[스토리라디오 시즌3 11편] HBM이 한개도 없는데, 세계에서 제일 빠른 칩 - LP판만 한 칩, 세레브라스 #세레브라스 #HBM #AI반도체

Semiconductor chips are typically the size of a fingernail. However, this chip is the size of an LP record. Cerebras, the company that manufactured a single, solid chip without slicing a wafer. Ranked number one in the world for inference speed based on independent measurements, despite possessing not a single HBM—how was this possible? And why did the stock price of the company that created such an incredible product halve immediately after its IPO? We will dissect the structure from Gene Amdahl's trilogy in 1980 to designs that bypass defects, the fortress known as CUDA, the reversal of cloud computing where showrooms became the main business, and the next test paper awaiting Korean semiconductor companies that rely on HBM for their livelihood. ※ This video is not a buy or sell recommendation for any specific stock; it is content that covers the structure of companies and industries. Mentions of Korean companies are made from an industry positioning perspective. ※ Some performance figures are based on company announcements and are indicated separately within the video. ※ Fixed exchange rate: 1 USD = 1,519 won (based on Fed H.10) Description (EN) A normal AI chip is the size of a fingernail. This one is the size of a vinyl LP. Cerebras refused to dice the wafer — the whole wafer IS the chip. With zero HBM on board, it now holds the fastest independently-measured inference speed in the world. So why did the stock get cut in half right after the biggest US IPO of the year? From Gene Amdahl's 1980 Trilogy to fail-in-place design, the CUDA moat, the showroom that became the main business, and the next exam Korean memory makers will face — a structural deep-dive. Not investment advice. Some figures are company-reported and labeled as such. FX fixed at 1 USD = 1,519 KRW. Chapter 00:00 Cold Open 01:20 Top 1 03:18 Top 2 05:59 Top 3 09:08 Top 4 10:50 Top 5 14:53 Top 6 17:22 Closer