Meta is reportedly in early discussions with Google Cloud about using its Gemini AI models to improve ad targeting. According to The Information, Meta employees have discussed adapting Gemini and the open-source Gemma models with Meta’s ad data.

The idea would be to strengthen Meta’s ad recommendation system, including tools that summarize ads before showing them to users most likely to engage.

At this stage, the talks are described as early and may not lead to an agreement. The report suggests that Meta’s consideration of a rival company’s technology highlights difficulties in scaling its own AI models, despite years of heavy spending on research and infrastructure.

Meta’s response to the reports

Meta has denied the claim that it is integrating Google’s Gemini into its ad systems. According to PYMNTS, Meta’s spokesperson said the company is not pursuing Gemini for its ad systems but is instead using external models for benchmarking. 

“We’ve always built our own industry-leading, proprietary ad targeting and recommendation systems,” the spokesperson said in an email, PYMNTS reported. “Separate from that, we regularly evaluate third-party tools for the purpose of benchmarking, which is the case here.”

Meta’s plans for full ad automation and AI investments

Meta and Google are two of the biggest companies in online advertising, and both companies have tied recent revenue growth to AI. In Q2 2025, Meta’s ad business grew strongly. Ad revenue was around $47 billion, up about 21% year-over-year. Ad impressions increased by roughly 11 percent, and the average price per ad went up about 9 percent.

The company attributes much of this growth to improvements driven by AI in its ad systems. Meta says its new AI-powered recommendation model has been applied to more “surfaces.” It also reported that a meaningful portion of its ad revenue now comes from campaigns using its generative AI features.

Meta has also been highlighting automation in ad creation and delivery, while encouraging advertisers to adopt Advantage+ campaigns powered by machine learning. The platform has stated that by the end of 2026, it aims to fully automate ad creation, targeting, and budget optimization. Under this plan, a brand might only need to provide a product image and a budget. The AI would then generate the ad (text, video, images) and decide where and to whom it should be shown.

To support this vision, Meta is investing heavily in AI infrastructure and talent. The company has boosted capital expenditures and made investments in companies like Scale AI to support training and infrastructure.

Meta Superintelligence Labs (MSL) is organized with units for research, training, product, and infrastructure. Sources say it is working on newer Llama models, versions 4.1 and 4.2, and aiming for Llama 4.X by the end of 2025.

CEO Mark Zuckerberg has called AI “the backbone” of Meta’s future, describing its vision as building “personal superintelligence” that serves both users and advertisers. Zuckerberg has also said that AI will make Meta’s ad business work “a lot better.”

Despite these efforts, reports suggest Meta may be struggling with the pace of its AI rollout. The Information noted that Meta’s new Superintelligence Labs team is still refining its in-house Llama models. Turning to a competitor’s models, even temporarily, would point to challenges in meeting advertiser demands at scale.

‍