Author here. I've run a Strix Halo box daily for six months, and the question I get most is whether to buy now or wait.

Two things make waiting a worse deal than usual right now.

First, the memory is soldered and it's getting more expensive, not cheaper. LPDDR5X is up 89% in a single quarter and analysts don't expect real relief before late 2027, so the capacity you buy is a permanent decision made at a bad time. The same box I paid under $2,200 for is around $2,800 today.

Second, for token generation these boxes are bandwidth-bound, and the announced successors add capacity at the same bandwidth. A 192GB or 768GB successor lets you load a bigger model and runs it at the same speed, so waiting for the bigger one only helps if you're capacity-limited, not if you want it faster.

Prompt processing is the exception. It's compute-bound, and there the CUDA box (Spark) is much stronger while Apple and Strix Halo are weak. The M5 Mac is interesting precisely because it targets that with matrix accelerators, and it's the one place I'd actually consider waiting, with the caveat that Apple just raised prices and the timing isn't certain.

Happy to answer questions.