I remembered hearing that same line when I bought a Raspberry Pi in 2012. "It's useless! You can't run x86 software on it, so what's the point?"
Flash-forwards a decade and now Graviton instances are blowing up like nothing else in the industry. RISC-V is in a very similar position to ARM 10 years ago; the groundwork has been laid, standards have been ratified and base packages/several kernels work perfectly fine on it. The only difference is that ARM is more expensive to license and is less flexible.
> Only the base RISC V is guaranteed thanks extensions.
Yeah. Is that a problem? The situation on ARM is equally bad if not worse (frequent iterations end up throwing even relatively recent CPU models under the bus), and the reason why RISC-V divided itself into extensions is so that you didn't have to start from scratch when John RISC decides to add in 3 new floating point instructions. It's a pretty damn good compromise if you ask me, and it certainly doesn't have any bearing on software availability; RISC-V programs run on RISC-V processors. ARM does not have that same liberty.
There are plenty of genuine constraints for RISC-V (the majority of them in the manufacturing/mass production side of things, now), but the majority of these software issues have been solved and taped out years ago.
The problem is dreamers thinking RISC V will be any different than other CPUs in the industry when big players come playing.
LLVM, contributions level at the scale of Linux kernel, C++20 support now lags behind everyone else.
When money comes to play, the rainbows and flowers eventually turn into wall street yuppies.
Only the base RISC V is guaranteed thanks extensions.
Also you are forgetting that just like Android and ARM, there are other forces at play that don't make it as easy in practice as FOSS advocates wish for.