Cloud Next Google's cloud business last quarter achieved an annual run rate of $36 billion, more than five times what it was five years ago, announced Alphabet CEO Sundar Pichai at the Google Cloud Next 2024 conference in Las Vegas on Tuesday.
While that's about a third of the annual revenue generated by AWS and Microsoft Azure, it's not inconsequential.
"I want to highlight just a few reasons Google Cloud is showing so much progress," said Pichai. "One is our deep investments in AI. We have known for a while that AI will transform every industry and company, including our own."
AI software, with its preference for costly GPU infrastructure, will generate a lot of business for cloud service providers. That's the reason Google and its cloud rivals won't shut up about it. But it's also potentially useful.
"Today, Google AI can scan 100,000 lines of code in two minutes to spot and fix bugs," the search giant declared in its introductory video. Spoiler: There are still bugs in software.
"Today, AI impacts lives for the better and understands the work the way you do," the video voice continued, making no mention of the ways in which AI has made things worse: misinformation, hallucination, usage of resources like energy and water, capturing the intellectual work of others and repurposing for profit without permission, and bias, among other issues.
But that's not what Google execs wanted to talk about. Thomas Kurian, CEO of Google Cloud, came not to bury AI, but to praise it.
"Today, we're going to focus on how Google is helping leading companies transform their operations and become digital and AI leaders, which is the new way to cloud," Kurian proclaimed.
Toward that end, Google has devised many products and services that it is offering through Google Cloud and adjacent business units.
Next-generation generative AI
Kurian said the Chocolate Factory's biggest announcements have to do with generative AI. "Customers have quickly gone from experimenting with generative AI, helping it to answer questions, to make AI predictions, and are now building generative AI agents," he explained. "Agents are intelligent entities that take action to help you achieve specific goals."
As an example, he cited a scenario in which an agent helps an online shopper find a desired dress. That possibility was subsequently demonstrated on-stage when an ecommerce shopping search box was fed with a YouTube video URL and a request to find a shirt like one worn by the keyboard player in the video. Sure enough, the AI bot proved capable of scanning the submitted video, identifying the keyboard player's shirt, and searching the shopping site's inventory for a match.
"Agents process multimodal information simultaneously, conversing, reasoning, learning, and making decisions," said Kurian. "Agents can connect with other agents and with humans, and they will transform how each of you interact with computing devices and the web itself."
It was left to the imagination how much it might cost a business to provide such a service.
- H-1B visa fraud alive and well amid efforts to crack down on abuse
- Google joins the custom CPU crowd with Arm-based Axion chips
- Microsoft hiring Inflection team triggers interest from EU's antitrust chief
- Stability AI reportedly ran out of cash to pay its bills for rented cloudy GPUs
Goldman Sachs CEO David Sullivan appeared in a video testimonial to highlight how the financial firm was finding uses for AI. "We're already seeing signs of promise in a few areas of our experimentation, and we're very optimistic about that," he enthused.
"There's evidence that generative AI tools for assisted coding can boost developer efficiency and productivity by as much as 40 percent," Sullivan continued. "And we're exploring different ways to use AI, whether it's to summarize public filings, extract sentiment and signals from corporate statements, or to gather and interpret information like earnings reports."
That's a way of saying we're not quite sure yet we can fully rely on AI advice.
Gearing up the kit
In terms of products and services, Amin Vahdat, VP for the machine learning, systems, and cloud AI team in Mountain View, talked up Google Cloud's hardware, including the general availability of Cloud TPU v5p, not to mention A3 Mega VMs with Nvidia H100 Tensor Core GPUs.
Vahdat also mentioned HyperDisk ML, a preview block storage service optimized for AI inference and serving workloads. "It accelerates model load times up to 11.9x compared to common alternatives and offers over 100 times greater throughput per volume versus competitors," he said.
In addition, GCP's Cloud Storage FUSE and Parallelstore got a new caching feature that's supposed to accelerate training by storing data closer to a customer’s TPU or GPU.
Vahdat also highlighted open software options like JetStream, an optimized inference engine that offers better performance per dollar for large language models. What's more, Google Cloud plans to offer Nvidia's Grace Blackwell chips in early 2025: HGX B200 and the GB200 NVL72.
Perhaps the biggest news hardware-wise was that Google Cloud now has an ARM-based CPU called Axion. Google claims it offers 50 percent better performance and 60 percent better energy efficiency than comparable x86-based compute instances. Beyond that, there are now N4 and C4 VMs, and bare-metal C3 machines.
Vertex AI, Google's enterprise AI platform, is now offering access to a wider variety of models, including Gemini 1.5 Pro in public preview, Imagen 2.0 family image generation models, and the CodeGemma software assistant.
Checking the facts
Google is also trying to make its models less prone to just making stuff up.
"Because response accuracy is critical for gen AI services, we are expanding our grounding capabilities in Vertex AI, including the ability to directly ground responses with Google Search, now in public preview. Vertex AI users now have access to fresh, high-quality information that significantly improves accuracy of model responses," Vahdat explains in an accompanying statement.
That's right – Google Search, magnet for web spam and affiliate marketing [PDF], can serve as a reality check for hallucination-prone AI models. What a time to be alive.
Google Workspace was not spared the AI augmentation. There's a Google Vids app coming to the suite in June to let people make AI-assisted videos.
"Vids is your video, writing, production, and editing assistant, all in one," explains Aparna Pappu, GM and VP of Google Workspace, in an explainer. "It can generate a storyboard that you can easily edit, and after choosing a style, it pieces together your first draft with suggested scenes from stock videos, images, and background music."
Google Meet now has AI note taking as a preview and in June will get machine learning translation as well. Later this year, Google Chat is scheduled to add AI translation and summarization. This is available through the new AI Meetings and Messaging add-on for $10 per user, per month.
Oh, and if you want the AI Security add-on, which lets IT teams scan and automatically classify and safeguard sensitive files in Google Drive, that'll be another $10 per user per month.
- Google joins the custom CPU crowd with Arm-based Axion chips
- Back of the net? Google's DeepMind is coming for football tactics
- Google ponders making AI search a premium option
- Developers are calling the shots on AI planning, judging by your experience
"We are at an inflection point of sorts where enterprises are evolving from ideating about Gen AI and AI to implementing AI factories of the future," said Garter VP Chirag Dekate told The Register. "Everything enterprises do and everyone that is part of these value creation journeys will be augmented with AI productivity boosts."
Dekate considers the AI Hypercomputer, Google's term for its AI stack, a point of differentiation from rivals. "Here the workload optimized part is important because parts of the workflow will benefit from TPUs and others from GPUs and CPUs," he explained.
"The AI Hypercomputer enables performance optimized access to the workload optimized compute capabilities that are needed to power AI native cloud experiences. Google's differentiation here is a decades-long innovation in purpose designed AI hardware."
Further up the stack, Dekate sees value in Google's use of various home-grown and third-party models, its Vertex AI platform, and its AI Ready Data foundation (Big Query, Looker, and AlloyDB).
"Bringing it all to life is core AI Agent frameworks," said Dekate. "The AI Agents are really important in that they enable enterprises to ground all of the above in enterprise truth and bring AI and Gen AI to life in enterprise data and execution contexts." ®