Semiconductor story, about Samsung Electronics and Hynix
Haebom
2
Recently, while having a meal with someone from Samsung Electronics DS, we got to talking about a company called Padu, which has become a hot topic. It was about semiconductor design and production, and I don't have much knowledge, so I didn't understand 100%, but among Padu and the new domestic semiconductor manufacturing and production companies that have been mentioned recently, only one has a proper production line, and the rest only design and produce small quantities of that design as OEM on other production lines.
In the AI chip industry, where prices are at their highest and everyone is lining up to get a number, Nvidia and TSMC occupy a very strong position and have already set a high barrier to entry for new competitors. Nvidia has about 95% of the GPU market, and TSMC has about 90% of the world's high-end chip market. The companies boast high profitability, recording sales profit margins of 74% and 60%, respectively.
The strength of these companies is due to a variety of factors. For example, Nvidia holds thousands of patents on hardware and related software libraries, while TSMC has invested more than eight years in advanced packaging technology, giving it an edge over rivals like Samsung. In addition, the extreme ultraviolet lithography machines needed to produce advanced chips are monopolized by ASML in the Netherlands, and even if ordered now, it usually takes an average of two years to receive the machines.
This is the value of semiconductor-related stocks listed in the United States. Comparing this figure with the market cap of Samsung Electronics' semiconductor division and SK Hynix...
Against this backdrop, it takes a lot of capital and time for new chip manufacturers to enter the market, and even if they overcome technological challenges, it is very difficult to catch up with the on-site advantage enjoyed by Nvidia and TSMC. The semiconductors that domestic companies currently have are positioned as general-purpose and frequently used ones rather than high-performance chips. Of course, demand for these will not decrease immediately. However, in the end, unless they build a good consortium or find a new path within 2-3 years, it seems that it will be difficult to have fun in the semiconductor industry in the future.
Subscribe to 'haebom'
📚 Welcome to Haebom's archives. --- I post articles related to IT 💻, economy 💰, and humanities 🎭. If you are curious about my thoughts, perspectives or interests, please subscribe. Would you like to be notified when new articles are posted? 🔔 Yes, that means subscribe. haebom@kakao.com
엔비디아가 3분기 마이크로소프트와 메타에 각각 15만대의 H100 AI GPU를 판매했다고 한다. 이에 반해 구글 아마존 오라클 텐센트는 각각 5만대. 코어웨이브 4만대, 바이두 3만대, 알리바바 2.5만대. 네이버가 얼마전 연 세종시 '각' 데이터센터는 엔비디아 A100 텐서 코어 GPU 2240개로 구성되었다. 이걸 4로 나누면 서버 560대라는 소리다. (하나의 서버에 칩 4개 꽂는다. SK텔레콤이 연초에 AI 인프라 두배로 늘렸다고 보도자료 뿌렸는데 알아보니 65대 있던 거 65대 늘린 거였다. 그래봐야 130대다.)
