Since the rise of generative artificial intelligence, Meta has been progressively reorganizing its workforce, investments, and industrial ambitions. Behind the layoffs, the concentration of research capabilities, and the expansion of data centers lies a deeper transformation: that of an advertising giant seeking to become a major force in computing infrastructure and artificial intelligence, without yet knowing precisely what economic model will emerge from this transition.

For nearly two decades, Meta's power rested on a remarkably effective mechanism. Facebook, followed by Instagram and WhatsApp, enabled the company to bring billions of users into a single ecosystem of services. Data generated by their interactions fed increasingly sophisticated recommendation systems, while advertisers financed a digital infrastructure that had become almost indispensable. Algorithms organized attention, advertising converted that attention into revenue, and revenue financed further improvements to the algorithms.

That architecture has not disappeared. Indeed, it remains extraordinarily profitable. But it is no longer sufficient to define the future Mark Zuckerberg intends to build.

Since the emergence of large generative artificial intelligence models, Meta has undertaken a transformation affecting almost every dimension of its organization. Teams are being redistributed, research priorities redefined, infrastructure dramatically expanded, and financial commitments raised to levels that would have seemed disproportionate only a few years ago.

This movement is not simply another exercise in technological diversification. It reveals a change in the company's underlying conception of itself. Meta no longer wants merely to own the platforms through which content and interactions circulate. It seeks to control the systems capable of producing, interpreting, and increasingly organizing those interactions.

The stakes are considerable: a company that built its fortune on distributing information now wants to help construct the intelligence that could determine how information is accessed.

From Permanent Expansion to Organizational Discipline

The current transformation has some of its origins in the crisis Meta experienced in 2022.

Following massive hiring during the pandemic, the slowdown in digital advertising, competition from TikTok, and restrictions on advertising tracking introduced by Apple exposed the vulnerabilities of a business model long accustomed to almost uninterrupted growth. At the same time, investments in the metaverse were consuming substantial resources without generating proportionate revenue.

Zuckerberg's initial response was organizational.

Beginning in November 2022, Meta launched several rounds of layoffs. The company officially designated 2023 its Year of Efficiency, reducing management layers, simplifying processes, and increasing pressure on costs.

This period is often interpreted as a correction of excessive hiring. It also represented a break with the culture of expansion that had accompanied the company's growth.

Meta discovered that a larger organization did not necessarily produce more innovation, and that multiplying teams could slow decisions as much as it increased capabilities.

Organizational discipline therefore preceded the acceleration of artificial intelligence. It subsequently provided the framework for that acceleration: fewer intermediate structures, greater concentration on priority projects, and more direct intervention by senior management in resource allocation.

But this first movement remained essentially defensive. The transformation became genuinely strategic when Meta began treating artificial intelligence not simply as a technology serving its platforms, but as the potential foundation of their next evolution.

Llama: Rebuilding a Technological Advantage

Meta did not discover artificial intelligence with ChatGPT. For years, its recommendation, moderation, translation, and advertising systems had relied on machine learning techniques.

What changed was the nature of the competition.

For much of the digital era, the competitive advantage of major platforms depended primarily on their audiences, data, distribution networks, and ability to optimize interactions. Generative models introduced a new technological layer between users and digital services.

An assistant capable of searching, synthesizing, generating content, and executing tasks could gradually transform the conditions under which users access platforms. Companies controlling these systems could acquire influence comparable to that previously exercised by search engines, operating systems, or social networks.

For Meta, this prospect carries a particular risk: the emergence of new intermediaries between its billions of users and the services it controls.

The development of the Llama models is partly a response to that threat.

By distributing models whose weights are accessible under certain conditions, Meta adopted a strategy different from that of competitors favoring entirely proprietary systems. The objective is not exclusively scientific. It also involves encouraging the adoption of a technological environment in which Meta retains a central position.

This approach can stimulate experimentation, attract developers, and reduce the ecosystem's dependence on a small number of closed-model providers. It guarantees neither technological dominance nor direct profitability.

The paradox is already apparent: Meta invests heavily in technologies whose use it sometimes facilitates for other companies, hoping that their wider diffusion will indirectly strengthen its own position.

This strategy becomes more difficult to sustain as competition shifts toward models that are increasingly expensive to train and operate.

Superintelligence as a New Structure of Power

The year 2025 marked a decisive acceleration.

Meta reorganized its advanced research capabilities and established Meta Superintelligence Labs, a structure intended to concentrate its efforts on the most capable models and next-generation artificial intelligence systems.

The arrival of Alexandr Wang, founder of Scale AI, accompanied this evolution. An investment of approximately $14.3 billion in Scale AI illustrated the company's determination to assemble expertise, data infrastructure, and execution capabilities that could no longer be developed exclusively through organic growth.

The significance of this operation extends beyond recruiting an executive or acquiring technical expertise.

