The 44-Minute Call That Became Meta’s $130 Billion Problem

(SeaPRwire) –   By: Oliver Hawthorne

Fourteen and a half billion dollars. A 49 percent stake. And then a 44-minute WhatsApp call that reshaped one of the largest AI gambles in tech history. The industry has plenty of stories about failed pivot attempts. OpenAI has one. Microsoft has two. But Meta’s current trajectory is unlike any of them. Not because of the scale alone. It is because the man running it never trained a frontier model. Alexandr Wang, 29, MIT dropout, built Scale AI into the dominant data-labeling platform. He has never personally trained a frontier-class AI model. Zuckerberg hired him anyway. The CFO has reportedly told him, in private, multiple times, that if this fails, the financial damage will be severe. She pulled him aside again and again. Meta is now projecting capital expenditures between $130 billion and $145 billion for 2026. That is nearly double the $72.2 billion it spent in 2025. Investors flinch at every upward revision. Zuckerberg once poured tens of billions into a metaverse that few people wanted to live in. Now he is spending even more on AI. On the Cheeky Pint podcast with Stripe cofounder John Collison, Zuckerberg discussed the scale. The numbers reached $100 billion or $200 billion. He asked Wang directly whether he was sure this was going to work. The market does not yet know whether to treat this as visionary or as a repeat of the metaverse experiment. Both outcomes would define the next chapter of the largest social media company on earth. What is certain is that no other company in the AI race is making its next move with this little confidence. Whether the model will be good enough remains the open question.

The timeline is compressed enough to look like a plot device. In April 2025, Meta’s Llama 4 models landed to a lukewarm reception. Zuckerberg started reaching out for advice. Wang was in a staff meeting when the WhatsApp notification pinged. “Have time for a call?” That call ran about 44 minutes. Two days later, Wang sent a long memo. Zuckerberg described it as Wang laying out what he would do if he were in his position. The first principle was talent density. You do not need many people, Wang argued. You need the right ones. The second was a willingness to start from scratch. Meta was not happy with the Llama 4 trajectory. Zuckerberg was prepared to rebuild the team around someone who could recruit the best researchers away from their labs. That recruiting push lasted months. Zuckerberg called it “supernova.” Meta paid $14.3 billion for a 49 percent stake in Scale AI to bring Wang aboard. Reports of $100 million compensation offers extended to OpenAI researchers followed. Zuckerberg cleared his calendar for two months. He and Wang did not sleep much during that stretch, by Zuckerberg’s own admission. The result was a core research team of fewer than 200 people at Meta Superintelligence Labs. Wang says the number is closer to 100. Wang described the culture he cultivated as one of “insecure overachievers” — researchers who consistently under-promise and over-deliver. Zuckerberg had been quietly impressed by Wang for years. He said Wang felt like the sharpest entrepreneur of his generation. The pick still raised eyebrows. Wang had built Scale AI into a data-labeling powerhouse. But he had never trained a frontier model. Meta has a deep bench of AI PhDs. Zuckerberg was not swayed. He talked to all the lab heads and research leaders. And he said no. Alex is the guy. The team delivered Muse Spark in April. A string of updates followed. Then came Muse Wonder, which Meta calls frontier-class. The Muse app launched in the United States on September 8. It became the most popular free app on Apple’s App Store within two weeks. Over 2.5 million downloads, according to Sensor Tower data. But independent benchmarks tell a more measured story. On Artificial Analysis’s Intelligence Index, Muse Wonder Full lands roughly level with top models from OpenAI and Anthropic. It sits a couple of points behind the current leader. Wang claims state-of-the-art for personal and professional agent tasks based on Meta’s own measurement. That gap between internal measurement and external scorekeeping is where the real risk sits.

The commercial loop is straightforward once you strip away the hype. Meta needs AI monetization to justify a $130 billion capex cycle. The personal AI agent market is the most plausible revenue vector. But it competes directly against ChatGPT and Claude on user retention. Developer stickiness and API economics matter too. Meta has the distribution advantage through Facebook, Instagram, and WhatsApp. It has the capital advantage through an advertising machine that funds these capex projections. But it does not have a proven frontier model yet. That distinction matters when your competitors have already locked in developer relationships. Muse Wonder’s benchmark performance, while competitive, is not the category-leading result that would attract developer lock-in. Wang’s talent-density thesis is credible on paper. A small elite team outperforming a larger one has happened before. But Meta’s execution context is not a research lab. It is a 75,000-person company. Institutional inertia runs deep. The talent pool that Zuckerberg built around Llama was not the talent pool Wang needs now. Wang said Zuckerberg has fully internalized power laws in a way most people have not. Once he commits to a belief, his speed and magnitude of pivoting toward that worldview is unusual. The AI PhDs who lost their research direction in the Llama 4 rewrite may be watching from the sidelines. If the model works, Meta’s distribution advantage amplifies its moat. If the model is merely adequate, competitors with stronger models will win developer mindshare despite weaker distribution. The end-game is binary. If Muse Wonder and its successors reach frontier-class parity on independent benchmarks, Meta becomes a genuine third pillar of the AI race. Its capex bet is vindicated. If the gap persists, the $130 billion capex becomes the most expensive wrong turn in Meta’s history. There is no comfortable middle ground. Wang has the team. He has the memo. He has Zuckerberg’s trust. What he does not yet have is proof that his talent-density playbook survives contact with Meta’s institutional gravity. The market is watching. The clock is not on his side.

Author bio: Oliver Hawthorne, principal correspondent permanently stationed at an international technology review, covering AI infrastructure strategy, capital allocation trends, and frontier model development across major tech hubs.