News

Google Cloud Next 2024 Roundup: Vertex AI Agent Builder, Google Vids, New Silicon, and More

• By John K. Waters
• 04/12/2024

Google's annual Cloud Next conference drew thousands of partners, developers, and customers to Las Vegas this week for three days of sessions, product announcements, and a look at the company's ambitious generative AI roadmap.

Google introduced 10 advanced AI-optimized offerings that cut across infrastructure, Workspace, and Vertex AI platforms.

"We last came together just eight months ago at Next 2023," said Google Cloud CEO Thomas Kurian during his conference opening keynote, "but since then, we have made well over a year’s progress innovating and transforming with our customers and partners.... Last year, the world was just beginning to imagine how generative AI technology could transform businesses, and today, that transformation is well underway."

The biggest news for developers at the event was the addition of two features to Vertex AI, Google's AI development platform for building and using GenAI. Google's new Vertex AI Agent Builder, currently in preview, was designed to allow developers to build and deploy GenAI "experiences" using natural language or open-source frameworks, such as LangChain, on Vertex AI. Also, the Vertex platform will soon include  two new text embedding models, currently in public preview: the English-only text-embedding-preview-0409 and the multilingual text-multilingual-embedding-preview-0409. "Embeddings" are numerical representations of real-world data, such as text, speech, image, or videos. They are how the foundation models powering generative AI understand relationships within data, Google explains on its website. And Vertex AI has new machine learning operations capabilities, including prompt management tools, side-by-side comparisons, certainty scores, and a Rapid Evaluation feature.

The public preview of Gemini 1.5 Pro was announced at the event. The newest member of Google's Gemini family of GenAI models will soon be available on the Vertex AI platform. Among other features, it comes with a first-ever native audio (speech) understanding capability and a new File API  to make it easy to handle files. Two other new—system instructions and JSON mode—were included to give  developers more control over the model’s output.

Google also announced AI-related enhancements to its Workspace collaboration and productivity suite, including enhancements to Gemini for Google Workspaces across Gmail, Meet, and Chat. New AI features for Workspace include a "take notes for me" option in Meet, automatic translation for 69 languages, and data loss prevention controls for sensitive files.

The company also previewed Google Vids, an all-in-one video generator that will sit alongside Docs, Sheets, and Slides. The app was designed for the workplace, and the company says it can make everyone a "great storyteller" using the power of AI. Vids is set for release to Workspace Labs in June.

The showstopper on the hardware side was Google’s first in-house-designed, Arm-based CPU, Axion, which was designed for data centers and will be available to Google Cloud customers later this year. Coming at a close second was the announcement of the general availability of Google's Cloud TPU v5p AI accelerator. Google says it will roll out Axion to its business customers later this year. Kurian emphasized the importance of optimization for every cloud workload, noting Axion's role in improving performance, interoperability, sustainability, and cost.

The company also announced that Nvidia's latest Grace Blackwell platform would soon be available on Google Cloud, alongside a newly developed A3 mega instance that uses Nvidia’s H100 GPUs. In early 2025, Google Cloud will introduce Nvidia's Grace Blackwell platform in two versions, aiming to empower businesses to deploy real-time GenAI on large models. Google also announced two new machine series for its general-purpose virtual machine portfolio and the general availability of Hyperdisk Advanced Storage Pools.

Finally, Google has expanded access to AI models, including Anthropic’s Claude 3 and a new fine-tuned version for code generation, CodeGemma, along with Imagen 2.0 for advanced text-to-image technology.