With so many events and big news this week, I wish I could have cloned myself, but instead I shuttled between events in San Francisco and Silicon Valley and tried to keep up with all the news on my phone.

The headline stories: Arm finally went public and its stock rocketed, auguring well for more IPOs coming next week. Dreamforce focused a lot on the issue of trust in the generative AI era. And gen AI looks to blow up how computing gets done all the way from the chip to applications to how data gets wrangled. Oh, and Google’s big antitrust trial began, the first days of what will be a long slog in coming weeks (and months).

You can get more analysis of this and other news from the new theCUBE Pod podcast from SiliconANGLE analysts John Furrier and Dave Vellante. And don’t miss Vellante’s weekly Breaking Analysis Saturday for a deep dive on CrowdStrike ahead of its Fal.con event next week.

Here come the IPOs

In this year’s biggest IPO, Arm stock closes up 25% as chip designer raises $4.87B

• As the story says, Arm Chief Executive Rene Haas is positioning the company as a key driver of the fast-emerging era of artificial intelligence, pushed into high gear by generative AI chatbots such as OpenAI LP’s ChatGPT. “AI on Arm is literally everywhere,” Haas said in a post the morning of the IPO. “Seventy percent of the world’s population relies on Arm technology today, putting us in a unique position to advance AI across all devices.” Perhaps, but it’s no Nvidia just yet, at least from a perception point of view.

• Then again, as Vellante rightly points out in theCUBE Pod this week, it doesn’t need to be, because Nvidia’s GPUs are based on Arm. So are Amazon Web Services’ AI chips, and many others. So it’s already in AI in a very big way, and likely to increase that as AI moves more and more to the network edge.

• Indeed, in an interview with SiliconANGLE, Dipti Vachani, senior vice president and general manager of Arm’s automotive line of business, said the IPO will enable Arm to take an even greater role in burgeoning AI workloads, especially as the execution of AI models moves increasingly to edge devices rather than in data centers and the cloud.

• “We were AI before AI was cool,” Vachani said. “Arm and AI is synonymous. AI workloads have run on Arm from the start.”

• Vachani noted that despite Nvidia’s Corp.’s leadership in AI thanks to its powerful graphics processing units, AI also requires central processing units like Arm’s designs, such as preprocessing data. “As that grows, Arm grows,” she said. “Accelerating AI within the CPU is necessary because of the low-power requirements that CPUs can provide.” Moreover, “it’s built on one architecture that goes from the lowest end to the highest end.”

• The IPO is a talent magnet as much as anything, Vachani added, providing liquidity for employees and potential stock options for prospects.

And another couple of IPOs are on the way: Instacart seeking to raise up to $616M in IPO at $9.3B valuation And then there’s Klayvio.

AI moves to the edge and enterprise suppliers scramble to keep up

As generative AI takes off, promising to bring the power of AI to everyone, it’s becoming clear that every company from chip startups to software giants to cloud providers is having to change how it does business to keep up with the computing demands of the revolutionary technology.

That trend was on full display at the AI Hardware & Edge Summit, a small but information-rich conference in Santa Clara this past week where a lot of interesting new products debuted as well. We’ll be exploring the implications a lot in coming months, but for now, a few tidbits:

• Inference, the process of running those large language models after they’ve been trained with massive amounts of data, is increasingly moving to the network edge to reduce latency for all manner of applications, such as self-driving cars.

• Big changes are coming to data centers to contend with the massive compute needed for generative AI. Meta and Google both talked about plans already underway to build entirely new kinds of data centers to handle generative AI, including liquid cooling and all kinds of optimizations all the way from the chip to networking to software.

• Alexis Black Bjorlin, Meta’s vice president of infrastructure hardware, said every hyperscaler is scrambling to keep up with gen AI demands. “Compute and storage hasn’t kept up,” she said. “We’re already looking toward liquid cooling,” as opposed to the strictly air-cooled server racks Meta has used to date.

