Monday, July 13, 2026 | by Keren Lentschner
The race for artificial intelligence has been anything but smooth sailing for Meta. In early July, during an internal all-hands meeting, Mark Zuckerberg voiced frustration over the company’s slow start despite aggressive strategic shifts decreed in recent months, according to an audio recording obtained by Reuters. “In retrospect, the trajectory of AI agent development (software designed to execute tasks autonomously) over at least the past four months hasn’t really accelerated at the pace we anticipated,” the CEO lamented.
The company’s “AI-First” pivot, which included 8,000 layoffs in late April (10% of its workforce) alongside the redeployment of 7,000 employees to AI-related initiatives across products, services, and agentic systems, was far from seamless and is struggling to “bear fruit,” Meta’s chief executive acknowledged. Furthermore, executive leadership misjudged the timeline for these structural shifts. Having recently established a Chief Data Officer role to accelerate execution, Zuckerberg now estimates that Meta may begin seeing tangible results from the restructuring within three to six months.
Zuckerberg’s candid remarks underscore the staggering stakes the social media giant has placed on artificial intelligence. Locked in a fierce arms race against Google, OpenAI, and Anthropic, Meta has repositioned the emerging technology at the core of its corporate strategy. The company projects capital expenditures of $125 billion to $145 billion in AI this year (covering foundational research, data center construction, and cutting-edge silicon purchases) double its expenditure from last year. Last autumn, Meta completed a record $30 billion bond offering to fund this aggressive spending strategy and is reportedly weighing an equity offering.
Some shareholders are growing increasingly uneasy over the sheer scale of these expenditures and Meta’s ability to yield returns. The company is struggling to persuade investors of its strategy’s merits. While its new AI-enhanced ad tools have driven incremental growth in ad revenue, Meta remains far behind its rivals in establishing itself as a dominant force in consumer AI (via its Meta AI assistant) or enterprise solutions. Meanwhile, sales of its AI-powered smart glasses have yet to achieve mainstream breakthrough status. Over the past 18 months, the stock has experienced volatile trading; valued at $1.7 trillion, Meta has slipped out of the top ten most valuable global market caps, shedding more than 7% of its market value over the past year.
“Investors feel that Meta is currently lacking a clear sense of direction,” says Hanan Ouazan, Head of AI Strategy at consulting firm Artefact. “The company is struggling to bridge its long-term ambition (achieving superintelligence, going all-in on infrastructure, and integrating AI into advertising to drive monetization) with short-term market pressures to yield ROI on massive capex.”
A year ago, Meta made the pursuit of human-level “superintelligence” an absolute priority, establishing a dedicated research lab for the endeavor. Zuckerberg was reportedly rattled by disappointing benchmarks from its Llama 4 language model, which trailed competitors in performance, prompting a sweeping reorganization of its research teams.
Zuckerberg spared no expense last year, offering lucrative signing bonuses and compensation packages worth hundreds of millions of dollars to poach elite Silicon Valley talent. He also spent $14 billion to acquire a 50% stake in Scale AI, securing its founder, Alexandr Wang, to lead Meta’s flagship Superintelligence Lab. The shift, however, triggered high-profile departures, including renowned French computer scientist Yann LeCun and key members of his team, who resigned to launch AMI Labs to explore “world models”, an alternative paradigm in AI research.
Yet, deployments of next-generation AI models have hit delays. Meta was severely set back by its ill-fated acquisition of Manus, a Chinese agentic AI startup acquired for $2 billion last December amid a frothy market. Just four months later, Beijing vetoed the transaction, forcing Meta to abort an integration process that was already underway.
Meanwhile, Muse Spark (the inaugural model built by Meta’s Superintelligence Lab and billed as its most powerful to date) finally debuted in April. Positioned as “intentionally compact and fast, yet equipped with sufficient reasoning capabilities for scientific, mathematical, and healthcare queries,” it was followed on July 7 by Muse Image, a generative image synthesis and editing model integrated into Instagram, WhatsApp, and the Meta AI assistant in the U.S. However, the rollout sparked immediate privacy controversies: unless users explicitly modify their privacy settings, third parties can manipulate public Instagram photos using generative AI without creator consent.