Meta brought fundamental research, model development, and industrial decision-making closer to the center of corporate power. Artificial intelligence became a priority whose governance increasingly rested directly with senior management.

This shift altered the implicit hierarchy of the company's activities. Social networks remained the principal sources of revenue, but advanced research projects began determining a growing share of investment and organizational decisions.

The underlying logic was one of strategic concentration: when a technology is considered capable of redefining an entire industry, resources cease to be distributed primarily according to the current performance of individual divisions. They are redirected toward capabilities deemed essential to the next generation of products.

Such concentration, however, introduces difficulties of its own.

Researchers specializing in frontier models constitute an exceptionally constrained labor market. Recruiting them requires substantial compensation packages, while integrating them into large organizations does not automatically guarantee proportionate scientific progress.

More fundamentally, scientific advances do not necessarily follow the financial calendars of the companies funding them. Models may become more capable without their improvements immediately producing new products, additional revenue, or measurable productivity gains.

Meta is therefore attempting to reconcile two timelines that do not always coincide: that of research, inherently uncertain, and that of a publicly traded corporation subject to recurring performance expectations.

The Automated Company: Ambition Meets Reality

The reorganization acquired another dimension in 2026, as Meta sought to transform its own working methods through artificial intelligence.

The objective was no longer merely to produce assistants for users. It was to turn AI into an internal instrument for software development, coordination, and execution.

An investigation published by Reuters in August 2026 described an internal initiative known as Project OT, centered on a more automated organization, smaller teams, and agents capable of handling an increasing proportion of tasks.

This prospect challenges the traditional relationship between headcount and productive capacity.

In a conventional technology company, expanding products and infrastructure generally requires more engineers, project managers, and support functions. Under the model Meta is exploring, part of that expansion could be absorbed by automated systems.

The organization would no longer necessarily be built around large specialized teams, but around smaller groups whose execution capacity is multiplied by artificial intelligence tools.

The promise is considerable. Its realization remains uncertain.

Reuters reported that the project's most radical scenarios had not been fully implemented and that some internal results revealed a gap between increased code output and genuine productivity improvements. Technical difficulties, incidents, and organizational resistance also accompanied the transformation.

These findings matter because they distinguish automating a task from actually improving a production system.

Generating more code does not necessarily mean developing better software. Reducing the number of people involved does not guarantee better coordination. Replacing certain human operations with autonomous agents may simply transfer costs toward supervision, verification, and correction.

Meta's workforce stood at 66,775 employees on September 30, 2026, representing a 15% decline year over year. This contraction is significant, but it cannot be attributed entirely to AI-driven productivity gains. It reflects a combination of restructuring, redeployment, and cost-rationalization decisions.

The transformation of work therefore remains an industrial experiment as much as an administrative reorganization.

Meta has not yet demonstrated that a technology company can sustainably operate with substantially fewer employees because of intelligent agents. It is testing, on a large scale, the conditions under which that hypothesis might become economically credible.

Fewer Employees, More Capital

The transformation is perhaps most clearly visible in Meta's investment structure.

The company now expects capital expenditures of between $135 billion and $145 billion for 2026, compared with approximately $72.2 billion spent in 2025.

This acceleration profoundly changes its economic profile.

Historically, Meta belonged to a category of digital companies capable of generating enormous revenues from infrastructure that was relatively inexpensive compared with the scale of its audience. Data centers, networks, and computing systems already required substantial investment, of course. But profitability largely came from the ability to distribute digital services globally.

Generative artificial intelligence changes that balance.

Training advanced models requires enormous computing capacity. Operating them involves specialized processors, high-speed networks, cooling systems, and reliable electricity supplies. As applications become more complex and their use expands, spending no longer ends with development: it continues throughout the daily operation of services.

Meta is becoming increasingly dependent on physical assets, industrial supply chains, and energy constraints.

This evolution brings certain characteristics of its business model closer to those of major infrastructure operators.

The company must secure equipment, plan capacity years in advance, and commit considerable capital before knowing precisely what future demand will look like.

The nature of risk changes.

In social networking, an unpopular feature can be modified or abandoned at relatively limited cost. A data center built for several billion dollars represents a far less reversible commitment.

Technological competition also becomes a competition between balance sheets: employing the best researchers is no longer sufficient if a company lacks the financial capacity to sustain their work.

Advertising Finances an Economy That Does Not Yet Exist

Meta's third-quarter 2026 results reveal the full ambiguity of this transformation.

The company generated $64.6 billion in revenue, an increase of 26% year over year. Its platforms continued to reach an immense audience, with an average of 3.64 billion people using its family of applications daily in September.

Advertising remained central to this performance.

Artificial intelligence was already improving recommendations, advertising relevance, and monetization. Meta emphasized that these investments were producing additional advertising revenue, distinguishing operational applications of AI from more speculative research initiatives.