• The crunch extends beyond hardware. “We need to innovate on everything, from models to the underlying silicon,” she said, as well as reducing model sizes through techniques such as quantization, compression and pruning–which collectively have already reduced model sizes by 10 times without a significant loss in accuracy.

• ”The next frontier of innovation is hardware-software codesigned systems,” she added.

• And gen AI won’t be the last word in AI. Amin Vahdat, a Google Fellow and vice president of machine learning, systems and cloud AI, raised the likelihood of artificial general intelligence down the line. “Responsible AGI will require new infrastructure like the internet did,” he said. He showed a mind-boggling chart of the demand for machine learning compute rising 10 times a *year*.

• Specialization at every level of the computing stack likely will be required. “We cannot rely on general-purpose computing anymore,” Vahdat said. That could mean for for some time, standards could go out the window in favor of specialized hardware, networking and data representations.

• Another angle on what’s driving the explosion of compute came from Bratin Saha, vice president and general manager for machine learning and AI services at Amazon. He noted a 1,600 times increase in the size of AI models just from 2019 to 2022. “That’s a level of scaling we haven’t seen in the IT industry,” he said in the understatement of the week.

• Software is in for big changes as well, something that was apparent at Salesforce’s Dreamforce conference this past week (more on that next). “A lot of software in our industry needs to be rewritten, now,” said CEO Mark Benioff.

There will be much more discussion of these trends at our upcoming Supercloud 4: Generative AI Transforms Every Industry. We’re lining up some great guests that we’ll let you know about soon, but meantime, register for the free virtual event and we’ll keep you informed. 

In other AI news

Regulation is coming, even if no one knows in what form: Musk says AI could ‘kill us all’ as tech luminaries attend closed-door session at Congress

A startup has an ambitious plan to create explainable AI: Howso launches fully auditable and explainable open-source AI engine

It’s not just energy that gen AI is hogging: Report: Data centers guzzling enormous amounts of water to cool generative AI servers

Meta’s determined not to be left behind after years of AI research: Meta Platforms reportedly building open-source generative AI system to rival OpenAI’s ChatGPT

Given how much gen AI is going to require changes in how and where computing is done, this startup looks like a company worth watching closely: Enfabrica raises $125M to scale GPUs, memory and storage for distributed AI workloads

And this one too, as AI increasingly gets done at the network edge: Machine learning chipmaker SiMa.ai enables low-code AI model development at the edge

And even more AI at the edge: Edge AI chip startup Axelera debuts Metis AI Platform

Best not to forget China’s fully in the AI game too: Alibaba opens Tongyi Qianwen AI model to the public to drive adoption

SAS ups AI features in its analytics and CRM platforms

Amazon launches generative AI tool for sellers to create product listing information

Salesforce’s heady AI dreams

Say what you will about Mark Benioff, but the man knows how to put on a show, again taking over a few blocks of San Francisco around Moscone Center this week for the company’s Dreamforce conference and entertaining 40,000 people with cuddly characters and a whole lot of trucked-in greenery.

Benioff also knows which way the wind’s blowing, and as generative AI scares the pants off a lot of people, his company has settled on a key message to underlie the flurry of new gen AI-powered services it introduced this week: trust.

Benioff acknowledged the problems with the current obsession with generative AI. “These things are good but they’re not great,” he said during his keynote. “They have a lot of answers that aren’t exactly true. We call them hallucinations. I call them lies. These LLMs are very convincing liars.… They can turn very toxic very quickly.”

Although something as amorphous as trust should be looked at with skepticism when corporations stake a claim to the term, many folks seem to acknowledge that Salesforce is doing more than “trust washing.” It has a chief trust officer, Vikram Rao, plus Kathy Baxter, principal architect for responsible AI and tech.

