Metas (NASDAQ:META) data center capital expenditures ('capex') have been both praised by AI enthusiasts and criticized by investors concerned with return on invested capital. In reality, Meta is constructing a market-leading infrastructure moat that underpins its AI models. This strategy creates dependency for training architects and application developers on its computeshould third-party access ever be grantedeven though Metas robust Family of Apps currently negates the need to monetize its computer clusters externally. Instead, Meta AI acts as a strategic barrier around its social media empire by locking in engagement, boosting advertising revenues, and paving the way for innovations such as smart glasses. However, while I remain bullish on Metas AI capex, the stocks current overextended valuation makes it a poor entry point.
We anticipate our full-year 2025 capital expenditures will be in the range of $6065 billion. We expect capex growth in 2025 to be driven by increased investment to support both our generative AI efforts and our core business. The majority of our capex in 2025 will continue to be directed toward our core business. Susan Li, CFO, Meta Q4 2024 Earnings Call
Li emphasized that most of the 2025 capital expenditures will be allocated to Metas Family of Apps, even as AI developments transform the business. Investments in state-of-the-art compute clusters are already driving enhanced user engagement, improved content recommendations, AI-driven ad targeting, streamlined content moderation, and automated infrastructure scaling.
She further noted that servers would represent the largest portion of fiscal year 2025 capex. With Meta AI exceeding 700 million monthly users and its deep-learningdriven tools growing in complexity, the company requires expanded server capacity to support training and inference workloads at scale. Unlike users of external cloud services (e.g., AWS (NASDAQ:AMZN) or Microsoft (NASDAQ:MSFT) Azure), Meta owns and operates its global infrastructure. This self-sufficiency mandates sustained investment in data center expansion and server procurement to preserve its leadership in AI-powered applications.
Mark Zuckerberg also remarked during the call:
2025 will be the year when it becomes possible to build an AI engineering agent that has coding and problem-solving abilities comparable to those of a good mid-level engineer. Meta AI is already used by more people than any other assistant, and once the service reaches that kind of scale, it usually develops a durable long-term advantage. Mark Zuckerberg, CEO, Meta Q4 2024 Earnings Call
In comparison, while Meta AI currently has 700 million monthly active users, OpenAIs ChatGPT reported over 300 million weekly active users as of August 2024. This data lends credence to Zuckerbergs comments. Regardless of usage metrics, Meta AIs rapid adoption is remarkablepartly because its Family of Apps serves as an organic funnel into its AI assistants. Unlike ChatGPT, which facilitates primarily deep, information-based interactions, platforms such as Facebook, WhatsApp, and Instagram offer digital social experiences. As a result, Meta is well positioned to consolidate its status as the market-leading consumer-based AI company.
Despite skepticism regarding the return on investment for the hundreds of billions allocated to its AI infrastructure, my research into AI capex trends and training architecturesespecially following the release of DeepSeek-R1 (Chinas efficient reasoning model)suggests that compute clusters remain the primary competitive moat. Meta is establishing an indisputable source of computational power through its data center investments that can be leveraged in the future. Furthermore, the company is poised to capitalize on lessons from DeepSeek-R1 by adapting its internal training architectures to reduce costs while maintaining market-leading compute clusters that dominate in raw inference and training potential.
Advanced GPUs are challenging to design and procure globally. Thus, Meta holds a privileged position as a Western-based company, where export controls are less likely to restrict the computational capacity of its data centers. While short-term evaluations of Metas AI capex may focus on immediate returns, the long-term computational potential of its data centers represents a far more enduring and impenetrable moat than algorithms or training architecturesassets that, although designed with talent, are limited by compute constraints. Moreover, developing competitive data centers demands exceptional expertise (as evidenced by Nvidia (NASDAQ:NVDA) in semiconductor design and TSMC (NYSE:TSM) in manufacturing), and Meta is fortunate to possess both the requisite financial resources and talent, enabling it to leverage these assets more effectively than most competitors.
Valuation and sentiment
Meta stock is on a long-term uptrend; however, its current valuation appears overextended in the short term. A Relative Strength Index (RSI) above 80a signal that the stock is heavily overboughtsuggests that a short-term pullback or even a correction is likely.
Meta Is Building a Compute Moat Amid Overvaluation
Metas one-year EPS (excluding non-recurring items) grew by 60.8%, rising from $14.98 in FY23 to $24.09 in FY24. For FY25, consensus forecasts an EPS of $25.35 (excluding non-recurring items), indicating only moderate growth. Although the market appears unconcerned at present, this vulnerability implies that the current valuation and price are poor entry points for both long-term fundamental and short-term technical investors.
The companys valuation multiples have been on a long-term downtrenda trend that is likely to continue. Assuming that market bullishness remains strong over the next five years but that the stock is revalued based on more moderate growth prospects, a P/E ratio (excluding non-recurring items) of 27.5 appears reasonable in a bull case. This figure aligns with the companys 10-year median P/E ratio of approximately 28. Additionally, consensus estimates project EPS (excluding non-recurring items) to reach around $45 in five years. Although this forecast may be slightly bearish given Metas potential for higher accretion via its AI moat (through innovations such as AI smart glasses and improved operational efficiencies), my higher estimate of $47.50 reflects this upside. In a bull case, Meta stock could reach approximately $1,300 in five yearsa 12.7% compound annual growth rate ('CAGR') from the current price of $715. Even with robust fundamental growth and sustained goodwill, these valuation constraints suggest that only limited alpha can be expected.
Meta Is Building a Compute Moat Amid Overvaluation
Bear case
In a bear-case scenario, if the five-year outlook is less bullishassuming consensus EPS (excluding non-recurring items) of approximately $45 and a more pronounced decline in the P/E ratio to 25 (reflecting a rational market response to lower growth)Metas stock price by February 2030 would be around $1,125. Compared to the current price of $715, this implies a CAGR of 9.5%, which would likely underperform the S&P 500 (SPY) over the period due to valuation factors amid moderating growth rates.
Additional operational factors could also negatively affect sentiment. For example, an unclear return on investment for the companys data center capex could raise concerns. However, given Metas progress in automated coding and the widespread benefits that its Family of Apps derives from AI, this risk appears limited. Moreover, should DeepSeek and other Chinese firms further dominate training architectures, Metas data center infrastructure might become temporarily redundant as it adjusts to harness its compute power effectively. Such developments are more likely to trigger short-term volatility than lasting sentiment corrections.
Conclusion
In conclusion, my analysis suggests that Meta is likely to only marginally outperform the S&P 500 over the next five yearseven in a bull caseprimarily due to its current overextended valuation following a robust FY24. Metas AI capital expenditures have been judicious investments that have built a formidable and largely impenetrable compute moat, one that only other Big Tech companies could challenge. Even China, despite its ambition to catch up with the West in supercomputer capabilities, faces hurdles due to its lag in semiconductor development and current U.S. export restrictions on advanced semiconductor-related products. Although Meta is operationally strong, alternative investments may offer better alpha potential at this time.
