Hardware is well understood to be OpEx intensive, and have high up-front costs. Look at the pricing for injection mold dies for example. The opposite of a good fit for an iterative approach.
In my defense I was thinking of the GPUs and servers, not the client devices (does stadia have clients?). In that scenario, might HW be capex -- a durable good you buy and use for many years, and can sell if you no longer require it?
> While Stadia can use any HID-class USB controller, Google developed its own controller which connects via Wi-Fi directly to the Google data center in which the game is running, to reduce input latency.[8] Google is also exploring further ways to reduce latency, using an idea called "negative latency" which involves prediction of user input through various means so that any apparent network lag between controller and game response is minimized.[12] During its GDC 2019 keynote reveal, Google confirmed that the controller would also feature Google Assistant, which will automatically search YouTube for relevant, helpful videos related to the game they are currently playing at the touch of a key.[13]
I don't think any of Google's manufacturing capex is recoupable here; it's specific to making that thing, and if you don't want to make any more of that thing it's trash. (I'm not a hardware guru though, so that's not a statement made with high certainty. There might be bits that are reusable.)
