Meta Bets $115 Billion on AI Dominance as DeepMind CEO Warns of ‘Jagge

When Mark Zuckerberg announced Meta’s AI capital expenditure plans for 2026, the number was so large it barely seemed real. Between $115 billion and $135 billion—more than the GDP of many countries—will flow into artificial intelligence infrastructure this year alone. For context, that’s roughly equivalent to building the entire Apollo moon program, adjusted for inflation, every twelve months.

“If we don’t disrupt ourselves, someone else will. You’re better off doing it on your own terms.” — Demis Hassabis, CEO of Google DeepMind

The Scale of Meta’s Wager

The centerpiece of Meta’s bet is a data center project codenamed Hyperion, currently under construction in Louisiana. When completed in 2027-28, it will be the world’s largest AI facility—a 5-gigawatt behemoth roughly 50 to 100 times the size of a typical data center. To put that in perspective, it could power a small country.

This isn’t speculative planning. Meta has already crossed a milestone that took ChatGPT nearly two years to achieve: one billion monthly active users for its AI assistant. The company reached this figure in just 18 months by embedding AI directly into WhatsApp, Instagram, Facebook, and Messenger—apps that collectively serve over four billion people.

The distribution advantage is staggering. While OpenAI and Anthropic ask users to download new apps or visit websites, Meta made AI feel like a natural extension of texting friends on WhatsApp or scrolling Instagram feeds. Approximately 630 million users interact with Meta AI through WhatsApp alone, representing nearly two-thirds of all AI interactions across Meta’s platforms.

The Nvidia partnership underscores the seriousness of the commitment. In February 2026, Meta expanded its deal with the chipmaker to what analysts estimate at $35-67 billion, securing millions of Blackwell and next-generation Rubin GPUs at a time when competitors face months-long backorders.

“Today’s AI systems are very good at certain things, but very poor at certain things. This dichotomy must be fixed before artificial general intelligence is achieved.” — Demis Hassabis

The Jagged Intelligence Problem

While Meta races to scale, one of the field’s most respected scientists is urging caution. Speaking at the India AI Impact Summit in New Delhi this week, Google DeepMind CEO Demis Hassabis warned that current AI systems suffer from what he calls “jagged intelligence”—a phenomenon where models can solve complex problems like International Mathematical Olympiad challenges while failing at elementary arithmetic.

Hassabis, who shared the 2024 Nobel Prize in Chemistry for his work on AlphaFold, predicts that true artificial general intelligence remains five to eight years away. The path forward, he argues, requires solving fundamental limitations in planning capabilities, continual learning, and what he terms “true creativity”—the ability to not just solve problems but to formulate the right questions and hypotheses.

The risks divide into two categories, Hassabis explained. Societal risks emerge when bad actors misuse AI, requiring international collaboration and shared standards to mitigate. Technical risks arise when systems behave in unexpected and potentially harmful ways—a concern that grows more acute as models become more autonomous.

The scientific promise remains Hassabis’s primary focus. He envisions a “new golden era of discovery” within 10 to 15 years, where AI enables personalized medicine, unlocks new materials for fusion energy, and eventually allows humanity to “travel the stars and explore the galaxy.” At Isomorphic Labs, his drug discovery spinoff, AI-designed cancer drugs are already entering preclinical trials.

Two Visions, One Destination

The divergence between Meta’s approach and DeepMind’s reflects a broader schism in the AI industry. Meta is building toward what it calls “radical abundance”—AI woven into every aspect of daily life, generating an estimated $15-20 billion annually from AI products already. DeepMind is focused on scientific breakthroughs that could take decades to materialize.

Meta’s upcoming models tell the story. Codenamed Avocado and Mango, these flagship systems target release in the first half of 2026. Avocado aims to leapfrog competitors in coding and reasoning, while Mango focuses on image and video generation—capabilities that could unlock $20-30 billion in additional revenue through Instagram Reels and Facebook Stories.

Yet both companies face the same underlying challenge: transforming AI from a fascinating technology into something reliable enough to bet the future on. For Meta, that means converting its billion monthly users into daily active participants. Current figures show only 40 million daily active users—a 4% daily usage rate compared to 84% for Meta’s overall app ecosystem.

The Infrastructure Reality

Behind the headlines about models and users lies a more prosaic concern: power. Meta’s planned 30 data centers through 2028 represent an enormous strain on electrical grids. The Prometheus facility in Ohio and Hyperion in Louisiana together will consume more electricity than many industrialized nations.

This infrastructure arms race has environmental implications that remain underexplored. Communities hosting these facilities face rising power bills, water shortages from cooling systems, and constant noise pollution. AMD and Microsoft are developing more efficient chips in response, but the fundamental tension between AI capabilities and resource consumption continues to intensify.

For enterprise technology leaders watching these developments, the strategic implications are clear. The companies that dominate AI will be those that can marshal not just talent and algorithms, but energy, silicon, and data center real estate at unprecedented scale. The $115 billion question is whether that scale will translate into the kind of reliable, general intelligence that Hassabis believes remains years away—or whether we’re building a jagged foundation on shifting sand.

This article was reported by the ArtificialDaily editorial team. For more information, visit AI Funding Tracker and Observer.