Jay Cuthrell covers AI infrastructure trends including silicon photonics partnerships between Corning and NVIDIA, policy-as-code approaches to compliance, and aerospace developments like NASA's X-59 quiet supersonic aircraft. He discusses how Anthropic and other vendors are reshaping the market, plus emerging concerns around consumer AI hardware and data control.
Chapters
And welcome to the What the Fudge podcast for May 10th, 2026. I'm going to actually try to get this into 5 minutes this time. I've been working on my workflow, doing some Python, using some tools. Here we go. First up, the AI arms race, if you will, of infrastructure and overhaul. We're starting to see now that co-packaged optics and networking is coming into visibility. You're seeing references to Corning with NVIDIA and that partnership. I'm a materials engineer by training in college. I didn't really go into that, but it is interesting to see how ceramics-type companies, ceramicist-type companies, are interested in being and participating in this. And this is the whole idea that you're replacing all the copper with optical fiber. The next thing you'll start hearing about, of course, is the notion of these silicon photonics systems. I'll maybe cover that, in a later date. But the, the notion that this, you, that you need all these things is also part of the vertical, um, integration of some of the companies. So for example, Anthropic has decided that they're going to, uh, I believe, uh, possibly purchase, um, some inference accelerators from Fractile. Uh, we'll see how that goes. Um, there's also some other interesting things happening in the world of, uh, battery innovation. So, uh, I've quipped about this in the past with, uh, Apple with as much money as Apple has in terms of cash, just the amount of research they could do and just the battery chemistry is just kind of phenomenal. But on the other fronts around data control as it relates to AI, I think that even hearing names that are more what you would call traditional, like Teradata, Teradata has now also updated their data platform. And this is, again, everyone's trying to get into this world of policy control, policy as code, centralization, automation of the policy. Compliance, and that's just going to be part of that Agentic AI story. The next piece would be is around the UI or the interface around AI, which would be these concepts around agents, harnesses. Now you have Google who's got Remy, which is, you know, an equivalent to what you had seen previously. We also now know that infrastructure OEMs like HPE, Dell, and others are starting to roll out their own AI agents to automate the workflows associated with things like provisioning or diagnosing issues that might be specific to network equipment since that's a fairly straightforward. That coming to servers, physical compute, you know, embodied in a rack mount form factor or something, the edge will be interesting. But there's obviously going to be some consumer parts of this, but on the enterprise side, I think that's just where a lot of this is going to begin. And in the consumer realm, you'll just have a series of possibly a camera explosion of different devices with ill-equipped or ill-advised uses of AI on the hardware. So do I really need to talk to my toaster, for example? And then the other element will be around the software side of where the new interface comes in. And so watching how AI engineers and traditional software development engineering executes as though they've talked to each other. And the next piece will be around, I would say, the disruption, the security, the governance element of what's also happening. I don't necessarily put a lot of stock in the workforce reductions because of AI stories. I think that there was overhiring. That's my opinion. I do see some of the data coming out, but I also think in terms of the amount of code shipping with poorly tested and poorly scanned for security vulnerability is real. We will definitely see some AI slop in terms of security. But on the other hand, I also think there's gonna be those that will use some of the new models to actually harden and secure and shift left a lot of that security in the DevSecOps pipeline. I'm gonna wrap up there. Again, the goal this time was to stay underneath 5 minutes And then just the last minute that I've got here, I think it's awesome. I always call out something that's happening in the space value chain. And so if you remember, I mentioned NASA's X-59 experimental, you know, them hitting just under sub-Mach speeds is actually really, really interesting. And so the path from sub-Mach speeds to now ever quieter subsonic travel over, supersonic travel, I should say, over land will actually be pretty interesting also. If we can see things before they become things that we have to worry about. If you remember the dystopian asteroid defense challenges of the movies of the '80s and the '90s and the 2000s, that's also happening with the Near Earth Object Surveyor. And so that's, that's everything. Again, I told you I'd get it under 5 minutes.
