Zuckerberg Says Increasing Computing Power Requires Corresponding Workforce Reductions
Reuters reports that Mark Zuckerberg explained to Meta employees that a large-scale increase in AI capital expenditures means a corresponding reduction in workforce: the company's two largest costs are computing infrastructure and personnel, and more budget allocated to the former will compress the latter.
Meta subsequently confirmed a plan to reduce about 10% of its workforce. The company accounted for $1.18 billion in severance costs related to layoffs in its Q2 2026 financial report; as of June 30, the total number of employees was 75,472, including about 8,000 affected employees, most of whom are expected to no longer be counted in the company’s headcount by the end of Q3.
Meta plans to invest at least $130 billion to $145 billion this year in AI chips, servers, data centers, energy, networks, and related infrastructure. The company’s capital expenditures for Q2 were $31.08 billion, with an annual capital expenditure range of $130 billion to $145 billion.
Zuckerberg's logic does not suggest that the amount saved from layoffs is sufficient to directly cover the entire computing bill. Meta's annual AI infrastructure budget far exceeds the labor costs saved from a single round of layoffs; workforce reductions are more about maintaining the overall growth rate of expenses, adjusting organizational structure, and prioritizing limited operational budgets for AI infrastructure and high-end technical talent.
Meta is still increasing its AI-related technical staff. The company’s financial report states that one driver of rising employee compensation is the recruitment of technical talent over the past year, especially in AI; thus, the so-called "workforce reduction" mainly refers to cuts in general functions, some management, and non-core teams, occurring simultaneously with investments in top AI research and infrastructure positions.
According to Reuters, Meta internally studied transforming the company into an AI-native organization through "Project OT," with some scenarios even considering reducing certain teams by up to 60%; after employee backlash, Zuckerberg paused a subsequent round of larger layoffs. The report is based on internal documents and interviews, and Meta stated that company leadership canceled the second round of layoffs before determining the final total.
In market mechanisms, Meta is shifting capital from back-office personnel, recruitment, management, and some engineering functions to GPUs, data centers, electricity, and AI research talent, which will increase demand for chips, cloud, networks, cooling, and energy infrastructure; employees with capabilities in AI system design, model training, inference optimization, and data center operations will benefit. Positions executing repetitive processes, basic operations, and traditional support functions will face higher pressures from automation and organizational streamlining.
Source: Public Information
ABAB AI Insight
Meta's choice represents a shift in the cost structure of tech companies from "labor-intensive internet platforms" to "capital-intensive AI infrastructure operators." In the social networking era, new users primarily consumed server and content moderation resources, but core growth was driven by product, advertising, engineering, and sales teams; in the frontier AI era, model training and inference directly consume GPUs, electricity, land, data centers, and network capacity. By listing computing power and personnel as the largest costs, Zuckerberg indicates that the company internally views GPUs as a production factor competing for budget with labor.
The capital path is not simply about "machines replacing people." Meta's planned capital expenditure of $130 billion to $145 billion far exceeds the $1.18 billion in severance costs, indicating that layoffs cannot directly fund computing investments; the real effect is to reduce the growth rate of fixed operating costs, allowing the company to control profit margin pressures while maintaining advertising business, AI research, and infrastructure development. The company is reducing about 8,000 positions while continuing to pay higher salaries for AI researchers and technical staff, reflecting a restructuring of talent rather than a complete depopulation.
Historically, this is comparable to the rise of cloud computing, where companies shifted some operational personnel and self-built data center budgets to AWS, Azure, and Google Cloud; the difference is that Meta is not outsourcing computing power but building its own super-scale infrastructure. Its data centers, electricity procurement, and chip orders form long-term fixed assets, which will temporarily depress free cash flow, but if personal AI assistants, advertising recommendations, and enterprise AI services generate high-revenue businesses, these assets will become a scale advantage. Meta's free cash flow in Q2 was only $784 million, reflecting cash flow pressures under high capital expenditures.
This represents a concentration of capital. When AI competition depends on tens of gigawatts of power, millions of chips, and long-term data center contracts, companies with advertising cash flow, financing capabilities, and global infrastructure can use capital expenditures to exchange for model capabilities and inference cost advantages. The mechanism is that the greater the investment in computing power, the more likely the unit model service cost will decrease, but the entry barrier also becomes higher; small and medium-sized companies, even with similar algorithms, find it difficult to simultaneously bear the costs of training, online services, and product distribution. Workforce reductions are merely an external manifestation of this capital-intensive process on organizational structure.
ABAB News · Cognitive Laws
- When computing power becomes a means of production, budgets flow from payroll to data centers.
- Layoffs save on expense growth, buying space for capital expenditures.
- The most expensive people in the AI era are not the executors, but those who can manage the infrastructure.