Google Cloud is upping the ante on generative artificial intelligence (AI) with a slew of services and tools to help organisations harness the rapidly emerging technology that has taken the world by storm.

Speaking at the Google Cloud Next conference in San Francisco this week, Google CEO Sundar Pichai and his executives outlined the company’s efforts to make AI more accessible to businesses and workers.

Noting that AI will touch every sector and business function, significantly changing the way people live and work, Pichai said Google has been preparing for that pivotal moment for some time now.

“For the last seven years, we’ve taken an AI-first approach, and applied AI to make our products radically more helpful,” he said. “We believe that making AI comfortable for everyone is the most important way we’ll deliver on our mission in the next decade.

“That’s why we’ve invested in the very best tooling, foundation models and infrastructure across both CPUs [central processing units] and GPUs [graphics processing units]. These underlying technologies are helping us transform our products and businesses, and they will help you transform yours.”

That Pichai kicked off this year’s Google Cloud Next with a focus on all things generative AI was notable, and comes at a time of heightened interest in the technology.

Google Cloud’s generative AI play runs the gamut, from building its capabilities into Google Workspace and optimising its infrastructure to support AI workloads, to helping developers build generative AI applications through its Vertex AI platform.

Large language models
At the event, Google Cloud announced several improvements to Vertex AI, including the addition of large language models (LLMs) such as Meta’s Llama 2 and Code Llama along with the Technology Innovation Institute’s Falcon LLM.

The company is also enhancing its own first-party foundation models such as PaLM, which now come with higher quality outputs, a 32,000-token context window that makes analysis of much larger documents simple, and grounding capabilities for enterprise data.

It also claimed that its Codey model for code generation and chat now offers 25% improvements in quality for major programming languages, while Imagen, its model for image generation, features style capabilities so that organisations can create images that comply with specific brand guidelines.

“With these additions, Google Cloud now provides the widest variety of models to choose from, whether it’s first-party models from Google, third-party models from our partners, or open-source models from a variety of vendors,” said June Yang, vice-president of cloud AI and industry solutions at Google Cloud, during a media briefing ahead of the event.

Yang said organisations can also tune the models with their enterprise data, as well as leverage Vertex AI extension tools to connect models to real-time data and drive real-world actions.

“Developers can now build extensions to popular enterprise APIs [application programming interfaces] or build their own extensions to private or public APIs using a schema that’s compatible across Google,” she said. “For example, if an enterprise wants to create an HR chatbot, they can add an extension to access an HR database to answer questions about vacation balance, making their foundation model much more useful.”

With the size of LLMs growing by 10 times per year on average, purpose-built hardware, along with an integrated software stack, are needed to support new computational demands.

Earlier in May 2023, Google Cloud unveiled its A3 supercomputer based on Nvidia H100 GPUs. The A3 will be generally available next month, enabling organisations to train, tune and serve demanding generative AI workloads and LLMs.

Mark Lowmeyer, Google Cloud’s vice-president and general manager for compute and machine learning infrastructure, said the A3 supercomputer leverages networking technologies, such as Google’s infrastructure processing unit, which offloads the host to provide the massive scale and performance needed by generative AI models.

“We’ve also been focused on making our own homegrown Tensor processing units or cloud TPUs accessible to a much broader range of customers and use cases than ever before,” he added, claiming the new Cloud TPU v5e offers two times better performance per dollar for training and 2.5 times better performance per dollar for inferencing.

Google Cloud is seeing growing momentum around the adoption of generative AI, with the number of generative AI customer accounts growing by over 15 times in the past quarter, said Yang. The number of generative AI projects that run on Google Cloud has also grown by 150 times.

According to a study by Enterprise Strategy Group, 42% of organisations are already using generative AI for a variety of business and IT use cases, and an additional 43% are currently in the planning or consideration phase.

Among survey respondents, expectations were high for targeted use cases, with organisations believing generative AI will most benefit customer service (48%), marketing (45%), software development (43%), IT operations (38%) and product development (37%).

Crucial initiatives required for successful digital transformation will also benefit from generative AI, with more than half of organisations expecting the technology to improve process and workflow automation, data analytics and business intelligence initiatives, and employee productivity.