Meta is planning to integrate large language models (LLMs) directly into the recommendation systems that power Facebook, Instagram, Threads, and its ads business, according to Chief Executive Officer (CEO) Mark Zuckerberg during the company’s Q4 FY25 earnings call.
Zuckerberg said current recommendation systems are “primitive compared to what will be possible soon.” As a result, he said Meta’s apps will evolve from feeling like “algorithms that recommend content” into systems where “you’ll open our apps and you’ll have an AI that understands you,” with the aim of shaping feeds, ads, and commerce around individual goals rather than static signals.
At the centre of this effort, Zuckerberg outlined what he described as “building personal superintelligence”, AI systems that understand “our personal context, including our history, our interests, our content, and our relationships.”
Meta framed agentic AI as the layer that translates this shift in recommendations into concrete commercial and business outcomes.
Beyond changes to feeds and recommendations, Meta mentioned how agentic AI will be applied across commerce, advertising, and business tools, with a particular focus on transactions and messaging.
As part of this effort, Meta is developing “new agentic shopping tools” that will allow users to surface “the right, very specific set of products from the businesses in our catalogue.” Zuckerberg said these tools are designed to operate across multiple surfaces. “We’re focused on making these experiences work across both our feeds and across business messaging,” he said, adding that this is intended to “significantly increase the capabilities of WhatsApp over time.”
At the same time, Meta positioned agentic AI as a way to deepen its engagement with businesses. During the quarter, the company began testing a Meta AI business assistant for advertisers, which Chief Financial Officer (CFO) Susan Li said can help with “campaign optimisation and account support.”
Meanwhile, Meta pointed to early traction for business AIs on messaging platforms, with Li noting that “over 1 million weekly conversations between people and business AIs” are already taking place in markets such as Mexico and the Philippines. This year, she said, Meta plans to expand these systems so they can “help people get things done right within WhatsApp.”
However, Meta’s ability to deliver on this roadmap is constrained by access to compute. Alongside its product roadmap, Meta repeatedly returned to the issue of compute constraints, describing a widening gap between demand for AI capacity and available supply. During the question-and-answer session, Li said plainly that the company “do[es] continue to be capacity constrained,” despite having ramped up infrastructure through 2025. According to her, “demands for compute resources across the company have increased even faster than our supply,” driven by AI training, recommendation systems, and ads ranking.
Nevertheless, Meta expects additional capacity to come online over the course of 2026. Li said the company anticipates “significantly more capacity this year as we add cloud,” but cautioned that Meta will “likely still be constrained through much of 2026” until new, internally built facilities become operational later in the year.
At the same time, Zuckerberg framed infrastructure investment as a strategic priority rather than a temporary bottleneck. He said Meta will “continue to invest very significantly in infrastructure to train leading models and deliver personal superintelligence,” arguing that efficiency in building and operating compute will become a competitive advantage. As part of this effort, Meta has launched Meta Compute, which Zuckerberg described as a push to be “the most efficient at how we engineer, invest, and partner to build our infrastructure.”
Meanwhile, Li said Meta is diversifying its chip supply and redesigning data-centre development to maintain flexibility. Importantly, she framed AI infrastructure spending as the company’s highest-order capital priority, noting that “the highest-order priority for the company is investing our resources to position ourselves as a leader in AI.” Meta expects capital expenditure in 2026 to be between $115 billion and $135 billion, while still projecting operating income above 2025 levels.
While much of Meta’s AI investment is concentrated in back-end infrastructure, the company also outlined how these systems are intended to surface directly to users.
Meta also used the earnings call to position AI glasses as a key hardware layer for delivering personalised AI experiences, rather than as an experimental hardware category. Zuckerberg said, “Glasses are the ultimate incarnation of this,” referring to Meta’s efforts to build AI systems that can operate continuously and respond to a user’s surroundings. He explained that such devices would be able to “see what you see, hear what you hear, talk to you and help you as you go about your day, and even show you information or generate custom UI right there in your vision.”
While Meta did not disclose absolute shipment volumes, Zuckerberg said “sales of our glasses more than tripled last year.” He linked this growth to the scale of the addressable market, noting that “billions of people wear glasses or contacts for vision correction.” As a result, he argued that widespread adoption is structurally likely.
At the same time, Meta outlined a narrower focus for Reality Labs. Zuckerberg said the company is now “directing most of our investment towards glasses and wearables going forward,” while separately working to make Horizon successful on mobile and to build VR into “a profitable ecosystem over the coming years.” On costs, he said Reality Labs losses in 2026 are expected to be similar to 2025, which he described as “likely to be the peak” before losses begin to decline.
For You
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。