Summary Bullets:

• Google Cloud held its first annual summit at the Tobacco Docks in East London (England) in October 2024.

• Announcements included Google’s expanded data residency to help customers undertake Gemini 1.5 Flash ML processing and data storage entirely in the UK, if they handle sensitive data that cannot leave the country.

Google Cloud held its first annual summit at the Tobacco Docks in East London (England) in October 2024. The event included keynotes, workshops, demos, and the prominent presence of an exciting ecosystem of startups, driven by the launch of the Google Cloud Startup Hub, a new community space for developers and entrepreneurs.

Many enterprises in the UK and Europe are running on Google Cloud Platform (GCP), said Google Cloud’s president of EMEA, Tara Brady, during the summit. This includes around 90% of unicorns (startups valued at more than $1 billion). Brady highlighted Google’s early role in engineering GenAI technologies, which the firm is invoking to position itself as an “AI first” company in the enterprise.

AI Processors

Google emphasized its vertically integrated AI strategy, from silicon chips, cloud, platform, models, and applications. At the pure infrastructure level, updates included the wider availability of Axion, Google’s first ARM-based CPU as part of its proprietary chipset architecture. The aim is to offer customers a wide range of choices at the chipset level, running the gamut from GPUs, and TPUs (tensor processing units), to CPUs. On the GPU front, the company supports Nvidia’s Blackwell chips. On the TPU side, Google’s release of its sixth-generation custom architecture, Trillium, brings with it a 4.7X increase in peak compute performance per chip compared to the previous generation, the company claims. The chip is available to Google Cloud’s customers but isn’t sold directly to them: they can only access Trillium through GCP.

TPUs have a long heritage in AI, harking back to 2015, and are considered one of the few viable alternatives to Nvidia’s processors. When it comes to AI, silicon technology is directly linked to cost effectiveness, and using the right chipset architecture for each specific use case is paramount: the most advanced models in the Gemini family will have very different requirements compared to low-cost inference of a local small language model deployed at the edge. By offering a wide range of alternatives, Google is trying to cater to firms with different AI requirements.

On-Premises Cloud

Google Cloud announced its on-premises offering, Google Distributed Cloud (GDC), for enterprises in highly regulated industries, such as central banks in finance, or national critical infrastructure utility firms. GDC is a portfolio of hardware and software solutions to help customers bring Google Cloud into their data center or any other location if they have special sovereignty, compliance, and/or latency requirements. Organizations can scale from 1R single-node servers in edge deployments all the way up to hundreds of racks in massive data centers. GDC offers interoperability with other platforms and flexibility to support technologies by HPE, Dell Technologies, NetApp, Cisco, Palo Alto Networks, Nvidia, and others.

Agentic AI

Google has 130 models in the Vertex AI platform to help organizations get started with Agentic AI. Vertex AI Agent Builder (first released in April 2024) is Google’s approach to Agentic AI, which has taken the AI world by storm (see “DreamForce 2024: Tableau Einstein Marks Strategic Shift for Salesforce Thanks to Agentforce Capabilities“). Agentic AI represents a step forward from large language models (LLMs). Whereas traditional LLMs focus on content generation, AI agents focus on decision-making. LLMs generate responses based on the data they were trained on and are static; AI agents can dynamically interact with their external environment, choosing the right data points to collect for a specific task. Because it does away with the need to create prompts, Agentic AI helps firms get closer to solving the human in the loop (HITL) conundrum.

Business Intelligence Driving AI

As part of Google Cloud’s strategy to strengthen the link between data and AI, important announcements included the integration of BigQuery with the Gemini models and the availability in Looker Conversational Analytics. The company also launched the enterprise edition of Gemini Code Assist. The tool has a natural language interface and is available in many popular integrated development environments such as Visual Studio Code, JetBrains IDEs, Cloud Workstations, Cloud Shell Editor, and supporting programming languages such as Java, JavaScript, Python, C, C++, Go, PHP, and SQL. Other updates included new synthetic data capabilities, with BigQuery Dataframes.