Google’s AI ambitions had been on full show at its Cloud Subsequent convention this week, however for me, two issues stood out: It is now firmly selling AI brokers and is laser-focused on effectivity.
As normal, there was a whole lot of AI hype. “It is a distinctive second,” mentioned Google Cloud CEO Thomas Kurian, with CEO Sundar Pichai including that it may well “allow us to rethink what’s attainable.”
Pichai pointed to how Google helped recreate The Wizard of Oz for The Sphere in Las Vegas, changing it from a standard film format to at least one that works on a 160,000-square-foot display screen. “Even just a few years in the past, such an endeavor would have been almost unattainable with typical CGI,” in accordance with Google, with the staff needing to “account for all of the digicam cuts in a standard movie that take away characters from components of sure scenes, which wouldn’t work on the new, theatrical scale that was envisioned.”
The AI-enhanced Wizard debuts on Aug. 28. However the “likelihood to enhance lives and reimagine issues is why Google has been investing in AI for greater than a decade,” Pichai mentioned this week.
A Surprisingly Quick Ascent for Brokers
AI brokers had been the true wizards at Cloud Subsequent, nonetheless, with Kurian saying a number of new interoperability-focused options for Google’s AgentSpace platform.
Notably, the brand new Agent Growth Equipment inside Vertex AI helps the Mannequin Management Protocol (MCP) that permits brokers to entry and work together with numerous information sources and instruments, quite than requiring customized integrations for plugins. MCP was introduced by Anthropic only a few months in the past, and now evidently all the key AI software program firms are supporting it.  Â
As well as, Google introduced a brand new Agent2Agent protocol that permits brokers to speak with one another whatever the underlying mannequin and framework they had been developed with.Â
Google CEO Sundar Pichai (Picture by Candice Ward/Getty Pictures for Google Cloud)
Google presents its personal purpose-built brokers and instruments for letting you construct your individual brokers, however is now has a multi-cloud platform that “means that you can undertake AI brokers whereas connecting them along with your present IT panorama, together with your databases, your doc shops, enterprise functions, and interoperating with fashions and brokers from different suppliers,” Kurian says.
Salesforce CEO Mark Benioff appeared by way of video to speak about how Salesforce is working with Google to develop and join its brokers.
The thought is that you may use Google’s brokers, create your individual, or combine with third-party brokers. And naturally, Kurian talked about how Google helps you create AI techniques whereas addressing issues about sovereignty, safety, privateness, and compliance.
Among the many brokers Google is producing are Buyer Brokers for name facilities with human-like speech and dialog in Google’s Buyer Engagement Suite; Inventive Brokers for media manufacturing, advertising, promoting, and design groups; Information Brokers for Large Question; and plenty of Safety Brokers. Google is also introducing an Agent Gallery, no-code Agent Designer, Thought Technology agent, and a Deep Analysis agent.
In the meantime, brokers in Google Workspace embody a “Assist Me Analyze” agent for Sheets, Workspace Flows to assist automate duties, and audio overviews, which turns Docs into audio summaries.
It is attention-grabbing to me how shortly the brokers idea has advanced. It was solely a 12 months in the past that firms began speaking about constructing them, versus chatbots, which simply answered questions. To me, it looks as if a whole lot of brokers are chatbots linked to robotic processing automation (RPA) instruments, however that is advantageous if it may well really assist companies be extra environment friendly. Now it looks as if each main AI firm is competing to create platforms that work with brokers throughout software program firms.
Gemini Goes Professional
AI is not low cost; Google invests round $75 billion in capital expense, largely for servers and information facilities, Pichai says. The 2 most attention-grabbing areas listed below are the underlying fashions and the next-generation chips that may energy them.
Get Our Finest Tales!
Your Day by day Dose of Our High Tech Information
By clicking Signal Me Up, you affirm you’re 16+ and conform to our Terms of Use and Privacy Policy.
Thanks for signing up!
Your subscription has been confirmed. Control your inbox!
