Looking at the essay more, I would say: the chip fabrication labs are amazingly cheap and even highly coordinated efforts by all terrorist groups can not make even a small dent in the progress (excluding scenarios such as big nuclear war)

The several billion dollars apiece for a fab is cheap . The margins may be thin when everyone else, too, is making chips. But I can pay 10x of what I am paying for my electronic equipment, if needs be, and all chances are you can too if you are posting here. The income distribution being what it is (power law, see Pareto distribution), the costs can be raised massively while retaining the revenue as long as the cheap alternatives are not available. More than 50% of your customers can and would pay >2x (if the alternatives weren't available). Think about it. 20% own 80% of everything (or even more skewed), and its like this all the way to the top. With this kind of income distribution the luxury market alone can support entire industry along with progress. Look at Apple, and how successful it is even though the products are not all that popular worldwide - they make huge margin.

Anyhow, with regards to the de-novo AGI, really, all we have is extremely shaky guesses as to whenever de-novo AGI would be more or less disruptive than mind uploads, arguments which have extremely low external probabilities. Due to this uncertainty bombing fabs would result in negligible change of expected post-AI utility, as calculated by any rational agent that does not delude itself with regards to the strength of it's evidence (and argumentation) towards de-novo AGI being safer than mind uploads. (note that the negligible difference of expected value is different from negligible expected square of difference)

(the pascal's mugging, i think, is only a problem due to our self delusion with regards to strength of the prediction of superpowerful entity based on what superpowerful entity says; it can be that the superpowerful entities too would usually lie when doing pascal's mugging, and sufficiently often lie in an evil way involving eternal tortures when you accept what they say. This overconfidence in own prediction is a huge problem, manifesting itself not just in theoretical examples like pascal's mugging but realistic and soon to be practical examples like attitude towards de-novo AGI vs mind upload)