General-purpose computing isn't going anywhere

Welcome to Runtime! Today: Why Nvidia's Jensen Huang might be getting a little ahead of himself, Hugging Face makes cute with enterprise vendors, and this week's enterprise tech moves.

(Was this email forwarded to you? Sign up here to get Runtime each week.)

And, not or

Let's all take a breath.

Nvidia CEO Jensen Huang is having one of the best years in modern tech history. His company's prescient investments in AI-centric hardware and software design are paying off with astonishing earnings results, as the generative AI boom pushes demand for its AI training chips to new heights.

But that's not enough for Huang. On a call with financial analysts Wednesday, he argued that Nvidia's "accelerated computing" approach to computing will soon displace the general-purpose computing approach that was the platform for the last 15 years of enterprise tech applications.

• "Using general-purpose computing at scale is no longer the best way to go forward," Huang said, as transcribed by Motley Fool. "It's too costly, it's too expensive, and the performance of the applications are too slow, right?"
• He argued that Nvidia's chips — which require software developers to follow a different blueprint than most applications — will soon be able to take on more of the distributing computing tasks currently handled by CPUs designed by Intel, AMD, and Arm licensees like AWS.
• "Going forward, the best way to invest in a data center is to divert the capital investment from general-purpose computing and focus it on generative AI and accelerated computing," Huang said.

The generative AI boom has surfaced something that's been true in datacenter computing for several years: GPUs like Nvidia's absolutely do handle the specialized workloads needed for AI better than CPUs.

• More than a decade ago, Nvidia realized that parallel computing — in which several different tasks can be worked on simultaneously — was the best way to handle certain kinds of compute-intensive workloads, rather than waiting for a traditional CPU to process those tasks sequentially.
• As cloud computing exploded and interest in AI awakened after a long slumber, cloud providers snapped up Nvidia's GPUs and began designing and deploying their own GPU-like chips for customers.
• That trend has been in motion for a long time now, and is the main reason why Nvidia suddenly can't build chips fast thanks to the surge of interest in training AI models.

But what Huang is suggesting takes that trend a step further; that all types of applications will run better on Nvidia's "accelerated computing" chips, and that cloud providers and data-center operators should stop spending so much money on CPUs.

• Such a shift would require a massive amount of reinvestment around Nvidia's technology by both infrastructure providers and the software community, both in terms of time and money.
• It assumes a relatively basic enterprise application — the opposite of high-performance computing — would run so much better on an expensive Nvidia GPU than a relatively basic CPU as to demand that level of reinvestment.
• And it assumes that the world's cloud providers and software developers are eager to throw all their eggs in one vendor's basket for the next decade.

Change comes slowly to the enterprise, no matter how many billions of dollars are spent every year trying to convince CIOs and developers that they'll be left behind if they don't jump on the trend du jour.

• Nvidia's accelerated computing vision will be a major part of every enterprise infrastructure computing strategy for the foreseeable future, even if demand for generative AI applications wanes in the coming years as the hype cycle runs its course.
• But it took more than a decade for mainstream tech buyers to embrace Nvidia's AI computing philosophy.
• It's hard to imagine that shift happening any faster when it comes to all applications.

Keep your friends close

After Microsoft cemented its relationship with OpenAI last year, its rivals have found a love connection in Hugging Face, which has positioned itself as the more-open version of OpenAI. This week a who's who of enterprise tech vendors sank $235 million into the AI platform startup, valuing it at $4.5 billion.

The new investors include Amazon, AMD, Google, IBM, Intel, Nvidia, Qualcomm, Salesforce, and Sound Ventures. "It solidifies our position as a neutral platform, or the Switzerland ... for AI," Hugging Face CEO Clement Delangue told Axios.

It also likely solidifies a lot of inbound business for those investors from Hugging Face, which has already inked a partnership with AWS. Hugging Face has ambitions to become the GitHub of AI, which will require a substantial investment in infrastructure tech should it want to become a daily destination for those working on the generative AI boom.

Enterprise moves

Carsten Thoma is the new president of Celonis, after working with the process-mining company as its first external investor and board member since 2016.

Lori Williams is the new CEO of Caylent, joining the AWS partner from Salesforce partner Traction on Demand after it was acquired by Salesforce last year.

The Runtime roundup

AMD acquired Mipsology for an undisclosed amount to bolster its AI software assets in competition with Nvidia.

The days of unlimited storage at Dropbox are officially over, after the company said cryptojerks and storage resellers were ruining it for everybody else.

Snowflake beat Wall Street expectations for revenue and maintained its guidance, which was interpreted by the day traders as a sign that enterprise tech spending might have hit bottom and is starting to stabilize.

Splunk bolstered that argument with beats for both revenue and profit, and it increased its guidance.

Ahead of Google Cloud's big event next week, it released two security updates that use AI to help detect if employees are sharing sensitive information and prevent administrators from making big changes unless two people sign off.

A Danish cloud provider lost all of its customers' data in a ransomware attack, and cloud provider CISOs around the world winced.

The Oracle-ByteDance shotgun marriage to manage TikTok's U.S. presence is as much of a clusterfuck as anyone might have expected, according to Forbes.

Thanks for reading — see you Saturday!