But the financial results also revealed the scale of the transition's cost.

Capital expenditures reached $34.91 billion during the third quarter, while free cash flow turned negative at $2.12 billion.

Reported operating income declined by 37% to $12.99 billion. This decrease, however, requires careful interpretation: it included an exceptional $9.89 billion charge associated with a legal settlement concerning the effects of social media on adolescent mental health. Excluding that charge, operating income would have reached approximately $22.88 billion.

It would therefore be misleading to attribute the decline in reported operating margin solely to artificial intelligence investment.

The fundamental tension lies elsewhere: established businesses continue generating growing revenues, but a substantial proportion of financial resources is now being directed toward capabilities whose economic returns remain difficult to measure.

Meta is using the profits of a relatively mature advertising economy to finance an artificial intelligence economy whose revenue models have not yet stabilized.

This strategy could prove highly effective if the new systems strengthen existing platforms and create additional markets. It could also become costly if investments consistently grow faster than the incremental revenues they generate.

The distinction between improving the core business and creating a new business therefore becomes essential.

Advertising AI already has an identifiable monetization logic. Superintelligence, personal agents, and new conversational interfaces still depend, to a considerable extent, on strategic expectations.

The Metaverse Loses Its Centrality Without Disappearing

This shift also requires Meta to reconsider the ambition that justified its name change in 2021.

The metaverse was supposed to constitute a new generation of digital environments in which social interactions, work, entertainment, and transactions would gradually take place within immersive spaces.

That vision has not disappeared, but it no longer occupies the same strategic position.

Reality Labs continues developing augmented reality, virtual reality, and connected devices. However, Meta now expects to reduce the division's losses and redirect part of its investment toward artificial intelligence priorities.

In October 2026, the company anticipated Reality Labs losses of between $18 billion and $19 billion for the year, followed by approximately $15 billion in 2027.

The change does not necessarily represent an abandonment of immersive interfaces. Rather, it reflects a modification in the industrial sequence.

Instead of first building an entirely new digital environment in which users might operate, Meta is placing greater emphasis on developing intelligent systems that can be integrated into existing interfaces and future devices.

Connected glasses acquire particular importance in this context. They could become an interface for assistants capable of understanding speech, images, and the user's immediate surroundings.

Hardware would then cease to be merely an instrument of immersion. It would become a means of accessing artificial intelligence capabilities continuously.

There is continuity between the two strategies, but their center of gravity has shifted.

Meta is no longer seeking only to construct the next digital space. It is seeking to control the intelligence that could accompany users across multiple spaces.

The New Financial Architecture of Power

The industrial transformation is accompanied by changes in financial instruments.

In October 2026, Meta announced the creation of a new class of non-voting Class C common shares, scheduled to become effective in January 2027.

The arrangement is intended, among other purposes, to provide greater flexibility for financing, acquisitions, and equity-based compensation while preserving the group's existing control structure.

The decision is significant for a company that must simultaneously finance infrastructure, attract exceptionally sought-after researchers, and retain the ability to undertake strategic acquisitions.

It also highlights the distinctive nature of Meta's corporate governance.

Mark Zuckerberg retains decisive influence over the company's long-term direction. This concentration of power can help sustain major investments despite immediate uncertainty. It also limits the ability of ordinary shareholders to redirect certain strategic decisions.

The model therefore rests on a particular trade-off between continuity of control and financial discipline.

That trade-off becomes more consequential as capital requirements increase.

On October 30, 2026, S&P Global Ratings revised the outlook on Meta's credit rating from stable to negative, citing uncertainties surrounding artificial intelligence spending and future cash generation.

The agency expressed concern that investment could persistently exceed the company's internally generated financing capacity. Meta disputed that assessment, emphasizing the strength of its existing businesses and the economic prospects of its projects.

This disagreement does not establish that the strategy is unsustainable. It does, however, demonstrate that Meta's transformation is beginning to alter perceptions of its financial risk.

The company retains considerable resources, but it is no longer entirely insulated from the constraints associated with large, capital-intensive industrial strategies.

Financing artificial intelligence is becoming a governance issue as much as a technological one.

A New Dependence on Infrastructure

One of the least visible aspects of the transformation concerns the geography and material foundations of digital power.

For years, the dominance of major platforms appeared to rest primarily on software, data, and network effects. Those factors remain essential, but the artificial intelligence race is strengthening the role of much more tangible elements.

Semiconductor manufacturing capacity, electricity availability, transmission infrastructure, cooling systems, and data center construction timelines are becoming strategic variables.

Meta can design its own models and organize its teams with considerable autonomy. Nevertheless, it remains dependent on equipment suppliers, energy producers, network operators, and territories capable of accommodating its facilities.

This dependence introduces a new dimension into its international strategy.

