This content was created by Jay Cuthrell. This content was uploaded by Jay Cuthrell. This is a YouTube podcast episode titled "What The Fudge for June 07, 2026: The IPO Parade and Device-Level Agents" (published June 7, 2026). A single host (labeled SPEAKER A) delivers a short monologue covering two main topics: the large anticipated AI-related IPOs (SpaceX, Anthropic, OpenAI) and their market effects (NASDAQ vs. S&P 500 inclusion, potential trillions in market value and FOMO), and the shift toward device-level/sovereign agentic AI (smaller localized models such as Google’s Gemma-class models, running on limited RAM, examples like Ollama, and implications for cloud vs. local inference). The host also briefly raises ethical and governance questions about when technologies should or should not be used and notes emerging product and chip competition (Google, Microsoft, NVIDIA) for on-device AI.
Chapters
And welcome to Want the Fudge podcast for June 7th, 2026, the I took a few days off edition. This time around it's all about the IPO parade, and in this case it's, uh, gigantic IPOs. There's also this notion of what's going on with the concept of agentic AI and what we're seeing in terms of adoption. So 2 big things, this huge capitalization event happening in the public markets, then what's going to be taking place as you go from AI that could be moving from very large capital-intensive projects to the alternative side, which is AI moving more discreetly to devices closer to where arguably a lot of the data might be created, whether that's humans or environmental. Or the Earth models that we're about to be learning more about. So first up, there is this concept of SpaceX is going to IPO, Anthropic is going to IPO, OpenAI is going to IPO. And when they do, when these companies do go public, there's this discussion around the trillions in new market value that will come out of that process or series of IPOs. However, there are some challenges on the NASDAQ versus the S&P 500 versus if you have a 401, you know, which, which organizations are more or less committed to buying into all of this. And so I think S&P right now, based on the latest news, was that they've sort of— they're sticking to their timelines of when you would be able to include any of these as part of the S&P 500. That means that effectively, if you do have parts of your portfolio, if you have a portfolio that would involve purchasing or auto-purchasing of S&P 500 versus NASDAQ, you'll certainly be seeing the NASDAQ pick up, but probably not the S&P 500 until after the traditional 12-month period. This means that if you look at what happens, there's probably going to be— if you remember bonds. If you remember the jokes in movies about junk bonds, which were very much a thing in the past, it is interesting that there's a lot of interested parties and being in this fear of missing out wave. So really interesting to see what's happening there, getting away from just the pure infrastructure and, you know, trillions of dollars related to what's going to happen most likely in the market. There is also this shift of what happens when AI becomes less about AI itself and more as a design concept being part of your overall user experience. And so you will see provocative things in the device level layer coming from companies like Google, for example, where they push out things like a Gemma 4-12B, which is a a model that is designed to run on something that has a limited amount of RAM. And in, in some instances, this means that you're going to probably see smaller language models, more discrete, more powerful, but smaller language models that don't require these many, many, many, many trillions of parameters. And so sometimes billions of parameters is more than enough. And in these cases, you'll see the Googles, the Microsofts, the NVIDIAS, all trying to either push models and/or device chips that could fit into things that look more like what we would call, hey, I could carry that around. More importantly, it's sovereign AI because you don't necessarily have to move to a network to take advantage of those particular models. Now, to set this up, there is a lot of steps, to be clear, but if you're familiar with, um, Ollama or any of the other on-desktop approaches to getting this done, it is going to be fascinating to see who can run localized models and who will be running exclusively cloud models and what kind of a match or mix and match there might be along the way. So, um, as I said, we're always trying to hit that 5-minute mark with this podcast, so I'm going to also break this into another section which I'll call what happens if just because you can doesn't mean you should. So I think there's also going to be an increased amount of ethics and scrutiny and moral philosophizing around whether we should or shouldn't do things with these technologies that we've talked about. So if you, you know, go into a post-mythos world, I'm going to eventually write about this as a separate blog post because it won't fit into 5 minutes, but if you're familiar with the shift left, of moving security further to the left in the CI/CD pipeline. If you think about organizations that ship their org charts, that old bromide, I want to tie that together, but it won't be in this podcast. So at a high level, big things that you want to keep an eye on are related to— of course, hey, look at that, went over 5 minutes, man, and I was doing so well, but such is the case. So I'm going to wrap it up. Thank you again for hanging out, for The What the Fudge podcast for June 7th, 2026. Be safe and be well.
