A while back I came across someone’s write-up of a 1U firewall appliance they’d picked up from AliExpress for $290. Intel N100, four 2.5G Intel i226-V ports, a PCIe slot for SFP+ modules. Runs pfSense. One year on, they reported zero freezes, zero reboots, quiet as you like, cool enough in a home office with no air conditioning.

That last detail stuck with me. No AC and it just keeps going. That’s the kind of reliability review I actually trust, because nobody’s trying to sell you anything.

There’s a post doing the rounds that stopped me mid-scroll last week. Someone has assembled what they’re calling the infinity stones of local AI inference: 2.3 terabytes of RAM, 400-plus vCores, a Blackwell GPU for prefill, and a mesh of Mac Studios for decode. They want to connect the whole thing via RDMA over Thunderbolt and run disaggregated inference across heterogeneous hardware, essentially splitting the “thinking” work across fundamentally different architectures.

So AMD apparently just dropped some news at their AI Dev Day about releasing their own in-house Ryzen AI 395 mini PC box, coming in June. And the tech corners of the internet are… cautiously underwhelmed? Which, honestly, is a pretty reasonable reaction when you dig into what it actually is.

The short version: it’s a 395 with 128GB unified memory. Same as what you can already buy from a dozen different vendors right now. No extra bandwidth, no architectural magic, just AMD putting their own name on the box. One person who was actually at the event confirmed it directly with an engineer on the floor — just a standard 395 system, nothing more.

I’ve been following a fascinating discussion about Minecraft server performance lately, and it’s got me thinking about something that frustrates me to no end: the way we’ve been conditioned to believe that cloud services are always the answer, even when they’re not.

Someone ran some tests comparing Minecraft chunk generation speeds across different setups - from expensive Hetzner cloud instances to a decade-old CPU that’s barely worth anything. The results? That old hardware was holding its own remarkably well against modern cloud offerings that cost significantly more per month.

Reading about someone’s 14x RTX 3090 home server setup this morning made my modest 32GB VRAM setup feel like I brought a butter knife to a nuclear war. This absolute unit of a machine, sporting 336GB of total VRAM, represents perhaps the most extreme example of the local AI computing arms race I’ve seen yet.

The sheer audacity of the build is both impressive and slightly concerning. We’re talking about a setup that required dedicated 30-amp 240-volt circuits installed in their house - the kind of power infrastructure you’d typically associate with industrial equipment, not a home computer. The cooling requirements alone must be enough to heat a small neighbourhood.