Mark Zuckerberg is making a wager that dwarfs his metaverse investment: Meta will spend between $115 billion and $135 billion on artificial intelligence (AI) infrastructure in 2026 alone, with over $600 billion committed through 2028. This represents a dramatic pivot from the company's costly metaverse bet, which erased roughly $60 billion in value. The question now is whether Zuckerberg's open-source AI strategy can actually compete with the proprietary models of OpenAI, Google, and Anthropic.

Why Is Meta Spending This Much on AI Infrastructure?

The scale of Meta's AI commitment reflects Zuckerberg's belief that artificial general intelligence (AGI), a theoretical AI system capable of performing any intellectual task a human can do, will define the next decade of technology. Meta told investors in January 2026 that the company expects its AI assistant to serve more than one billion people, positioning it as a serious contender in the frontier AI race.

To support this ambition, Meta is building massive computing clusters. The Prometheus cluster, scheduled for 2026, will be the world's first gigawatt-plus computing facility, meaning it will consume enough electricity to power a small city. Hyperion, a follow-up facility, will eventually scale to five gigawatts over several years. These aren't theoretical projects; they're under construction now.

The financial commitment is extraordinary even by Big Tech standards. Meta's AI-related capital expenditures jumped from $72.2 billion in 2025 to a projected $115 billion to $135 billion in 2026. For context, that's roughly equivalent to the annual revenue of many Fortune 500 companies, all dedicated to training and running AI models.

How Does Meta's Open-Source Strategy Differ From Competitors?

Unlike OpenAI and Google, which keep their most powerful AI models proprietary and behind paywalls, Meta has chosen a different path. The company released Llama, an open-source large language model (LLM), a type of AI trained on vast amounts of text to understand and generate human language. In 2025, Meta launched the fourth iteration of Llama, introducing Scout and Maverick models alongside a preview of the much larger Behemoth model. Llama 5 is scheduled for release in 2026.

This open-source approach creates a developer ecosystem that proprietary competitors cannot easily replicate. By making Llama freely available, Meta builds loyalty among thousands of developers and enterprises who integrate the model into their own products. If Llama becomes the industry standard, Meta gains influence over how AI is deployed globally, even if it doesn't directly monetize every use case.

What Are the Key Signals Investors Should Watch?

• Llama 5 Performance: The 2026 release of Llama 5 will be a critical test of whether Meta can match the frontier model capabilities of closed-source competitors like OpenAI's GPT series and Google's Gemini.
• Meta AI Assistant Adoption: Zuckerberg expects Meta AI to become the leading assistant serving more than one billion people by the end of 2025, a claim that will be verified by user engagement metrics and market share data.
• Advertising Revenue Integration: Meta released Muse Spark, a new AI model designed to improve its core online advertising business, suggesting the company is finding ways to convert AI research into revenue.
• Workforce Efficiency: Meta laid off approximately 8,000 employees, or 10 percent of its workforce, in May 2026 to improve business efficiencies while simultaneously scaling AI infrastructure investment, a balancing act that will determine whether the company can sustain profitability.

How Has Zuckerberg's Leadership Style Evolved?

The 41-year-old founder's approach to business has shifted dramatically over the past three years. His early philosophy of "move fast and break things" gave way to strategic discipline following the metaverse collapse. The 2023 "Year of Efficiency" streamlined Meta's operations, and the company has since recovered significantly in stock price. Chief Financial Officer Susan Li has earned praise from Wall Street for transparency around capital expenditures and a focus on return on invested capital.

This disciplined approach extends to the AI bet itself. Rather than making vague promises about AGI, Zuckerberg has set specific, measurable goals: Llama 4 becoming the leading state-of-the-art model, an AI engineer contributing increasing amounts of code to Meta's research and development efforts, and the Meta AI assistant reaching one billion users. These are testable claims that will either validate or undermine his strategy.

What Does This Mean for the Broader Tech Industry?

If Meta successfully converts its advertising dominance into an AI infrastructure and services business, it would fundamentally alter competitive dynamics across cloud computing, enterprise software, and consumer AI, sectors currently led by Microsoft, Google, and Amazon. Meta's 3.2 billion daily active users across its family of apps, including Facebook, Instagram, and WhatsApp, provide an unmatched testing ground for new AI features.

The stakes are high not just for Meta, but for the entire AI industry. A successful open-source competitor to proprietary models could democratize access to frontier AI, lowering barriers for startups and smaller companies. Conversely, if Llama fails to keep pace with OpenAI and Google, Meta will have spent hundreds of billions on a strategy that didn't deliver competitive advantage.

Zuckerberg holds the majority of voting power at Meta through Class B shares, making him the company's controlling force regardless of market sentiment or investor pressure. This means he can pursue the AI strategy without fear of shareholder revolt, a luxury few other tech leaders enjoy. Whether that autonomy becomes an asset or a liability depends entirely on whether the bet pays off.