Today: As is tradition, AWS released all the news that won't make the re:Invent keynote ahead of time, the Allen Institute for AI introduces a powerful and truly open-source AI model, and the quote of the week.
This era of enterprise software is either the dawn of a new era of corporate productivity or the most hyped money pit since the metaverse. ServiceNow's Amit Zavery talks about the impact of generative AI, how SaaS companies should think about AI models, and his decision to leave Google Cloud.
Today: Google Cloud expands its hardware strategy and (of course) talks up AI, why Intel has nowhere to go but up when it comes to its own AI strategy, and the latest funding rounds in enterprise tech.
Welcome to Runtime! Today: Google Cloud expands its hardware strategy and (of course) talks up AI, why Intel has nowhere to go but up when it comes to its own AI strategy, and the latest funding rounds in enterprise tech.
(Was this email forwarded to you?Sign up here to get Runtime each week.)
Chipping away
LAS VEGAS — Five years ago today, Google Cloud CEO Thomas Kurian hosted the first Google Cloud Next in San Francisco after he took over the top job earlier that year. A lot has changed since 2019, when the main topics at Google's annual customer event were Kubernetes and multicloud management software like Anthos; everyone knew what this year's event would cover a long time ago.
Kurian spent most of Tuesday morning's keynote evangelizing the power of generative AI technologies through new additions to its Vertex AI platform and Gemini models. But Google also joined rivals AWS and Microsoft by announcing plans to let customers run workloads on a custom Arm-based CPU, arguably the freshest thing unveiled at this year's event.
Google said the Axion processor, which is already running several internal Google services and will be available for customer use later this year, is 30% more powerful than other Arm processors available on the cloud and 50% more powerful than comparable x86 chips from Intel and AMD.
It seems likely that Google is comparing Axion to Ampere's Arm CPUs — which are available from Google, Microsoft, and Oracle — rather than AWS's Graviton or Microsoft's Maia, but it didn't specify and cited "Google Cloud internal data'' as the source for those claims.
Google has built its own custom AI silicon for a long time, and said the latest version of its TPU chip for AI workloads is now generally available.
But like its rivals, Google also plans to make sure customers can run Nvidia's latest and greatest AI chips, and said the Blackwell generation will be available "early next year" on Google Cloud.
Despite all the hype, enterprises have taken their time evaluating generative AI technology over the last 18 months. Google introduced two new features Tuesday that Kurian believes will help them make the leap.
Both involve the use of "grounding," which is what popular generative AI techniques like RAG (retrieval augmented generation) are designed to do.
RAG allows companies to connect large-language models to their own corporate data and generate specific answers to prompts that might otherwise end up producing hallucinations.
Vertex AI customers will now be able to use Google's search technology, which it has been pretty good at developing over the years, alongside its Gemini model to search for internal corporate data or documents.
And it will also now be able to connect to external SaaS applications like Salesforce, Workday, ServiceNow, and Atlassian's Jira to find relevant information hosted on those platforms.
So much of the discussion about Google Cloud over the last five years has centered around the horse race; will it ever be able to catch AWS and Microsoft in cloud infrastructure computing? The pecking order has not changed, but Google Cloud is clearly a different company than it was five years ago.
Most cloud customers are interested and willing to experiment with new technologies, but they want to do so at their own pace while making sure they still have the tools to keep their revenue-generating applications running and improving.
Generative AI hasn't changed that fact, and if anything it has put a premium on vendors that can help customers incorporate those emerging technologies in whatever way makes the most sense for their business.
Before Kurian, that was always the knock on Google Cloud; it had a well-deserved reputation for building things The Google Way and just sort of assumed everybody else wanted to do that too.
While keynote demos are probably the least-reliable way to purchase cloud computing services, Tuesday's presentation focused on real-world business problems, not shiny tech.
But Google does show one sign of falling back into its old habits.
It is ramming its Gemini foundation model into basically everything it sells, which probably gives its sales people (excuse me, go-to-market function) something to tout when up against Microsoft and OpenAI.
Google mentioned other models like Anthrophic's Claude and Meta's Llama on Tuesday, but it clearly has its thumb on the scale when it comes to Gemini and that approach could backfire.
We’re excited to announce a new section coming soon from Runtime: Runtime Roundtable, a group of enterprise tech leaders and thinkers who will share their perspectives on this ever-changing slice of the technology world on a regular basis. We're looking for tech executives, founders, investors, and buyers who are willing to help their peers make decisions about how to acquire and deploy enterprise technology, which gets more and more complicated every quarter.
While chips like Google's Axion create new competition for Intel in its main market, it's on the bottom looking up when it comes to the specialized chips needed to train AI models. Intel unveiled its best-yet challenger to Nvidia's dominance of that market on Tuesday, introducing its Gaudi 3 processor.
According to Reuters, Intel said Gaudi 3 is 50% faster at training AI models than Nvidia's H100, the current workhorse of the AI boom. Of course, Nvidia is currently rolling out the H200 to its customers this quarter, which is also when Intel's customers expect to get their hands on Gaudi 3, and Blackwell is coming next year.
Dell CEO Michael Dell all but begged Intel to get the chip out faster during the launch event Tuesday, according to CRN. As much as they sing Nvidia's praises at every opportunity, cloud providers and server manufacturers are desperate to find an alternative lest they find themselves as beholden to Nvidia as they were to Intel a decade ago.
AWS might be having power issues of its own in Ireland, where it has imposed restrictions on spinning up new computing resources thanks to a maxed-out power grid, according to The Register.
Hugging Face changed the license for its TGI tool to the more permissive Apache 2 license, a rare case these days of a license going from kinda open to really open.
Tom Krazit has covered the technology industry for over 20 years, focused on enterprise technology during the rise of cloud computing over the last ten years at Gigaom, Structure and Protocol.
Today: As is tradition, AWS released all the news that won't make the re:Invent keynote ahead of time, the Allen Institute for AI introduces a powerful and truly open-source AI model, and the quote of the week.
Today: Microsoft shores up its AI strategy heading into a pivotal year, Meta is getting into the AI SaaS business with the former leader of Salesforce's AI division, and the latest enterprise funding.
Today: OpenAI would rather ChatGPT users spend more time using its tool than other "copilots," HPE rolls out a new supercomputer design, and the quote of the week.