and $72.2 billion in 2025.

For comparison, Sweden's 2024 national GDP was about $604 billion. Meta's annual CapEx alone approaches one-fourth of Sweden's entire economic output.

But why invest so much without seeing tangible results?

Money primarily goes to "infrastructure" rather than "products." The vast majority of Meta's CapEx flows into data centers, GPU clusters, and power infrastructure.

The company plans to double its total computing capacity to 7 gigawatts in 2026 and double it again to 14 gigawatts in 2027. These infrastructure projects require years from groundbreaking to operation, making short-term revenue generation impossible.

Meta CFO Susan Li also admitted that the company currently prioritizes computing capacity expansion for 2026-2027, with concrete deployment plans only considered for 2028 and beyond.

In other words, money invested now will only yield results—if at all—two to three years later.

AI's enhancement of core businesses is "invisible." Zuckerberg repeatedly emphasizes that AI improves ad recommendation efficiency and conversion rates, but this enhancement is indirect, gradual, and difficult to quantify precisely.

Unlike launching a new product that generates immediate market reactions, AI's impact silently permeates every algorithmic optimization detail. For investors, such "invisible returns" hardly alleviate anxiety about massive expenditures.

New business monetization paths are long and uncertain. Meta explores multiple AI monetization routes—large model APIs, compute rentals, enterprise AI agents, etc.

However, these businesses are either in their infancy or face fierce market competition. In cloud computing, Amazon AWS, Microsoft Azure, and Google Cloud have formed a stable triopoly.

As a latecomer, Meta faces significant challenges carving out a share in the compute rental market. Although Zuckerberg stated during the earnings call that "we've received substantial compute leasing offers at prices significantly above acquisition costs," he emphasized that "selling intelligence will continue to yield significantly higher profit margins than selling raw compute." The problem is that "selling intelligence" remains far from scaling (scaling).

Internal management chaos hinders execution efficiency. In May 2026, Meta conducted a massive restructuring affecting about 20% of employees—global layoffs of approximately 8,000 while transferring about 7,000 to AI-related departments.

However, Zuckerberg himself admitted internally that the layoff process was disorganized and uneven, with the restructured AI-oriented organizational system failing to quickly deliver value. Multiple engineers transferred to AI departments revealed intense work pressure (work pressure) in newly formed AI teams. Zuckerberg even acknowledged in a June internal memo that the company's AI transformation "paced too quickly."

EMARKETER senior analyst Jaime Minton's assessment was sharp: "These continuous product launches increasingly feel like shooting in the dark rather than advancing along a sustainable path."

From strategy to execution, infrastructure to products, organization to culture—Meta has issues at every level. No matter how much money gets invested, without resolving these fundamental problems, the outcome likely remains unoptimistic.

Will the History of the Metaverse Repeat Itself?

To understand Meta's dilemma in AI, we cannot ignore the 'lesson' from the metaverse.

In 2021, Zuckerberg renamed the company Meta and announced an all-in commitment to the metaverse. At the time, he painted an equally grand vision: billions of people would work, socialize, and entertain in the virtual world.

And the result? Five years have passed, and Reality Labs has accumulated losses of $85.6 billion, while the metaverse remains a distant concept for ordinary users.

Today, Meta's narrative logic in AI is almost identical to that of the metaverse back then:

Both start by drawing an extremely grand vision ('Billions of people will have personal AI assistants' vs. 'Billions of people will live in the metaverse');

Both involve investing astronomical sums of money ($130-145 billion in annual capital expenditures vs. cumulative losses of $85.6 billion);

Both go all-in before market demand has been validated;

Both rush to roll out infrastructure before core products are mature.

Mike Proulx, Research Director at Forrester, commented pointedly: 'This resembles Meta's missteps with the metaverse. Meta is once again burning money before demand is proven. The difference is that AI adoption and value are real. The technologies differ, but market patience is starting to look similar.'

This comment strikes at the heart of the issue: AI's value is real, but Meta's model of 'burning money first and figuring out how to make money later' is exhausting the market's remaining patience.

To be fair, Meta's AI layout (AI layout means AI strategy, but keeping HTML format, so not translating literally) is not entirely without bright spots.

AI's enhancement of the advertising business is real. With a 27% growth in ad revenue, 9 million small businesses using AI advertising tools, and an annual revenue run rate of $75 billion for Advantage Plus, these numbers show that AI is indeed strengthening Meta's core business.

Zuckerberg's statement on the earnings call that 'we are indeed a full-stack technology company' is not empty talk. From self-developed chips to Self built data center (self-built data centers, keeping HTML format), from self-developed large models to proprietary application scenarios, Meta's depth of layout (strategy) in the AI supply chain is unmatched by few companies.

NVIDIA CEO Jensen Huang even commented: 'No company deploys AI at a scale larger than Meta.'

But the problem is: scale does not equal return, and investment does not equal output.

When a company's free cash flow plummets from $8.5 billion to $700 million, when its stock price falls for 10 consecutive trading days and loses 13% of its market cap, when it admits that 'progress is far below expectations' in AI Agent development, the market and investors have ample reason to feel anxious.

Zuckerberg said on the earnings call: 'I understand this is a huge investment and a huge bet. My personal bet is that those who invest in this will be rewarded and feel very good over time.'

Meta's dilemma in AI is essentially a dilemma of 'strategic pace.'

Going all-in in the right direction is not wrong, but if the pace is wrong—if infrastructure is overbuilt before products are mature, if capacity is overexpanded before market demand is validated, if transformation is forced before organizational capabilities catch up—

then even the right direction can lead to a dead end.

The core question Meta needs to answer has never been 'whether to invest in AI,' but 'how to invest, how much to invest, and when will returns be visible.'

So far, Zuckerberg's answer appears to be: keep investing, keep doubling down, and keep drawing grand visions.

On July 29, 2026, the same day as the earnings release, Meta announced a collaboration with BlackRock to launch a $14 billion data center project in El Paso, Texas.

Earlier, the company also disclosed the 'Hyperion' data center project in Louisiana, with total investments exceeding $50 billion.

The visions keep getting grander, and the money keeps burning faster.

As for the fruits, they are probably still 'three to six months' down the road.'