Google reshapes the AI landscape at Cloud Next 2025
During the Google Cloud Next 2025 event, the company unveiled a series of innovations in artificial intelligence that strengthen its position as one of the leading players in the field. From hardware advancements to improvements in its generative models and the launch of new platforms for developers, Google’s commitment to AI is more ambitious than ever.
The most notable announcement from my point of view as a genai developer was the merging of Project IDX with Firebase Studio (thanks to a fork of Code OSS), resulting in a new development platform focused on intelligent agents, which natively includes the Genkit framework I mentioned in this other article. This combination creates a robust environment for building AI powered applications, directly competing with emerging solutions like Cursor or Replit. Google is not just providing tools, but creating an ecosystem that makes it easier to develop intelligent products from scratch. As for me, I will be testing it and will return soon with more updates.
In the hardware field, Google introduced Ironwood, its next generation artificial intelligence chip. This processor represents a significant evolution compared to previous generations, integrating over a decade of Google Cloud’s experience in AI computing, applied daily to global services like Gmail or Search. Among its most notable features is a performance per watt two times greater than Trillium (TPU v6), thanks to an optimized design and advanced liquid cooling solutions that double the performance compared to air cooling even under heavy loads. Ironwood also features 192 GB of HBM memory per chip (six times more than Trillium), enabling the processing of larger models and datasets with fewer data transfers. In addition, its memory bandwidth reaches 7.2 TBps per chip (4.5 times more), and the bandwidth of the Inter Chip Interconnect (ICI) has been increased to 1.2 Tbps bidirectional, allowing for much more efficient large scale training and inference. Altogether, Ironwood is nearly thirty times more energy efficient than the first cloud TPU launched in 2018.
The company has also introduced major improvements to its generative models. The new Veo 2 includes advanced camera control and video editing capabilities, now available in private preview with an allowlist, enabling enterprise clients to refine and reuse audiovisual content with greater precision. In the audio field, the launch of Lyria stands out, a model capable of generating music from text, also available in private preview on Vertex AI, and allowing for the creation of full production assets from a simple text prompt. Additionally, Chirp 3 incorporates Instant Custom Voice, a feature that allows the creation of personalized voices with just ten seconds of audio. On the other hand, Imagen 3 has improved both in generation and inpainting, enabling the restoration of damaged or missing parts of an image and further enhancing the quality of edits such as object removal.
Another major announcement was Gemini 2.5 Flash, an optimized version of Google’s main model. Designed to deliver faster responses and lower costs, this model allows for dynamically adjusting the level of reasoning, adapting to the specific needs of each task or application.
Although this summary highlights the most important points, it does not fully capture the magnitude of the moment. Google is not only advancing in creating more powerful or efficient models; it is redefining the artificial intelligence ecosystem. With cutting edge chips, multimodal generative models, and a new agent based development environment, Google is setting a new standard for the industry.
Without a doubt, These innovations will not only drive new applications, but also influence how companies and developers think, create, and deploy AI solutions in the coming years. I’ll be back soon once I’ve tried everything out!