A couple of weeks in the past, Google introduced Gemini 2.5 Pro, which Pichai describes as “a considering mannequin that may motive by means of its ideas earlier than responding.” Gemini 2.5 Professional is now Google’s high-end mannequin, obtainable by means of its AI Studio, Vertex AI, and Gemini app.
At Google Cloud Subsequent, Pichai announced that Gemini 2.5 Flash, a considering mannequin with low latency and essentially the most cost-efficient efficiency, is coming quickly.
Imagen 3 (Credit score: Google)
As well as, Google introduced enhancements to quite a lot of different AI fashions for particular makes use of. Imagen 3, its image-generating mannequin, now presents higher element, richer lighting, and fewer distracting artifacts, Kurian mentioned. Veo 2, the most recent model of its video-generation software, creates 4K video that’s watermarked, however with options comparable to “inpainting,” or eradicating components of pictures. Chirp 3 creates customized voices with simply 10 seconds of enter. And Lyria transforms textual content prompts into 30-second music clips.
With all these instruments, “Google is the one firm that gives generative media fashions throughout all modalities,” Kurian says.
All these fashions can be found on Google’s Vertex AI platform, which now supports greater than 200 fashions, together with these from Google, third events, and open-source ones. Different adjustments embody Vertex AI Dashboards to assist monitor utilization, throughput, and latency, new coaching and tuning capabilities, and a Vertex AI Mannequin Optimizer.
Really useful by Our Editors
Strike Whereas the Ironwood Is Sizzling
Within the infrastructure space, the most important bulletins was Ironwood, Google’s seventh era Tensor Processing Unit (TPU). Due later this 12 months, this chip is alleged to supply twice the efficiency per watt of the present Trillium chip. Pichai says it has 3,600 instances the efficiency of the primary TPU Google launched in 2013. In that point, Google has turn out to be 29 instances extra energy-efficient.
Amin Vahdat, Google’s VP & GM for Machine Studying, Programs, and Cloud AI, says demand for AI compute has elevated by greater than 10x a 12 months for greater than eight years, by an element of 100 million. Google’s latest TPU Pods help over 9,000 TPUs per pod and 42.5 exaflops of compute efficiency. (The pods will likely be provided in two sizes, one with 256 TPUs and the opposite with 9,216.) Nonetheless, these chips are “only one piece of our general infrastructure,” Vahdat mentioned.
Google Cloud CEO Thomas Kurian (Picture by Candice Ward/Getty Pictures for Google Cloud)
As an alternative, Kurian talked a few constructing an “AI Hypercomputer” that entails a number of applied sciences. As a part of this, Google additionally introduced new compute situations with Nvidia’s GPUs, in addition to a cluster director that lets customers deploy and handle a lot of accelerator chips; some new storage swimming pools, referred to as “hyperdisk exopools” in addition to an “anyplace cache” that retains information near the accelerators, and a zonal storage answer, which presents 5 instances decrease latency for random reads and writes in contrast with the quickest comparable cloud different.
As well as, the corporate introduced new inference capabilities for the Google Kubernetes Engine and Deepmind Pathways for multi-host inferencing with dynamic scaling.
General, Kurian claimed that placing all this stuff collectively signifies that Gemini 2.0 Flash powered by Google’s AI Hypercomputer achieves 24 instances greater intelligence per greenback in comparison with GPT-4o and 5 instances greater than DeepSeek R1.
And, in partnership with Dell and Nvidia, Kurian introduced that Gemini will now run on Google Distributed Cloud for native deployments, together with those who must be “air gapped” for notably delicate functions.
As a part of the infrastructure push, Google introduced that it’s providing its international non-public community to clients. Pichai mentioned the Cloud Wi-fi Entry Community (Cloud WAN) comprises over 2 million miles of fiber and underlies Google’s providers, delivering “over 40% quicker efficiency whereas decreasing whole value of possession by as much as 40%.”
I by no means take vendor efficiency numbers at face worth, and clearly Google’s rivals may have new choices of their very own. However it’s attention-grabbing to see such a give attention to not solely efficiency but additionally value. I do know many CIOs who’ve been unpleasantly shocked by the price of working AI fashions. This can be a step in the suitable path.
About Michael J. Miller
Former Editor in Chief