Concurrently, Meta continues to pour capital into compute infrastructure, breaking ground on a massive $10 billion data center complex in Mississippi and a $9 billion facility in Canada. It is also co-developing proprietary AI accelerators in partnership with Broadcom.
These setbacks have frayed internal morale at the Menlo Park giant. Meta is navigating widespread employee discontent stemming from spring layoffs, executive attrition, and controversial workplace monitoring practices. Software deployed in April logged employee keystrokes and mouse movements to generate training data for AI agents. The revelation that employees were inadvertently training autonomous systems designed to replace them sparked intense internal backlash, ultimately forcing Meta to uninstall the software last month.
Against this backdrop, Meta is doubling down on alternative avenues to monetize its models, leveraging its core moat: driving engagement and content sharing across its 3.5 billion social and messaging users. In late June, the tech giant quietly launched Pocket, an application that enables users to generate mini-video games via simple natural-language prompts, eliminating technical barriers to game creation. Users can discover and play these user-generated interactive experiences directly within their feeds.
Pocket builds on technology from Gizmo, an online gaming startup Meta acquired earlier this year with 635,000 historic downloads according to Appfigures data. The application positions itself as a “creative platform to build and share gizmos”, its terminology for interactive software objects.
Pocket is not Meta’s sole consumer experiment. Zuckerberg has reportedly greenlit the development of a predictive wagering app. Codenamed “Arena,” the stealth project draws inspiration from controversial prediction platforms like Polymarket and Kalshi (which are banned in France) that allow users to place speculative bets on real-world news events. According to The New York Times, Meta targets 100 million monthly active users, specifically focusing on the 18–34 demographic. Meta aims to capture a share of the booming prediction market, which generated an estimated $44 billion in transaction volume last year, while using the app to drive higher engagement and cross-platform sharing back to its core services.
“By wrapping AI applications in social loops, Meta is attempting to build behavioral data flywheels to refine ad targeting,” explains Hanan Ouazan. “It’s a direct strategy to monetize its AI investments. Meta remains one of the few entities with unprecedented consumer telemetry across its ecosystem.”
Additionally, the tech conglomerate is exploring enterprise infrastructure as a new revenue stream. According to Bloomberg, Meta is considering entering the lucrative cloud market by renting out surplus GPU compute capacity from its data centers to third parties. It is also contemplating commercializing API access to its proprietary AI models, managing the underlying infrastructure while billing developers on a consumption basis.
Speaking to investors on an earnings call in May, Zuckerberg indicated that a cloud infrastructure offering was “definitely on the table.” He noted, “Almost every week, companies reach out asking us to host AI platforms or purchase compute capacity at a premium over our cost basis.” This marks a strategic shift for the founder, who previously hoarded compute resources to stay ahead in the AI race. “Obviously, if we reach a point where we’ve provisioned excess capacity beyond our internal requirements, monetization via cloud access becomes a compelling option, which gives us added conviction to over-invest in infrastructure,” he added.
Such a pivot would pit Meta directly against hyperscale cloud leaders Amazon Web Services (AWS), Microsoft Azure, and Google Cloud, following a playbook similar to SpaceX, which recently began leasing data center capacity to Anthropic, Google, and Reflection AI. “It’s a de-risking mechanism designed to reassure Wall Street: if Meta doesn’t consume all its compute capacity internally, it can monetize the excess,” notes Ouazan. “However, if Meta attempts to commercialize proprietary paid models, its competitive differentiation against established offerings like ChatGPT, Gemini, or Claude remains questionable.”
Wall Street responded positively to the reporting, with Meta shares jumping over 7% at the opening bell the following day.
