Jay Cuthrell covers open-source AI models like Moonshot KIMI K3 and Apple's Siri beta, New York's data center moratorium, and enterprise shifts including IBM's stock drop and potential fintech consolidation. Regulatory pressure and a chip market correction tied to tokenomics are reshaping the AI landscape.
Chapters
Welcome to What the Fudge for Sunday, July 19th, 2026. This is the growing weekly habit edition. I am going to break down a couple of topics today. The first will be around open source AI, the notion of regulatory headwinds, and then we'll talk a little bit about what's going on within the enterprise capital world. So the first thing is on this notion of the AI ecosystem and what's going to be taking place either on your device in an AI factory near your device or in the large hyperscale public cloud. The question will be, is, are we looking at a shift from purely cloud-based, very proprietary frontier models from the bigger players that you would recognize, the Googles of the world, the Anthropics of the world, the OpenAI and ChatGPTs, or are we looking more like things that could be localized, be open source, be available on your device, be available for you to run in your own private cloud. So the question will become, how do we view some of those Chinese companies like Moonshot? They release something like KIMI K3. It's got literally trillions, uh, 2.8 trillion parameters in that model, showing that that's possible, uh, to compete against things like Claude, uh, their latest, uh, Opus 4.8, or ChatGPT OpenAI's 5.5. So that, that gap is kind of shrinking between these proprietary and open weight models. Um, you'll also find that there's, uh, coverage of all the ex-OpenAI folks that are starting their own open weights models as well, so you can train with your own data. So looking further downstream, you also know that there is the packaging of AI that will be accessible on iOS devices, Apple iPhones for example. And so you've got this new beta where Apple is sort of finally showing what Siri is capable of if it did actually have these true AI features, as opposed to what was previously just a, um, ask ChatGPT for that as a kind of workaround for the limitations of on-AI device that Siri had for the original rollouts. The next piece of this, of course, comes with a company I mentioned previously, PrismML. They're now demonstrating obviously the ability to take these very highly optimized large language models and compact them to have them operate on things like an iPhone. Granted, it's a, it's a high-end iPhone, but nonetheless, it's showing that you can ship a solution now. And so I think what's going to take place is larger enterprises are going to figure out they can transition over time going to more localized, more on-device use of AI and intelligence. And then if they have built internal AI factories, they can deploy that into a private cloud that is incredibly independent of what would be required to take those same things and say set fire to tokens, if you will, on a massive public cloud footprint. And so, so looking at all these things, it's important to understand that as that's taking place, the governments of the world are making their choices. And so, whereas the United States might be the Wild West, you now see a couple of different things. First is the precedent that New York has now put into law. They're blocking effectively anything greater than a 50-megawatt data center from getting an environmental permit. So if it happens in New York, it's a, it's a big state, so it could happen in other states. So this could set precedent for more nationwide adoption of those same kinds of moratoriums. Then if you look at what's happening with proposals and standards bodies, if you're familiar with FINRA in the financial space, there's the possibility that someone will lobby for Washington to make policy, and that policy would centralize all the frontier labs that are doing amazing AI research would need to submit those models for review prior to the general availability. We see some of this right now with Department of War, Department of Defense, some of those machinations, but this would be formalized. This would be an actual standards body, much like what happens with the financial industry that gets regulated. So the last piece, and I just, this is kind of funny, apparently you can't use generative AI to call in the plays if you're a coach in the dugout. My quip, of course, would be is I don't know that, uh, you do— do you eject the, the iPad and the coach, or it's just the coach gets ejected? I'm not really sure. But, um, last couple things real quick, wrapping up. IBM took a massive drop, um, something related to them saying that, well, the, the budgets went elsewhere. Instead of you spending money on mainframes or our software, you're putting into chips and AI infrastructure. The next piece, of course, is that there's going to be consolidation in fintech along the lines of, say, Stripe and Advent going in to buy PayPal. But the last two things I want to cover in the last 20 seconds is, uh, you know, Grok Build, uh, has done some oops-a-daisies. And then the last thing is just that if you look at the general markets, a lot of these, uh, chip and storage providers had a correction because over time, as I said, tokenomics is real and it costs real money. So that's the edition for today. Take care.
