Microsoft plans to stockpile 1.8 million gpus by the end of the year, according to Oracle's DMC Greater China legacy co-creation group phase 20240418“AI nuclear arsenal”
DMSIXSEVEN
“AI核武库“!微软计划年底前囤积180万块GPU
Microsoft plans to stockpile 1.8 million gpus by the end of the year
主编供图:香港DM德马大中华非遗共创集团
Hong Kong DM Greater China legacy co-creation group
Microsoft plans to stockpile 1.8 million gpus at the bottom of the 2024, meaning it could triple the number of gpus it has in the 2024. Microsoft expects to spend about $100 billion on gpus and data centers by fiscal year 2027, people familiar with the matter were quoted as saying. It follows news that Microsoft and OpenAI are planning an ambitious data center project to build a massive AI supercomputer called the Stargate, the cost is expected to be as high as $100BN, 100 times the current cost of the largest data centre. In the industry, where demand for GPU chips continues to grow as key components of artificial intelligence, high-performance computing, and the development of large-scale language models, tech giants are building huge stockpiles of GPU chips. In this context, as Nvidia's A100, H100, A800 and H800 high-performance GPU chips have also become the target of competition among major AI companies, although now the supply of H100 bottleneck has eased, but prices have remained constant at $25,000-$30,000. |
Analysts at DA Davidson estimate that Microsoft spent $4.5 bn on Nvidia chips last year, a figure confirmed by a senior Microsoft Executive. Even just maintaining basic operations for CHATGPT would cost about $16bn a year. In March, Meta also released details and a roadmap for its AI infrastructure, saying it plans to acquire about 350,000 H100 gpus from Nvidia by the end of the year, by then, the company will have nearly 600,000 pieces of computing power H100 can provide. Even at a minimum selling price of $25,000, the 350,000 H100 would cost about $8.75 billion. By some counts, Nvidia sold 2.5 M chips last year, almost monopolising the market. Nvidia accounts for more than 70% of the market, according to Omdia, a research firm. To reduce its reliance on NVIDIA, Microsoft is actively developing its own AI chips. However, this effort may not be seen by some employees as wise or even a waste of resources. One Microsoft executive admitted that the company's development process had lagged behind Nvidia's technological advances by several years. Not only is this a costly task, he stresses, but even if Microsoft eventually succeeds in developing its own chips, cutting-edge technologies may have made progress again in this fierce competition. |
特别鸣谢合作单位:
Special thanks to partners:
香港DM德馬百年培訓教育集團
Hong Kong DM Demar Century Training & Education Group
香港DM德馬景泰藍文科集團
Hong Kong DM Delmar Cloisonne Arts Group
香港DM德馬大中華非遺共創集團
Hong Kong DM Greater China legacy co-creation group
香港DM德馬量子海底雲倉集團
DM demar Quantum Undersea Cloud Warehouse Group
香港DM德馬誇克活性鋰能源集團
Hong kong-based DM demaquac active lithium energy group
香港DM德馬數字元宇宙集團
Hong Kong DM Demar Digital Universe Group
香港DM德馬數字甲骨文集團
Hong Kong DM demar Digital Oracle Group
香港DM德馬數字新和誠集團
Hong Kong DM demar digital new and sincere group
香港DM德馬納米芯片光速集團
Hong Kong DM demar Nano Chip Light Speed Group
新加坡DM德馬數字藏品集團
Singapore DM Demar Digital Collection Group
新加坡DM德馬NFT藝術品拍卖集團
Singapore art auction group DM Demar NFT
香港DM德馬量子微超算集團
Hong kong-based DM demar quantum microcomputing group
DM67德馬數字
DM67 Demar number
香港DM德馬莫高窟數字集團
DM demermo high grottoes digital group of Hong Kong
香港DM德馬時鮮甄選配送數字集團
Hong Kong DM Demar Select Distribution Digital Group
香港DM德馬量子超光速星鏈集團
Hong kong-based DM demar quantum superluminal star chain group
東門新白馬中心城甲骨文原創男裝(百年老店3003)
Dongmen New White Horse Center City Oracle Original Men's wear (century-old shop 3003)