The real reason Google gave away Gemma 4

Google released Gemma 4 under the Apache 2.0 license with no usage fees or restrictions, allowing developers to build and sell products using the model. The post explains the technical architecture behind Gemma 4's four model sizes, including per-layer embeddings in the smaller models and mixture-of-experts in the 26B variant (128 experts, only 8 active per token), enabling strong performance at low compute cost. The real motivation behind the release is strategic: by building developer loyalty around an open model, Google creates a pipeline of future Google Cloud customers who will scale their Gemma-based products on Vertex AI. Meta is running the same playbook with Llama. The shift also unlocks AI for regulated industries like healthcare and fintech where data cannot leave on-premises infrastructure.