Mark Zuckerberg has made his position clear: Meta will not be left behind in the race for artificial intelligence supremacy.
What started as a social media empire built on connecting people has transformed into one of the most aggressive AI infrastructure buildouts in tech history, with the company committing up to $600 billion over the next three years and fundamentally reshaping its identity in the process.
The shift has been dramatic and expensive. Meta’s capital expenditures for 2025 alone are projected to reach between $66 billion and $72 billion, a roughly $30 billion increase year-over-year at the midpoint.
CFO Susan Li made it clear during the company’s recent earnings call that this is just the beginning, stating the company expects “to ramp our investments significantly in 2026” as it pursues what Zuckerberg calls “personal superintelligence for everyone.”
This isn’t just throwing money at a buzzword. Meta is constructing massive AI superclusters with names that sound lifted from Greek mythology.
The first, Prometheus, is slated to come online in 2026 in Ohio and will be among the first AI data centers to hit one gigawatt of computing power.
Then there’s Hyperion in Louisiana, which Zuckerberg has described as having a footprint the size of Manhattan, with the potential to scale up to five gigawatts over several years. These aren’t data centers in the traditional sense.
They’re AI factories, purpose-built to train and run the next generation of large language models at unprecedented scale.
But the spending spree extends far beyond concrete and silicon. In June, Meta shelled out $14.3 billion to acquire Scale AI and bring its 28-year-old founder and CEO, Alexandr Wang, on board as Meta’s new chief AI officer.
Wang now heads an elite unit called TBD Lab, which is developing what insiders are calling Avocado, a new AI model that represents a potential departure from Meta’s long-standing commitment to open-source AI.
According to those familiar with the matter, Avocado may launch as a “closed” model that Meta can tightly control and monetise, marking the company’s biggest strategic pivot in its AI approach to date.
The shift hasn’t been without turbulence. The botched release of Llama 4 in April served as a catalyst for a major internal shake-up.
Chris Cox, Meta’s chief product officer and a 20-year company veteran who was the company’s 13th software engineer, no longer oversees the AI division after the failed launch.
The company’s obsession with its Llama brand has notably cooled. While Zuckerberg devoted significant time to discussing Llama during January earnings calls, he mentioned the brand only once during the October call.
Wall Street’s response to the spending has been mixed. After Meta announced it would raise its 2025 capital expenditure guidance, the company’s stock slid 12%, wiping out more than $200 billion in market value in a single day.
Investors are also growing increasingly nervous about the return on investment for these massive expenditures, particularly as Meta has yet to unveil a clear revenue-driving AI product that justifies the billions being poured into infrastructure.
The company is also exploring partnerships with financial firms to co-develop data centers, a sign that even Meta’s considerable cash reserves may need supplementing to sustain this level of investment.
The strategic rationale is becoming clearer, though. Meta sees AI not just as a feature for its family of apps but as the foundation for entirely new revenue streams.
WhatsApp, for instance, is increasingly becoming a primary customer service channel for businesses in India, Brazil, and parts of Europe.
By integrating advanced AI agents into these conversations, Meta can charge businesses for successful conversions, whether that’s booking a flight or selling a product through AI-powered chat. This represents a potentially massive new revenue stream that’s less sensitive to the cyclical nature of display advertising.
The company is also betting heavily on what it calls “agentic AI,” artificial intelligence that doesn’t just answer questions but actually performs tasks.
Meta’s platforms provide what executives describe as the perfect “surface area” for AI agents to operate, whether that’s ordering groceries on WhatsApp, scheduling appointments via Messenger, or editing photos on Instagram.
The pressure is mounting as competitors pour their own billions into the AI arms race. Microsoft plans to allocate $80 billion toward data center development in 2025, while Amazon anticipates spending upwards of $75 billion.
OpenAI recently raised $40 billion at a $300 billion valuation and is in talks for additional funding that could bring it close to a $1 trillion valuation ahead of a potential IPO.
Adding to the complexity, Meta just announced its acquisition of Manus, a Singapore-based AI startup that’s generated significant buzz since its April debut.
With millions of users and more than $100 million in annual recurring revenue from its AI agent platform, Manus represents something Meta desperately needs: an AI product that’s actually making money.
The startup’s Chinese founders and early backing from Chinese investors including Tencent have raised eyebrows in Washington, though Meta insists all Chinese ownership interests will be eliminated following the transaction and operations in China will cease.
For Zuckerberg, who once bet the company’s future and reputation on the metaverse only to watch the stock crater throughout 2022, this AI pivot represents both vindication and enormous risk.
The company has proven it can adapt and recover, staging what some analysts call one of the most significant comebacks in S&P 500 history, with shares up over 350% from their 2022 lows.
But the fundamental question remains unanswered: When will all this spending actually pay off? Meta’s advertising business remains robust, generating the cash flow that funds these AI ambitions. Yet investors want to see tangible returns, and patience is wearing thin.
As one KeyBanc analyst noted, “Meta has been the opposite of Alphabet, where it entered the year as an AI winner and now faces more questions around investment levels and ROI.”
The company’s position is complicated by its ongoing regulatory challenges. The European Union fined Meta nearly $1 billion in late 2024 and 2025 for various compliance issues, and the Digital Markets Act continues to pose threats.
Recently the company took a one-time, non-cash income tax charge of $15.93 billion related to U.S. tax legislation, though it expects the law to reduce federal cash tax payments going forward.
Despite the uncertainty, analyst sentiment remains overwhelmingly bullish. Of 55 analysts covering Meta, 44 rate it a “Strong Buy,” with price targets suggesting potential upside of 29% from current levels. The most aggressive forecasts see the stock climbing 72%.
What’s clear is that Meta is all-in on AI in a way few companies have ever committed to anything. Zuckerberg’s shirt at the September Connect event said it best: “Aut Zuck Aut Nihil” — all or nothing.
The social media giant that connected billions of people is now betting its future on building the infrastructure and intelligence to power the next era of digital interaction. Whether that gamble pays off may determine not just Meta’s future, but the broader trajectory of AI development itself.
Meanwhile, The U.S. Virgin Islands has filed a lawsuit against Meta Platforms, accusing the Facebook and Instagram owner of knowingly profiting from scam advertising while misleading the public, parents, and regulators about the safety of its platforms for both children and adults.
The lawsuit alleges Meta routinely fails to block scam content unless its automated systems reach at least a 95 per cent certainty threshold.
If Meta invested the same resources into automated AI systems to detect and remove scam advertising as it does into promoting platform growth, many of the fraudulent ads targeting users may never reach the public.
For now, the company continues to build. The concrete is being poured, the GPUs are being stacked, and the talent is being recruited at costs that would have seemed absurd just a few years ago.
