Google launched a new set of cloud infrastructure and tools for AI development, including its latest fifth-generation Tensor Processing Units (TPUs) at its Cloud Next 2023 event on Tuesday.
The tech giant says its new Cloud TPUv5e is its most cost-efficient, versatile, and scalable AI accelerator to date and customers can now use a single Cloud TPU platform to run both large-scale AI training and inferencing. Training is the process of teaching the AI model from data, and inferencing is the process of applying the AI model to new data.
Google has announced the release of new Virtual Machines (VMs) with NVIDIA H100 GPU, which will be available next month. It says these VMs are designed to help organisations achieve three times better training performance than the prior generation A2. The A3 VMs are open, workload-optimised, and offer ML-powered security to deliver zero trust.
Zero trust is a security framework that assumes that every network, user, device, and transaction is potentially compromised and needs to be verified before granting access. It is based on the principle of “never trust, always verify”, which means that no one is trusted by default, even if they are inside the network or have been authenticated before.
The following models are also now available on Google Cloud: Llama 2 and Code Llama, developed by Meta; and Falcon LLM, a widely used open-source model from Technology Innovative Institute. Google Cloud also reveals that Claude 2, a model from Anthropic, will be coming soon. Google Cloud is the only cloud provider that offers both adapter tuning and RLHF methods for Llama 2.
Google also announced some new clients including The Estée Lauder Companies, FOX Sports, GE Appliances, General Motors, HCA Healthcare, and more.
According to Google, their cloud services are used by 70% of the emerging AI unicorns such as AI21, Anthropic, Cohere, Runway and others. Many gen AI startups such as GitHub competitor Replit, Y Combinator-backed Typeface, Jasper, Contextual, CoRover, Elemental Cognition, Fiddler, Quora and more are also Google Cloud customers.
The company claims that more than half of all funded Gen AI startups are using Google Cloud as their platform.
Google also announced new partnerships with several companies to bring its latest-generation AI to various domains and applications.
These include: DocuSign, which is testing new AI features for its users, such as a “smart contract assistant” that can summarise, explain, and answer questions about complex contracts and other documents; SAP, which is creating new solutions that use SAP data and Vertex AI to help enterprises apply AI to important business use cases, such as streamlining automotive manufacturing or improving sustainability; Workday, which has launched its applications for Finance and HR on Google Cloud and is collaborating with Google Cloud to develop new AI capabilities within the flow of Workday, as part of its multicloud strategy.
Workday’s applications include the ability to generate high-quality job descriptions and to bring Google Cloud AI to app developers via the skills API in Workday Extend.
Edited by Saheli Sen Gupta
