In fact, I'd support Google with this policies if they'd also take full accountability if my phone gets hacked. Do they? No, so it's unfair and a bad authoritarian way to create another walled garden.
A computer created around a GPU and/or a shared ram to run LLMs should not be called a computer anymore. It should be called an inference engine.
I'm not against it but it could be stripped down to reduce some cost and it would be the perfect companion in a lab.
To me context means everything.
Tokens per second is a great metric but in the real world context window is the deal breaker when a real use case is on the table.