Meta's $135B AI Bet: Why Wall Street Cheered +9%

One hundred and thirty-five billion dollars. That is the figure Mark Zuckerberg just committed to artificial intelligence infrastructure in 2026. To put it in perspective, it exceeds Kenya's GDP, surpasses what most G20 nations spend on defense, and represents an 87% increase over what Meta spent in 2025. I won't sugarcoat it: we are looking at the largest corporate infrastructure bet in modern history, and the markets are celebrating like it's Christmas morning.

Meta delivered Q4 2025 results that blew past Wall Street estimates, sending the stock up 9% in after-hours trading and adding $176 billion to its market cap in a single session. But behind the fireworks lie uncomfortable questions: can any company justify $135 billion in annual capex? What happens if the AI bet fails to generate returns? And what does all of this mean for the developers who rely on Llama and Meta's open ecosystem?

My verdict is clear: the Q4 numbers are spectacular, but the check Meta just wrote for 2026 is the biggest and riskiest in its history.

Q4 2025 Results: The Money Machine Keeps Running

Before we talk about the future, let's look at the numbers funding this monumental wager.

If you ask me directly, these results are exactly why Zuckerberg can afford the $135 billion capex gamble. When your core business generates nearly $60 billion in a single quarter, you have the runway to swing big.

Digital advertising remains Meta's indestructible engine. With 3.58 billion daily active people across Facebook, Instagram, WhatsApp, and Messenger, Meta commands the largest audience on the planet. And AI is already supercharging those ads: recommendation models increased session time on Instagram by 8% and on Facebook by 6% in just the last quarter.

But here is a number most people are overlooking: full-year fiscal 2025 revenue hit $200.97 billion. It is the first time Meta has crossed the $200 billion threshold. For a company that was in an existential crisis just three years ago, this turnaround is nothing short of extraordinary.

What Does $135 Billion Buy? Data Centers, Chips, and Talent

The question everyone is asking: where does that staggering amount of money actually go?

Data centers: a global supercomputing backbone

Meta is building what it internally calls "the backbone of AI." This includes:

• New data centers in Louisiana, Montana, and several international locations
• Massive expansion of existing facilities in Iowa, Ohio, and Oregon
• A 2 GW data center that would be the largest in the world upon completion
• Renewable energy contracts exceeding 10 GW to power these facilities

Chips: the silicon arms race

A significant portion of the capex goes directly to purchasing NVIDIA GPUs. Meta was one of the earliest customers for the H100 chip and is now stockpiling orders for B200 and GB200 units to train next-generation models.

But Zuckerberg does not want to depend solely on NVIDIA. Meta is also developing its own in-house chips (MTIA) for inference workloads, aiming to reduce costs over the long term. NVIDIA's recent $20 billion acquisition of Groq underscores just how fiercely contested the AI chip market has become.

Talent: seven- to nine-figure compensation packages

This is where things get truly eye-opening. Meta is recruiting AI researchers with compensation packages ranging from seven to nine figures. Yes, you read that correctly: up to nine figures. We are talking about offers that can exceed $100 million for the most sought-after researchers on the planet.

To contextualize Meta's total capex range of $115–$135 billion, here is how it stacks up against the competition:

The combined capex of the Big Five will exceed $600 billion in 2026. That figure alone should make anyone pause and reflect on the sheer scale of this AI arms race.

Meta Superintelligence Labs: The AI Talent War

In January 2026, Zuckerberg announced the creation of Meta Superintelligence Labs, a new division dedicated exclusively to developing artificial general intelligence (AGI). He placed Alexandr Wang, the founder of Scale AI, at the helm. Wang left his own company to join Meta.

If you ask me directly, this is the clearest signal yet that Zuckerberg is dead serious about AI. You do not recruit the founder of one of the world's most important AI data companies for a side project.

Models in development

Meta Superintelligence Labs is working on several models simultaneously:

• Llama 4 Behemoth: The successor to the Llama family, designed to compete directly with GPT-5 and Claude
• Avocado: A specialized model for programming and code generation
• Mango: A multimodal model focused on image and video

The strategy is clear: Meta wants to be a top-tier player in every AI vertical, from text to code, from image to video. And it is willing to pay whatever it takes to get there.

The talent war has consequences

What rarely gets discussed is the domino effect of these astronomical compensation packages. When Meta offers $100+ million to a single researcher, it forces OpenAI, Anthropic, and Google to match or lose talent. This inflates costs across the entire industry and creates a salary bubble that only the wealthiest companies can sustain.

Zuckerberg publicly admitted that Meta "fell behind on AI" during the metaverse era. Now he is making up for lost time by writing enormous checks. The question is whether money alone can buy the innovation Meta needs, or whether it is simply inflating prices.

The Metaverse Retreat: Reality Labs Keeps Burning Cash

While Meta goes all-in on AI, the elephant in the room remains Reality Labs. The numbers are brutal:

I won't sugarcoat it: $80 billion in cumulative losses is a figure that should alarm any shareholder. And yet, Wall Street looks the other way because the advertising business more than compensates.

In January 2026, Meta laid off between 1,000 and 1,500 Reality Labs employees and shuttered three VR studios. Zuckerberg is redirecting those resources toward AI, but the Reality Labs hemorrhage has not fully stopped. Q4 2025 posted $6.02 billion in losses, one of the division's highest quarterly figures ever.

