To The Moon

I attended an investment forum today where finance counterparts, be it from fund managers or big banks come down to share their thoughts on what the outlook of tech will look like in the next few months. Or to put it in hyped lingo, how AI will generate insane profits.

And its really interesting to hear how the finance counterparts try to articulate how AI is not a bubble. To quote, "we see ASML ahead of the race... advanced 2nm chips... for end consumers device...". And I wonder, how true is this statement. An investment analyst who i also spoke to had strong convictions that AI isn't a bubble, or rather China won't be able to take over the AI race from the US any time soon (and hence the bubble shouldn't pop for the Mag7)

Let's break this down. ASML produces the machinery needed for the UV etching on silicon wafers. It is needed true that the Dutch company has the most advanced etching technology. Feel free to correct me if i am wrong, I'm not a EEE trained person or a chip expert, but being able to hold monopoly on such a device must mean that it's very complex and costly to produce these type of cutting edge chips. The industry goes to as far as terming chips 5nm, 3nm and now even 2nm. As end consumers, we are brought forth with the idea that the theoretical limits of a transistor gates are 5nm but now there are 2nm chips? Google it, and you'll find that this is a marketing gimmick, confusing the general public to think that ASML might have somehow solved some physics mystery.

We are then also told TSMC and ASML will create the most advanced 2nm chips that will revolutionise how we will train and use AI. How HBM is going to king because LLM eats RAM like crazy. We somehow associate AI = Chatgpt. I find this extremely misguided.

Most devices run on ARM chips not X86 processors, which are the said 2nm chips. We are not going to foreseeably run chatgpt on our edge devices, maybe a small LLM, but as we get lazier, i cannot imagine how we humans can accept sub par response from smaller models. Even if we tried running LLM on our gaming PC, it can't even compare to the massive compute requirements of LLM today. So let's put it this way, 2nm aren't going to suddenly make our AI faster if you're thinking LLM, but if you define AI as a scheduler or even excel then probably yes.

Another misleading statement i personally find is the claim on how 2nm chips which is now dominated by the US is the holy grail to AI training. There is a difference between CPU and GPU. CPU is meant to handle extreme complex logic, routing, logic processing. You can take it as the CEO of the company. While GPU excel in parallel work loads. Solving the same math problem with different underlying data input, take it as the ops people of a company, hired to do saikang work. Saying that you're going to use 2nm processors in GPU is like saying you think you can fire all your workers today to replace them with CEOs. Sure you could, but you probably are going to pay for the GPU with your kidneys and liver.

I find that the finance sector is increasingly misleading with their statements and analysis. The valuations and statements that they claim, very much so ill informed. Target prices that are based on "what i learnt from chatgpt in 5mins". And if you think China is extremely far behind in terms of AI, think again. Just look at the number of open source models that they are releasing. The number of uptakes of the open source models that companies in the US are adopting. Just because you are placed a constraint on hardware doesn't mean you can't improve and optimise software.

This to me, looks like an establishment of a foundation. A very deeply rooted one. And ill be extremely cautious about the numbers printed on analyst reports.

Do your own due diligence.

$ASML Holding NV(ASML)$  $Taiwan Semiconductor Manufacturing(TSM)$  $Intel(INTC)$  $Arrow Minerals Ltd(AMD.AU)$  $NVIDIA(NVDA)$