You are manifestly incorrect. The ‘people you know who know AI’ don’t know jack if in fact they told you that. Which if they really did ‘know AI’ they won’t tell you that.
We are at the stage of integration of generative LLM based ‘ai’. Lots of companies are using these outstanding systems poorly. They are given a Lamborghini but use it as a sheep carriage.
Trust me the public systems are very able to carry out complex tasks. Let alone military ones. Palantir has declassified examples of battle management systems which are publicly available.
Ignore your friends.
LLM and general AI are two different topics you seem to be confusing.
generative AI (creating images/videos/media) And general AI are two vastly different things. The latter is known as a computer that is able to think and process information more similarity to the human brain.
LLM have plenty of flaws in its current state with regards to correctness and all that is mere data regurgitation. No one is downplaying their significance to civil society, but they are not apt to be used for military means.
General AI is still far from military use unless DARPA is far ahead and hiding it. Nor is there any indication that Iran has some type of foothold in this area.
Quantum computing is still progressing slowly you can get brute power (but high incorrect entanglement states) or low power (but higher corrective entanglement states), but not yet both.
The first binary supercomputers that can exceed human brain synapse calculations per second have just started coming online. So again technology is very fresh.
Lastly you have yet to explain how Iran will circumvent the hardware problem. You glossed over it.
Iran has little to no GPU/semiconductor capability.
Powering a [future] ‘army of killer robots’ with Western tech; what could go wrong?