Meta Admits LLMs Don't Yet Power Core Ad Business
Despite its massive investment in Llama, Meta's CFO admitted that LLMs are not yet a meaningful part of its core ad targeting business. The admission is a reality check on the slow path from research to revenue, even for tech giants, suggesting smaller startups can still move faster in niche markets.
Meta's CFO, Susan Li, made the admission at the Morgan Stanley tech conference in San Francisco, stating, "We are not by and large using LLM architecture to do ranking and recommendations work yet." The primary bottleneck is a lack of sufficient computing infrastructure; data centers currently being built won't be fully operational until 2027 or later. The company's current ad-ranking engine, which drives over $160 billion in annual revenue, relies on traditional engagement signals like likes, shares, and watch time. While not powered by LLMs, this system is augmented by an AI-driven automation suite called Advantage+, which handles campaign processes like audience expansion, bidding, and budget optimization. While core ad *targeting* remains outside the realm of LLMs for now, Meta has aggressively rolled out generative AI tools for ad *creation*. More than 4 million advertisers are now using these features to automatically generate images, video, and ad copy, suggesting a strategy of applying generative AI to creative workflows first. The long-term goal is for LLMs to fundamentally change targeting by reasoning about content in real-time, without needing historical engagement data to learn. This would allow for a more forward-looking and context-aware ad system, but Li characterized it as a "longer-term research effort." Meanwhile, startups are moving to fill the gap in the market. Companies like AdCreative.ai have launched LLMs trained exclusively on ad copy to predict performance. Others, like Agentio, use LLMs to analyze creator content and advertiser data to automate ad sales for YouTubers, demonstrating how smaller players can innovate in niche applications.
