Meta Platforms, the social media and technology giant formerly known as Facebook, is undergoing a decisive transformation in its artificial intelligence (AI) strategy following recent missteps with its flagship Llama models and a rapidly evolving competitive landscape. Ahead of the company’s highly anticipated second-quarter earnings call, CEO Mark Zuckerberg is expected to justify Meta’s aggressive investments and high-profile hiring blitz as it seeks to reclaim its influence in global AI innovation.

AI Setbacks Spark Strategic Overhaul

Once lauded for pioneering open-source AI technology, Meta stumbled in early 2025 when its Llama 4 model—developed in a bid to outmaneuver rivals—failed to impress many third-party developers. The new model, designed to compete with recent breakthroughs from Chinese AI startup DeepSeek and industry stalwart OpenAI, was seen as less customizable and harder to integrate than its predecessor, Llama 3. The disappointment came to a head amidst internal controversy and mounting industry scrutiny over Meta’s AI benchmarking practices.

At the root of the setback was Meta’s decision to adopt a mixture-of-experts (MoE) architecture for Llama 4, inspired by DeepSeek’s R1 model. While MoE models, like those OpenAI and Anthropic have increasingly pursued, promise more efficient multi-tasking and computational savings, many developers prefer the ‘dense’ AI models that dominated previous generations due to their customizability and integration ease. The sudden pivot left the developer ecosystem divided, with many clinging to the more flexible Llama 3.

Industry observers say DeepSeek’s R1, released open source in January 2025, “caught Meta off guard” due to its training and efficiency advantages, prompting Meta’s hurried architecture switch. However, the hurried mimicry yielded only modest gains, and some experts argued that Llama 4 failed to leapfrog open-source models from China or the latest offerings from U.S. competitors.

Talent Wars Heat Up: Superintelligence Labs and Strategic Hires

Recognizing the imperative to catch up—and potentially surpass—its rivals, Meta recently launched Meta Superintelligence Labs, a new AI unit led by a dream team of industry luminaries. In June, Meta invested $14.3 billion in Scale AI, onboarding its CEO Alexandr Wang as Meta’s Chief AI Officer. Wang was joined by former GitHub CEO Nat Friedman and Daniel Gross, formerly of Safe Superintelligence, in a move likened to an AI “supergroup.”

Meta’s aggressive recruitment has not stopped there; the company has poached top researchers and engineers from leading AI outfits including OpenAI, Google, and Apple. Notably, ChatGPT co-creator Shengjia Zhao was tapped by Zuckerberg to serve as chief scientist of the newly formed lab, signaling Meta’s intent to play not just catch-up—but to aim for leadership in cutting-edge generative AI research.

While Meta’s projected 2025 expenses ($113–$118 billion) may only rise slightly as a result, Wall Street analysts from Cantor Fitzgerald and Bank of America suggest the larger outlays are a calculated risk, representing a strong vote of confidence in Meta’s strategy to regain AI leadership. Zuckerberg has openly declared plans to invest “hundreds of billions of dollars” in AI infrastructure, vowing that Superintelligence Labs will enjoy industry-leading compute resources per researcher.

Open-Source Strategy in Flux

Historically, Meta has won developer goodwill by open-sourcing breakthroughs like its PyTorch framework and earlier Llama models, contributing to the proliferation of AI innovation worldwide. However, the lukewarm reception to Llama 4 and internal debates have sparked a fundamental re-examination of the company’s open-source strategy. Executives and newly-hired experts are reportedly debating whether to skip future open releases (such as the touted “Behemoth” model) in favor of developing proprietary systems to compete more directly against closed, state-of-the-art models from OpenAI and Anthropic.

A Meta spokesperson stressed the company’s “position on open-source AI is unchanged,” reaffirming Meta will continue to make leading open models available when prudent, but will also balance this with proprietary research as needed. The shift is emblematic of a broader industry pattern, as rivals recalibrate openness and competitive advantage in pursuit of artificial general intelligence (AGI).

AI Investment: From Scepticism to Confidence

The stock market’s reception to Meta’s AI splurge has grown notably warmer compared to Wall Street’s wariness during the metaverse investment wave in 2022–23. Institutional investors now see Meta’s pivot as aligned with a global “AI gold rush” that has tech behemoths, from Alphabet and OpenAI to China’s DeepSeek and Alibaba, investing billions to shape the future of computing platforms and digital agents.

In the second quarter of 2025, Meta’s advertising revenue growth slowed to an estimated 15% year-over-year—its slowest pace since early 2023—according to LSEG data. Still, the company’s core digital ad business remains robust, underwriting the pursuit of high-reward AI breakthroughs. Analysts at Bank of America interpret Zuckerberg’s bullish infrastructure plans as a sign “of confidence in Meta’s revenue trajectory” and point to growing expectations for future capital expenditures tied to AI advances.

The Global AI Arms Race Intensifies

Meta’s urgency mirrors a broader global scramble for AI expertise and innovation. Competition for elite talent now rivals the self-driving car hiring frenzy of the late 2010s, as companies offer lucrative pay, perks, and research freedom to secure top minds. As Megh Gautam of Crunchbase notes, tech giants now view AI research and engineering as a winner-take-all battleground where leadership can define market dominance for decades to come.

OpenAI, Google, Anthropic, and DeepSeek are simultaneously pursuing both closed and open-source advances, contributing to a rapid evolution of model architectures and learning paradigms. OpenAI, once an open-source champion, has dialed back upcoming open projects amid safety and commercial considerations, while Chinese firms like DeepSeek have energized the developer community through large-scale public releases like R1.

Looking Ahead: Meta’s Quest for AI Supremacy

Zuckerberg and Meta’s board face high investor expectations to explain how recent AI investments will translate to competitive advantage, new revenue streams, and a long-term growth story beyond advertising. The tech world will watch closely as Superintelligence Labs and its all-star roster seek to deliver cutting-edge generative AI—and potentially, the foundations of the next great computing platform.

While risks abound—from costly talent wars to uncertain returns and evolving regulatory scrutiny—Meta’s bold moves exemplify the transformative stakes of today’s global AI race.