WhatsApp is not immune to the restructuring either: on January 15, 2026, Meta banned general-purpose AI chatbots from the platform, signaling its intent to control the AI experience within its own ecosystem rather than letting third parties define it.

Impact on Developers: The Llama Uncertainty

For developers building on Meta's open ecosystem, the signals are concerning. Llama adoption among developers dropped from 19% to 11% according to recent surveys.

Why the decline? Several factors are at play:

1. License uncertainty: Llama is "open source" with caveats. The commercial-use restrictions for companies with over 700 million users create doubt
2. Fierce competition: Claude, GPT-4, and Gemini offer more polished APIs and robust commercial support
3. Lack of continuity: Developers do not know whether Meta will maintain its open-source commitment once it has models competing with GPT-5
4. Fragmentation: Multiple Llama versions with different capabilities make long-term planning difficult

My verdict is clear: Meta needs developers just as much as they need Llama. If the open-source community loses confidence, Meta loses one of its biggest competitive advantages over OpenAI and Google. The 650 million Llama downloads mean nothing if the active adoption trend keeps falling.

Meta vs. Microsoft: Why Wall Street Treated Them So Differently

The same week Meta added $176 billion to its market cap, Microsoft lost $357 billion. Both companies are spending astronomical sums on AI. So why did the market celebrate one and punish the other?

The fundamental difference comes down to visible returns. Meta can point to exactly how AI improves its ads, boosts engagement, and reduces costs. Microsoft is still promising that Copilot will transform enterprise productivity, but the numbers have not backed that up yet.

If you ask me directly, Wall Street is saying something simple: "Spend whatever you want on AI, but show me the return." Meta is showing it. Microsoft is not there yet.

Pros and Cons of Meta's Strategy

Pros

• Indestructible ad business: $200 billion in annual revenue funds any bet
• 3.58 billion users: The world's largest audience to deploy AI at scale
• Llama open source: 650+ million downloads creating a de facto ecosystem standard
• Elite talent: Alexandr Wang and nine-figure packages attract the best researchers
• Diversification: Models for code (Avocado), image (Mango), and text (Behemoth)
• Ray-Ban Meta is working: Hardware with real traction, unlike Quest

Cons

• $135 billion is an existential risk: If AI fails to deliver returns, the fallout will be historic
• Reality Labs keeps bleeding: $80 billion in cumulative losses with no profitability in sight
• Llama adoption is declining: From 19% to 11% among developers is a red flag
• Talent bubble: Nine-figure packages inflate costs across the entire industry
• NVIDIA dependency: The bulk of capex goes to GPUs from a single supplier
• Regulatory overhang: The EU and U.S. could impose restrictions on training AI with user data
• WhatsApp locked down: Banning third-party AI chatbots limits the ecosystem

Frequently Asked Questions

Why is Meta investing $135 billion in AI in 2026?

I won't sugarcoat it: Zuckerberg has publicly stated that Meta "fell behind on AI" during the metaverse years and is now scrambling to close the gap. The $135 billion will go toward data centers, NVIDIA chips, proprietary model development (Llama 4 Behemoth, Avocado, Mango), and recruiting talent with packages reaching nine figures. The strategy is to position Meta as the leader in general-purpose AI and apply it to the company's $200 billion advertising business.

How does Meta's capex compare to other Big Tech companies?

Meta leads with $115–$135 billion, followed by Amazon ($100 billion), Microsoft ($80 billion), Google ($75 billion), and Apple ($30 billion). The combined capex of the Big Five will exceed $600 billion in 2026 — an unprecedented sum that reflects the intensity of the AI arms race. If you ask me directly, we have never seen this level of coordinated capital expenditure in the history of the technology industry.

What is Meta Superintelligence Labs?

Meta Superintelligence Labs is Meta's new division dedicated to developing artificial general intelligence (AGI), led by Alexandr Wang, former CEO and founder of Scale AI. The division operates under Zuckerberg's direct oversight and is developing Llama 4 Behemoth (text), Avocado (code), and Mango (image/video). Meta is recruiting researchers with seven- to nine-figure compensation packages to compete with OpenAI, Google DeepMind, and Anthropic.

Why did Meta's stock surge while Microsoft's dropped?

Both reported earnings the same week, but the market treated them in opposite ways: Meta rose 9% (+$176 billion) and Microsoft fell 6% (-$357 billion). The key difference is visible return on AI investment. Meta demonstrates exactly how AI improves its ads and engagement metrics, while Microsoft has yet to prove that Copilot generates significant revenue. My verdict is clear: Wall Street rewards AI spending only when it sees tangible results.

What will happen with Llama and the open-source developer community?

Active Llama adoption among developers dropped from 19% to 11%, raising serious questions about the future of Meta's open ecosystem. Although Llama has accumulated over 650 million downloads, commercial license restrictions, competition from polished commercial APIs (GPT-4, Claude, Gemini), and uncertainty about whether Meta will maintain its open-source model once it competes directly with closed rivals are all causes for concern. Meta needs to rebuild developer trust if it wants Llama to become a de facto industry standard.