From their pitch the main target for these first chips are mobile, it's seemingly meant to compete with the RX5500M and equivalent lines, they're still a notable step up from what's possible with on-board to be fair.
But it's important they can get a small simple example of the architecture out quickly for software development when it's a completely new arch. Even if you could dive straight into making 4000-shader core GPUs from day one (Which you can't bc it takes a ridiculous amount of care/R&D to scale up architectures in bandwidth or latency sensitive environments), they'd have nothing intensive to run it on properly.
They'll certainly scale up to larger GPUs since their targets include datacenter & AI where space efficiency is a big part, the question is whether or not they can maintain price competitiveness with the established competitors for the consumer sphere while doing it.