The availability of industrial and energy resources can influence investment locations, while trade restrictions, geopolitical tensions, and national regulations can alter the conditions under which infrastructure is developed.

Artificial intelligence is giving digital companies a larger industrial footprint while exposing them more directly to the constraints of the physical world.

For Meta, the challenge is no longer simply to capture users' attention. It is also to guarantee the material capacity to operate the systems that may organize their future interactions.

The Battle for the Interface

Behind the investments and restructuring lies a more fundamental strategic question: who will control the relationship between individuals and digital services in the age of intelligent assistants?

Social platforms prospered by organizing spaces in which users create, discover, and share content. Artificial intelligence assistants introduce a different logic.

Instead of navigating multiple applications to search for information, compare products, or complete a task, users could delegate some of these operations to conversational systems or autonomous agents.

This evolution does not necessarily imply the disappearance of social networks. It could, however, change how users access them and the economic value of the interactions they generate.

For Meta, the risk would be to retain an immense audience while losing part of its control over the interfaces through which that audience interacts with the digital world.

The integration of Meta AI into its applications, the development of proprietary models, and investments in connected devices can be understood as complementary responses to that threat.

The company already possesses a considerable advantage: global distribution that few competitors can replicate.

But distribution does not guarantee that users will prefer its assistant, any more than it guarantees the superiority of the models powering it.

Meta must therefore manage a delicate transition: using its platforms to distribute new services without assuming that their historical audiences will be sufficient to establish those services.

The company is seeking to preserve control over the user relationship at precisely the moment when the nature of that relationship could be fundamentally transformed.

The Bet on a Different Company

Meta's reorganization cannot be reduced either to workforce reductions or to a spectacular increase in technological spending.

It represents a simultaneous transformation of several balances that have historically defined the company.

Capital is becoming increasingly important relative to headcount. Advanced research is moving closer to the center of decision-making. Physical infrastructure is becoming as strategically important as software. Established businesses continue financing expansion, but they no longer determine investment priorities on their own.

At the same time, Meta is attempting to transform its own production methods through automation while building products that could fundamentally alter digital behavior.

These changes do not necessarily advance at the same pace.

Advertising already benefits from artificial intelligence systems. Autonomous agents have not yet demonstrated all the expected productivity gains. Advanced models require increasing amounts of capital, while their commercial applications remain partly uncertain. Data centers demand long-term commitments in a technological environment where architectures can evolve rapidly.

The strategic coherence of the entire undertaking therefore rests on a hypothesis: artificial intelligence capabilities will become sufficiently important within the digital economy to justify the resources being committed today.

That hypothesis is plausible, but its economic scale, timing, and future distribution of value remain uncertain.

Meta possesses considerable advantages in managing this uncertainty: a global audience, massive advertising revenues, extensive technological infrastructure, and a governance structure capable of sustaining long-term decisions.

Those advantages do not eliminate the risk of misallocating capital or the possibility that competitors could rapidly shift the technological frontier.

Meta's recent history also demonstrates that major industrial expectations do not always unfold according to their original scenarios. The metaverse was supposed to constitute the company's next horizon. Artificial intelligence shifted its center of gravity before that earlier ambition had reached economic maturity.

It would therefore be premature to regard the current reorganization as the culmination of an already validated strategy.

It is better understood as a bet of exceptional magnitude, financed by one of the most powerful advertising machines ever constructed.

For two decades, Meta learned to convert human attention into revenue. It is now attempting to convert that revenue into artificial intelligence capabilities, hoping they will allow it to retain a central position in the next organization of the digital world.

The question is no longer simply how many users its platforms can attract or how many advertisements they can display.

It is whether the company that dominated the attention economy can also become one of the foundational architectures of the intelligence economy.

And if that transformation succeeds, Meta may discover that its most valuable asset is no longer the algorithm deciding what users see, but the system helping them decide what to do.

Principal Sources

  • Meta Platforms, Third Quarter 2026 Results, October 21, 2026. Financial results, workforce, capital expenditures, Reality Labs outlook, and announcement of the new share class.
  • Meta Platforms, Form 10-Q, quarter ended September 30, 2026, filed with the Securities and Exchange Commission. Financial data, cash flows, and risk disclosures.
  • Reuters, Mark Zuckerberg had a bold plan to replace Meta staff with AI. Here's how it imploded, August 26, 2026. Investigation into Project OT, workforce reorganization, and the limitations encountered in automation.
  • Reuters, S&P Global revises Meta's outlook to negative on AI spending, cash flow worries, October 30, 2026. Analysis of changing credit risk and Meta's response.
  • Meta Platforms, institutional communications concerning Llama, Meta Superintelligence Labs, and its artificial intelligence strategy, 2023–2026.

Reported financial figures and announced corporate decisions are distinguished from strategic interpretations. Investment projections, profitability expectations, and future productivity gains should not be regarded as established outcomes.