Observing the growth of Broadcom, we can see

Observing the growth of Broadcom, we can see once again that the development point of IT technology is under the paradigm of "high speed," "high capacity" and "low cost."

Among the three elements above, technology is evolving in the order of "high speed -> large capacity -> low cost." Other elements are secondary.
.

Companies that provide faster, larger capacity, and cheaper solutions win.

Most technology companies approached "This speed is enough," "This capacity is enough," and "It'll be used if there's no alternative." Just as Oracle, which still operates around RDB in the age of big data, is no longer Oracle of the past.
.

Of course, there could be a counter-moon.

"NVIDIA equipment is too expensive. Is it low cost?"
As Nvidia equipment is the cheapest in terms of total cost (TC), it is appropriate to see that most companies choose and use it.

It's much cheaper to put in expensive Nvidia equipment than to design and build a TPU yourself like Google.

Nvidia equipment is the cheapest solution for high-speed and high-capacity processing of current generation AI models.
.

Broadcom's growth is centered around big tech companies that are economies of scale, because the total cost is lower when Nvidia equipment is replaced with its own equipment. The cost of the immediate few years could be more than Nvidia equipment, but it seems to have been determined by the purpose of optimizing each company's workload and de-dependence on Nvidia.

This means that it is difficult to design AI semiconductors through Broadcom for all companies. As the need for custom AI semiconductors grows, costs will gradually decrease, but before that, it is expected that AI semiconductors that are standard models that can smoothly operate the current generation of AI models will emerge.

I'm sure some of you remember that saying, "Little-scale production of many varieties," but how is it over time? It's nonsense. Same goes for semiconductors. With standardized semiconductors to drive AI models, the market is expected to focus on finding rules.

At this point, the actual AI semiconductor standard model is Nvidia, but it would be appropriate to judge that the huge margin rate is making other market players participate.
.

Broadcom's growth makes us reflect on many things. Broadcom did not build an HBM controller well from the beginning. Early HBMs were not very fast, either. However, it seems that Broadcom was able to rise to its current position by understanding the development principles of the IT semiconductor market such as "speed", "capacity", and "cost".

In Korea, it is very rare for companies to make steady investments while looking at long-term investments like this. I remember some of the companies that are left in my memory are Hyundai Motor (the DNA is different because Hyundai Motor is now based on technology internalization), Samsung Electronics Foundry (the current performance is not good, but I think it is great in that it consistently invests), and SK Hynix (I think there must have been a culture that steady investment is beneficial based on HBM). Other Korean companies seem to focus on short-term financial performance rather than steady investment.
.

And watching Broadcom's growth these days, there are analyst reports and YouTube videos saying, "ASIC is the future!" but it doesn't seem appropriate.

"ASIC is the future" is the same as "Engineering is the future."
It's just one word.

Broadcom's growth is due to its strategy in the ASIC market, its long-term strong drive, and its positional advantages as a California-based company in the United States, not an example of ASIC's future.

Thank you.