Now that Samsung's crisis theory seems to be the end, it is a story that fits we

Now that Samsung's crisis theory seems to be the end, it is a story that fits well with the end. If we have talked about crisis theory to the fullest, shouldn't we plant seeds of hope.

= Over the past month, there have been so many rumors of a crisis at Samsung Electronics. Everyone I met talked about the crisis at Samsung Electronics from their own perspective, adding to the anger of people who were bitten by the drop in stock prices.

Crisis doesn't come overnight. The causes are fragmented, but the results are comprehensive. Many years of bad fragmentation accumulates decision-making, resulting in a total catastrophe.

We need to look back on the fragmented causes and sort them out one by one to sort out the past. For good results in the future, we need to sow the seeds that will contribute to future success from now on.

So I've covered what those seeds might be. I don't know which of the various possibilities is the answer or if it's in sight.

Experts are confused because they have different judgments about the future in different fields, and sometimes they have mixed feelings. It's the ability of the top decision makers to pick out those things well.

0) First of all, I'll quickly make an HBM and catch up with it. I'll regain my memory competitiveness. If the bottom has drained from the jar, I'll add water and remove the bottom.

1) a gathering of the anti-NVIDIA camp
Against Nvidia, which wants to dominate the entire data center, each company fights through specialization of division of labor. To do this, it is creating communication protocols (UA Link) and infrastructure software (UXL). It should be a major player in its camp.

-> There are some experts' opinions on this. Some say it will be a great alliance, while others say it's a mess. If I'm not sure, I think I should put my foot on it.

2) Development of AI semiconductors
AI semiconductors are narrowing the position of the CPU, which is the historical ruler of computers. All parts run around AI semiconductors, and even the CPU that did everything can be reduced to a minimum role. I think it's a challenge for Samsung Electronics because AI semiconductors only have Nvidia for now, and no one knows what will happen after Nvidia. Even if it fails, you need to know about AI semiconductors in the era of AI semiconductors. I think a lot of resources will accumulate in the process of developing themselves.

-> Samsung Electronics unilaterally ended the Mach project it was working with Naver. I can't hear that it is making AI semiconductors by itself. Some say that it needs to sell HBM to Nvidia, but it cannot develop competitive products for Nvidia. Some say that Samsung Electronics, which ignores such contracts with partners, should give up consciousness.

3) Memory Foundry
In the days when CPU dominated, memory was a business where other companies came and bought it as long as Intel made it to the standard. The competitiveness of memory companies was how to make the same product and how cheap it was. Just like oil was a comedy in the days of internal combustion engine cars. Memory in the age of AI semiconductors can be a part of AI semiconductors. Instead of selling memory designed and made by a memory company, AI semiconductors design both the calculator and the memory and leave the manufacturing to the outside world.

The HBM that gets attached to Nvidia isn't just an HBM that fits the Zedek standard. Nvidia AI semiconductors aren't just GPUs, they're semiconductors that operate as a single chip with a calculator and an HBM. It's not how well you make it, it's how well it sticks.

Memory and memory foundry are different subjects. Memory is made and sold by a memory company. Memory foundry is a format that creates memory that AI semiconductor company wants to make. Just as fabless designs and TSMC creates them, memory can be in an era like that.

4) Standardized AI semiconductors and separate AI memory
AI semiconductors become standardized and memory changes. Currently, AI computation blurs the line between what hardware has to do and what software has to do. AI modelers are unhappy with Nvidia hardware that is too heavy, but there is no alternative, so they just use it. If the standard of what the hardware has to do is established, and the hardware has to be optimized to the right line, it can be easier to compete with Nvidia and other semiconductor companies.

There was the era of graphic card spring and autumn. As Microsoft built software boundaries with Direct 3D, it became clear what hardware should do. Nvidia, which was the first to recognize and optimize this, dominated the graphic card market by beating Vudu and others.  

Memory may not be customized in such an era, but you may not use the current memory DRAM or mobile DRAM as it is. Even with the same mobile low-power semiconductor, you don't have to stick to the same method used by mobile. What it is...

5) Preparing for the Long-Term Artificial Intelligence Future
There may be many pictures, but the super-giant world model that Professor Jung Ji-hoon told me looked great. This is the era of super-giant language models. Who makes a better super-giant language model is the subject of competition.

On the other hand, artificial intelligence cannot be implemented using only digital language data. Humans did not gain intelligence through language but gained intelligence through sight. There is a spatial intelligence system that enables artificial intelligence through visual data. Prof. Feifei, who discovered Professor Jeffrey Hinton through the ImageNet competition, recently created a startup called World Labs and became a unicorn in four months.

It is called a super-giant world model, and in this case, mobile phones, smart cars, smart class, and robots that will collect data and perform artificial intelligence offline are also connected. Since Samsung Electronics is a company that uses it in all areas from semiconductors to final sets, I think it will be able to see opportunities in this area as well.