How so? I don't see DL failing to live up to anything said about it by actual Deep Learning researchers. If anything, DL has been a stunning success. And I say this as somebody who is very much in the Gary Marcus camp of believing that hybrid systems (that include elements of what we might call GOFAI, like symbolic reasoning, etc.) are going to be required to get to "strong" AI or AGI.
That means the question becomes "Why does anyone continue to invest in AI?"
I've argued this before, but I'll reiterate my position this: we don't need "strong AI" or AGI to do amazing, useful, and valuable things. We continue to invest in AI because it continues to get better and create more value. Same reason you invest in anything, really.
Is Big Tech trying to create Roko's Basilisk ...
I doubt it. It's not even generally accepted that the argument for Roko's Basilisk is compelling enough to consider the idea a Real Thing. I seriously doubt many (if any) people involved in AI lose even a microsecond of sleep thinking about this.