What Is Meta AI and Its Market Impact

On October 12, 2026, as Meta prepares to launch its latest AI models (Muse Wonder and Muse Wonder Full) the company faces a puzzling contradiction. Despite a surge in revenue, its net income has shrunk significantly. This paradox highlights the high stakes and uncertain payoff of Meta’s aggressive AI push.

Meta Platforms reported net income of $15.85 billion for the second quarter of 2026, down 14% from a year earlier, even as revenue rose 28% to $60.8 billion, according to Variety’s report on the July 29 earnings release. The company exceeded revenue expectations but missed Wall Street’s $7.22 per share forecast, posting $6.18, after a $2.4 billion legal charge and $1.18 billion in severance reduced profits. This sets the stage for the October 12 launch of Muse Wonder and Muse Wonder Full, models Meta says bring it back into competition with OpenAI and Anthropic.

Removing those charges presents a less concerning picture. CFO Susan Li said on the investor call that operating income would have increased 9% year over year excluding the legal and severance charges. As reported, operating income fell 8% to $18.78 billion and the operating margin dropped to 31% from 43%, according to an Insider Monkey analysis carried by Yahoo Finance. Research and development costs rose 67% to $21.7 billion.

The advertising business remains strong. Ad impressions increased 14% and the average price per ad rose 12%, so growth came from extracting more value from existing user attention rather than adding users. Daily active users across its apps reached 3.60 billion in June 2026, up 3% year over year, and advertising revenue was $59.36 billion. The Family of Apps segment earned $23.39 billion in operating income, while Reality Labs lost $4.62 billion.

What Muse Wonder Actually Does

Meta unveiled Muse Wonder and Muse Wonder Full on October 12, 2026, marking a major step in its AI strategy after a year of intense development, according to the company’s announcement. The smaller model is designed for daily tasks, while Wonder Full is the more advanced version. Both are built to handle multi-step activities, such as disputing an insurance claim, reading a lease, or building a working app.

Two features distinguish them from simpler chatbots. Active Memory allows the models to remember details across conversations and recall relevant information. Active Focus helps keep the AI goal-oriented, especially when tasks become complex, by trying different approaches, recovering from errors, and asking for clarification before proceeding. Wired’s review of the launch describes these as solutions to a common AI failure mode: when agents lose track during long research or software development tasks, errors compound.

Availability is currently limited. Developers can access Muse Wonder via the Meta Model API and Muse Code, while Wonder Full is restricted to Muse Code subscribers. The models will power Meta’s new Muse personal agent, initially rolling out to subscribers. Internally, the series was called Watermelon, and in July, Meta AI chief Alexandr Wang told staff that Meta’s models now match or surpass OpenAI, Business Insider reported. Wang noted that Watermelon uses about ten times more compute than its predecessor, Avocado, part of the Muse Spark family.

Where the Benchmark Claims Stand

The most independent metric is the Artificial Analysis Intelligence Index. Muse Wonder Full scored 56, second only to Anthropic’s Claude Opus 5.5, according to CryptoBriefing. This places it among top-tier models based on a composite of multiple tests, and it’s the only benchmark not provided by Meta itself.

Meta’s own claims go further. The company asserts its largest model outperforms OpenAI’s GPT-6 Astra and both Claude Fable 5.1 and Claude Opus 5.5 on specialized legal and software development tests, according to Wired. These evaluations include Harvey’s Legal Agent Benchmark, which tests models on legal document analysis, and Vals Finance Agent V2 for financial tasks, Unite.AI reported. Other benchmarks measure long-term memory with MemoryBench, sustained focus with FocusBench, and long-running coding tasks on SWE-Marathon and DeepSWE.

Wired cautions that benchmarks aren’t always reliable indicators of real-world performance. Meta hasn’t disclosed how its models were trained or built. The legal and coding results are based on Meta’s own tests on selected benchmarks. Until independent labs verify these claims, the most trustworthy statement remains that Wonder Full ranks second on one widely used index, behind Anthropic’s top model.

The Spending Behind the Models

Meta has revised its 2026 capital expenditure guidance upward to $130 billion-$145 billion, from a previous range of $125 billion-$145 billion, compared to $72.2 billion spent in 2025, The Motley Fool reported. The midpoint suggests nearly a 90% increase over last year. The market reacted negatively, as the company’s minimum spending forecast signals it won’t cut back below $130 billion.

In the second quarter, Meta spent $31.08 billion on capital projects, while depreciation and amortization were just $6.36 billion, according to the Insider Monkey analysis. Free cash flow plunged to $784 million from $8.55 billion. Long-term debt stands at $83.66 billion. The company holds $90.26 billion in cash and marketable securities, but that’s only a buffer, not a solution. The stock declined more than 9% the day after earnings.

