At its Cloud Next 2025 event, Google unveiled its latest advancements in AI infrastructure and models, positioning itself for what it terms the “age of inference” – where AI proactively anticipates needs. Headlining the announcements was Ironwood, Google’s 7th-generation Tensor Processing Unit (TPU), promising enhanced performance for demanding AI workloads.
The tech giant also introduced a suite of new AI models designed to revolutionize content creation and communication. Lyria, now in enterprise preview, is a text-to-music model capable of generating high-fidelity, royalty-free music across various genres within minutes. Google highlighted its potential for creating custom soundtracks without licensing complexities.
In audio advancements, Chirp 3, an upgraded model, now offers HD voice synthesis in over 35 languages and the ability to clone voices from a mere 10-second audio sample. Additionally, a new transcription feature can differentiate between multiple speakers in audio recordings, proving valuable for call centers and podcasts. Google also emphasized ethical considerations, stating that its Instant Custom Voice feature includes safety measures and a rigorous verification process for voice usage permissions. DeepMind’s SynthID will now watermark all media generated by Google’s creative models to ensure ethical use.
Furthermore, Google announced Gemini 2.5 Flash, a new “workhorse” model optimized for low-latency and cost-sensitive applications like customer support. This model can dynamically adjust its processing time based on the task’s complexity and allows users to fine-tune the balance between speed, accuracy, and cost. These announcements underscore Google’s focus on developing AI that not only responds but also anticipates and acts proactively in real-world applications.