My perception at that time (autumn 2017) was that Tensorflow had the biggest unforced fumble in the modern history in artificial intelligence, not having the vision to make an API aiming the researchers.
They made the almost impossible and took down Theano and Torch in less than 2 years and fueled with a lot of money from Google got the ecosystem for at least 2 years.
I think they thought like “hey, we won’t the comercial ecosystem, so it’s over; academics and research will follow; even this inflexible and cranky API. If they complain, push Keras to them all”.
What happened in reality was, at the time that Facebook made an absolutely wise decision to lift Torch and unite research and a better UX, it just twisted the situation that for me it’s just bonkers to imagine how Google let that happen?