Runtime: Jensen fired, a Google Cloud post-mortem, Shell's new CIO

+ CoreWeave sighs in relief, decision models are having a moment, and Google Cloud's power problems in London.

Nov 12, 2026 3:14 PM 5 min read

Runtime: Jensen fired, a Google Cloud post-mortem, Shell's new CIO — Image credit: Credit: WEF/Mattias Nutt, CC BY-NC-SA 4.0

Welcome to Runtime! Today: Nvidia's board has swung the ax and is shopping internally and externally for a new CEO; Google Cloud's bad hair day in London revealed (when N2 becomes N1 and then N0...) and more.

Please forward this email to a friend or colleague! If it was forwarded to you, sign up here to get Runtime for free every week, or level up here.

So long, and thanks for all the GPUs.

Jensen Huang has "resigned." It's still looking for a permanent new CEO. It is hard to imagine the depth of personal relationships accrued across over three decades at the help of Nvidia and with its myriad suppliers (a Blackwell unit alone has over a million parts) that now need replacing/rebuilding. Hot off the press below. More reaction/analysis to follow in The Stack.

The rest of The Stack

Dropping Shells: In other people moves, Shell's longstanding Group CIO is also resigning. The supermajors don't get much public focus on their enterprise technology: They should, as they are often doing some extraordinarily cutting edge work with remote management.

Kate Platonova takes the helm from January,

Decision time: Companies are finally starting to deploy AI agents after years of starts and stops, but we're a long way from settling on the best practices and invoking the hard-won wisdom that accompanied the cloud computing platform shift. There's perhaps one central point of agreement: it's kind of expensive and hard to coordinate.

That explains much of the buzz that is currently surrounding decision models, which unlike large-language models don't need to load the world's entire corpus of written material to produce their output. Instead, decision models like TypeSafe's Jev — the ChatGPT moment for this movement — generate outcomes that translate very well to agents and cost a fraction of what companies are currently spending on LLMs.

"It’s hard to give a single universal percentage because it depends entirely on how long the text answer would have been, but you save a massive percentage of GPU, roughly 30-40% and also inference time and compute for typical workloads," Microsoft's Shreyan J D Fernandes told The Stack. AWS, Cloudflare, and Datadog are all using decision models and offering them to customers looking for alternatives.

Decision models won't replace LLMs, but they could take on some of the burden for workloads that don't require text-based output and further reduce the cost of deploying AI across the enterprise. That could make it much easier for companies currently on the sidelines to get started, and if they catch on could send a warning shot across the bow of the big AI model companies.

I'm not dead yet: CoreWeave's stock is still down almost 24% on the year thanks to worries about the long-term viability of the neocloud business model, but it got a big boost this week after better-than-expected earnings results. “Customer demand remains extremely strong, and our newest contracts are among the most profitable we have ever underwritten,” CEO Mike Intrator said on the earnings call.

Power laws: Finding enough electricity to stand up new data centers is hard enough, but keeping the power flowing through existing ones was thought to be a solved problem. Apparently not at Google Cloud's London availability zone, however, after a local power outage caused a series of cascading problems with backup power systems and triggered another problem when trying to replicate data to other zones. Critically, some customers with cross-zone failovers found they couldn't...

Ain't patching grand: The good news is that after months of incredible growth in the number of software vulnerabilities identified each month on Patch Tuesday, the overall number declined for November's release. The bad news is that drop still contained over 1,000 patches, and one of the flaws identified is being actively exploited.

Web swamped: Severe weather forced the organizers of Web Summit to cancel the conference on Wednesday, where executives from Arm and Lovable were scheduled to speak. But the show was expected to go on on Thursday.

Stacked up: The week ahead

SC26 kicks off on Sunday in Chicago, with the annual gathering of the supercomputing community expected to feature a keynote address from Caltech professor Animashree Anandkumar.

Nvidia is expected to report earnings on Tuesday, and it's tough to figure out what they're going to talk about.

Microsoft Ignite kicks off Tuesday in San Francisco, and I'll be on the ground meeting with Microsoft executives and customers during its most enterprise-oriented event.

Anthropic is expected to begin trading shares in public on Thursday, which promises to be a spectacle.

We're also reading:

More Suicides at Spy, Cyber Agencies Scrutinized by Congress: The US government's hacking machine is coming apart at the seams, at a terrible cost.

I Am a Silicon Valley Pioneer. We Are Sabotaging Ourselves: "Americans are not rejecting A.I. They are rejecting the way Silicon Valley is defining it for them." Jaron Lanier attempts to place a wake-up call to Silicon Valley in The New York Times.

Thanks for reading — see you Saturday!

Add The Stack on Google

Tom Krazit

Tom Krazit is The Stack's Editor-in-Chief. He 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.
Edward Targett

Ed is a co-founder of The Stack. He previously edited Computer Business Review. He has also covered energy markets. He started his journalism career on local papers. He left school at 15 and has made a living asking "but why?" ever since.