Meta in Talks with Google to Use Gemini to Improve Ad Targeting (finance.yahoo.com)

Meta has held early talks with Google Cloud about fine-tuning Google’s Gemini and the open‑source Gemma models on Meta’s advertising data to boost content understanding and ad summarization — signals that would feed Meta’s ad recommendation system to improve targeting. Reports say the discussions are exploratory and may not lead to a deal, but the news prompted a roughly 1.5% after‑hours bump in Alphabet shares. Meta has previously been evaluating models from both Google and OpenAI for services across the company, and this would involve adapting large pretrained models to domain‑specific ad signals (creative text, metadata, engagement signals) to produce better ad inputs and recommendations. For the AI/ML community this is noteworthy for several reasons: it underscores how even major tech firms are willing to consider cross‑company model use when internal development lags, highlighting the strategic value of high‑quality pretrained models and fine‑tuning pipelines. Technically, it raises questions about data governance, secure fine‑tuning workflows, and latency/serving integration between cloud providers and a proprietary ad stack. The involvement of an open‑source model (Gemma) alongside Gemini suggests hybrid strategies to balance performance, control, and transparency. If pursued, the move could accelerate advances in ad relevance modeling while intensifying discussions about vendor lock‑in, privacy, and competitive dynamics in AI infrastructure.