Salesforce AI CEO Clara Shih noted an “AI trust gap” between AI being the No. priority for CEOs while 52% of consumers don’t believe it’s safe and secure. That was the genesis of a “Trust Layer” that underlies” Salesforce’s platform and applications, covering principles such as data masking, grounding of models in customer data (not the broad internet) and a zero-retention policy for prompts to generative AI applications, as well as toxicity detection and an audit trail.

That resonates with one Salesforce customer I talked to, Kyall Mai, chief innovation officer at Esquire Bank. said the Trust Layer is key for the bank’s customer base of contingent law firms. Likewise, Ian Kahn, head of PricewaterhouseCoopers’ Salesforce practice, said the zero-retention policy is important to the customers he works with.

“The AI revolution is a trust revolution,” Benioff noted at his keynote. We’ll see whether his company walks the walk as generative AI busts into every industry.

Benioff saved perhaps the best news for last: Salesforce to hire 3,300 workers after raising revenue guidance

Cyber beat

David Strom analyzes why It’s the end of the line for outdated internet encryption protocols and what companies should do about that

Chrome, Firefox and other browsers affected by critical WebP vulnerability

And MGM Grand is still down: ALPHV/BlackCat ransomware group linked to attack on MGM Resorts

But Caesar’s paid up, and big: Data breach at Caesars Entertainment compromises customer information

Strom researches the rise of deepfake cyberattacks and how to prevent them: Deepfake cyberthreats keep rising. Here’s how to prevent them

California stays ahead on state privacy protection and is already influencing other states, says Strom

Elsewhere in enterprise, cloud and beyond

The first salvos in a long antitrust case that may well be appealed anyway: In antitrust case, Google insists its dominance of internet search is lawful

And Microsoft is in the crosshairs too: Microsoft Facing Formal EU Complaint Over Teams Video App

“The enemy of AWS is my friend”: Oracle to colocate in Azure data centers under expanded Microsoft partnership

Oracle Cloud may be rising but overall the company needs to start growing faster: Oracle’s revenue comes up short of expectations and its stock slides the most in many years.

Big data is still getting big funding: Finally that Databricks funding happens; doesn’t an IPO have to be next? Databricks raises $500M in new funding at mammoth $43B valuation; also: Data management provider Denodo raises $336M from TPG

Apple’s new phones, like clockwork every September — this time with a cutting-edge chip: Apple’s ‘Wonderlust’ event reveals iPhone 15, Apple Watch 9 and new A17 Pro silicon France doesn’t like older iPhones so much because, um, radiation? Guess that opens up a market for the new ones: As France warns Apple it might have to recall the iPhone 12, Apple tries to dampen radiation fears But it looks like Apple will fix it with a software update: Apple moves to update iPhone 12 in France to address radiation issue

Rob Strechay gets the lowdown from DataWorks co-founder Alex Hutchins: The state of building modern data teams

Even GPUs can’t do everything, but they can simulate the next or the next-next era of computing: Nvidia and Xanadu power quantum computing simulations with GPUs and Zeus Kerravala’s analysis: Nvidia’s cuQuantum software development kit sings PennyLane’s song

AI isn’t always a magic wand for stocks: AI helps Adobe beats expectations and boosts its operating margins, but its stock falls anyway

Not sure whose fault a 70% drop in sales is, but it’s certainly understandable why former Flexport CEO Dave Clark is gone.

What’s next

CrowdStrike Fal.Con Sept. 18-21 in Las Vegas, the heart of which will be covered by theCUBE Sept. 19-20. Vellante will be there.

mWISE from Google’s Mandiant threat intelligence unit, Sept. 18-20 in Washington, D.C. Furrier will be there.

Oracle Cloud World Sept. 18-21 in Las Vegas and virtual: We’ll be watching from afar this time, but we’ll have all the news as it happens.

Intel Innovation event Sept. 19-20 in San Jose and online, with CEO Pat Gelsinger and others keynoting, and we’ll be reporting the news.

TechCrunch Disrupt Sept. 19-21: We’ll be there for all the enterprise and AI news and interviews.

Image: Bing Image Creator