This spending pattern highlights a key difference with cloud giants like Amazon, Microsoft, and Alphabet. Those companies resell much of the capacity they build, recouping costs by renting infrastructure. Meta, however, primarily invests for its own products and can’t recover costs through reselling, as our analysis notes. Meta’s $31.1 billion in Q2 capital spending nearly matched its $31.9 billion operating cash flow. S&P Global Ratings projects that Meta and its peers will have negative free operating cash flow through 2026 and 2027.

Meta is trying to diversify revenue. CEO Mark Zuckerberg said the company has built an API, is rolling out business agents, and is receiving offers for compute at a premium. Variety. Analyst Nick Jones from BNP Paribas sees strong early interest in Muse, driven by Meta’s large user base, but questions whether this curiosity will turn into steady revenue. As of October 6, Meta’s stock trades at 24.10 times expected earnings (above its five-year average of 22.01) reflecting investor optimism about future returns.

Safety Claims and the Counter-Evidence

Meta states that it has trained Muse Wonder to avoid sharing private information without consent, to refuse harmful requests, and to bolster defenses against cyberattacks, per its announcement. The company claims Wonder has its strongest safeguards yet against chemical, biological, radiological, nuclear, and cyber threats, while still providing defensive tools for security teams. It employs the Meta Superintelligence Scaling Framework and delayed release to enhance safety.

Content safety remains mixed. Meta reports that its AI tools identified over 97% of 33.2 million child-exploitation content pieces before any report. However, Wired uncovered that hundreds of AI-generated child abuse ads, some featuring real children (including a member of a European royal family) slipped through the system. The company has been called to explain how these ads repeatedly appeared. This inconsistency affects trust, especially when AI agents act on behalf of users.

Legal risks are also prominent. CFO Susan Li warned that Meta faces youth-related legal trials in the U.S. this year, which could result in significant losses. The quarterly filing warns that damages sought could total hundreds of billions of dollars. For investors and vendors, this legal exposure adds a layer of risk to Meta’s AI investments. Our decision model guide explains how to weigh vendor self-reports against documented failures when assessing technology choices.

Frequently Asked Questions

What was Meta’s net income in Q2 2026?

Meta reported net income of $15.85 billion for the second quarter of 2026, down 14% from a year earlier, on revenue of $60.8 billion that rose 28%, according to Variety’s report on the July 29 release. Earnings per share were $6.18, below the $7.22 analysts expected.

Why did Meta’s profit fall while revenue rose?

Two charges caused the decline: $2.4 billion related to legal proceedings and $1.18 billion in severance from a layoff of about 8,000 employees. Excluding those, the CFO said operating income would have increased 9%. Research and development spending also rose 67% to $21.7 billion as the company expanded its AI efforts.

How does Muse Wonder compare to OpenAI and Anthropic?

Muse Wonder Full scored 56 on the Artificial Analysis Intelligence Index, second only to Anthropic’s Claude Opus 5.5. Meta claims its largest model outperforms OpenAI’s GPT-6 Astra and Anthropic’s Claude models on legal and coding tests, but those comparisons come from Meta’s own measurements and have not been independently verified.

How much is Meta spending on AI infrastructure?

Meta guided 2026 capital expenditure to $130 billion to $145 billion, compared to $72.2 billion in 2025. In the second quarter it spent $31.08 billion on capital expenditures, and free cash flow dropped to $784 million from $8.55 billion a year earlier.

Is Muse Wonder available to developers?

Muse Wonder is available through the Meta Model API and Muse Code. Muse Wonder Full is limited to Muse Code subscribers. The models will power the Muse personal agent, starting with subscribers, according to Meta’s October 12 announcement.

Related Reading

More in-depth coverage from this blog on closely related topics:

• Deloitte Falsified Abramovich Tax Documents, Investigation Alleges
• What Is Elisity Network Security?
• What Are Fingerprints and Cookies in Privacy
• Skrillex Music and Tour Updates
• Windbreaker Jacket Fall 2026 Trends

Sources and References

Sources cited while researching and writing this article:

• Meta Q2 2026 Earnings: $2.4B Legal Charge, Revenue Up 28% to Over $60B
• Microsoft vs. Meta Platforms: Which Is the Better Magnificent Seven Stock to Buy Right Now?
• Meta’s (META) Sales Jumped 28% While Free Cash Flow Shrank. Can Muse Close the Gap?
• Introducing Muse Wonder and Muse Wonder Full: Frontier Models for Agents
• Muse Wonder Puts Meta Back at the AI Frontier
• Meta says it’s launching the biggest and best AI models it has ever built
• Meta Unveils Muse Wonder Series, Its Most Capable AI Models Yet
• Meta Failed to Catch Hundreds of AI Child Abuse Ads. Some Included Images of Real Kids