مراكز بيانات الخيام لدى ميتا ليست مجرد ابتكار، بل هي عَرَض لمشكلة ذكاء اصطناعي بقيمة 145 مليار دولار

The flagship deployment is at the Prometheus campus in New Albany, Ohio, a facility expected to draw more than 1 gigawatt of power by 2026 . Beyond speed, the tent structures represent a calculated trade-off: they notably lack backup diesel generation, prioritizing raw capacity and deployment velocity over the traditional redundancy seen in standard data centers . The strategy was reportedly inspired by the unprecedented launch speed of Elon Musk's xAI, signaling Meta's urgency to keep pace .

The Product Gap: A Cascade of AI Model Delays

The rush to build "data tents" is happening in parallel with a series of high-profile stumbles in Meta’s AI research division. The company's frontier models have faced repeated, public delays, falling behind key competitors.

• Avocado (Next-Gen Frontier): Initially targeted for a March 2026 launch, this model was postponed to at least May or June 2026 after internal benchmarks revealed it lagged behind Google's Gemini 3.0 on critical tasks like reasoning and programming . A separate report indicated the project, originally slated for late 2025, had already been delayed into the first quarter of 2026 .
• Llama 4 'Behemoth': This flagship large language model was pushed from a June release to the "fall or later" in 2026, after senior engineers struggled to substantially improve its capabilities beyond what earlier, smaller models in the Llama family could offer .
• Muse Spark API: As recently as June 2026, the Wall Street Journal reported that Meta had delayed the release of its new Muse Spark AI API multiple times since April 2026 due to "technical bugs and infrastructure challenges," with no set release date on the calendar .

These shifting timelines are a major setback for Meta’s monetization strategy. The company is spending like a leader without shipping like one, creating an uncomfortable vacuum between its massive infrastructure buildout and the commercial products meant to justify the price tag .

The Financial Reality: A $145 Billion Bet with an Uncertain Payoff

Meta’s capital expenditure trajectory is the steepest in corporate history. The company's AI spending has entered a phase of doubling down, with the bill rising much faster than revenue from new AI products.

• 2024: Approximately $40 billion in capex .
• 2025: Jumped to $72.2 billion, a roughly 70% year-over-year increase .
• 2026 Guidance: Initially set at a shocking $115–$135 billion, Meta raised it again during Q1 2026 earnings to $125–$145 billion . Meta attributed the increase to higher component prices and additional data center costs .

The sum now nearly doubles 2025's record spend and represents roughly 1.3 times Meta’s annual EBITDA, a staggering ratio that leaves little room for error . To secure the necessary compute, Meta also signed a $27 billion infrastructure partnership with Nebius Group in March 2026 .

Wall Street's Verdict: From Cheers to Jitters

Investor sentiment has oscillated sharply as the scale of the gamble comes into focus. Initially, Wall Street gave Mark Zuckerberg a pass. When Meta announced its initial $115–$135 billion plan in January 2026, the stock surged over 10% as strong advertising revenue helped investors look past the dizzying numbers .

That patience has since frayed. After Meta raised its capex forecast to $145 billion during its Q1 2026 earnings report—despite reporting strong quarterly profits—the stock immediately tumbled 8.5% . The spending surge overshadowed all other results, a clear sign that investors are now heavily discounting future AI revenue. While institutional analysts had flagged "execution risk" earlier in the year, retail investor sentiment on platforms like Reddit had already cratered, dropping from a score of 71 down to 38 out of 100 by February 2026 .

Analysts have distilled the core financial risk into a blunt warning: the $145 billion capital allocation could become "irreversible if delayed models fail to generate ROI" . The tent data centers, then, are more than an engineering marvel. They are a physical manifestation of a high-wire act—a company trying to build its way out of a performance gap, hoping the sheer scale of its infrastructure will compensate for models that are not yet good enough.