Microsoft and OpenAI are planning a data center project worth over 15 trillion yen, including the AI supercomputer 'Stargate'

Microsoft and OpenAI are working on plans for a $100 billion data center project that would include an artificial intelligence (AI) supercomputer called ' Stargate ' that is scheduled to be operational in 2028, The Information reported.

Microsoft and OpenAI Plot $100 Billion Stargate AI Supercomputer — The Information

Microsoft, OpenAI plan $100 billion data-center project, media report says | Reuters

Report: Microsoft to build 'Stargate' supercomputer with millions of chips for OpenAI - SiliconANGLE

Microsoft & OpenAI consider $100bn, 5GW 'Stargate' AI data center - report - DCD

OpenAI and Microsoft reportedly planning $100 billion datacenter project for an AI supercomputer | Tom's Hardware

The rapid development of generative AI technology is creating a surge in demand for AI-specific data centers that can handle more advanced tasks than traditional data centers. In response to this situation, The Information reported that OpenAI is planning to build an AI-specific data center. According to an anonymous source familiar with the project who provided information to The Information, the AI data center is expected to be 100 times more expensive than the largest existing data centers, and therefore Microsoft is likely to be the one to fund it.

OpenAI has already been using Microsoft's infrastructure to train AI models, and in 2020, Microsoft announced that it would provide supercomputer services on its cloud computing service, Azure, as part of its partnership with OpenAI. This supercomputer used 10,000 GPUs and was the fifth most powerful supercomputer in the world at the time.

Microsoft develops supercomputer with 285,000 CPU cores and 10,000 GPUs in collaboration with OpenAI - GIGAZINE

Later, in January 2023, Microsoft announced that it had invested hundreds of billions of yen in OpenAI and entered into a long-term partnership . In addition, in March 2023, it was revealed that the supercomputer built for Azure above was upgraded and equipped with tens of thousands of A100 chips.

Microsoft spends over tens of billions of yen to build a supercomputer for AI development for OpenAI, the developer of ChatGPT - GIGAZINE

According to reports, Microsoft plans to build some additional AI infrastructure by 2030, with the plan divided into five phases, and the third phase is the time of writing. The fourth phase is planned to build a new supercomputer, which is expected to be around 2026. The fifth phase is planned to build a new AI supercomputer called 'Stargate', which will use millions of chips and is expected to be operational as early as 2028. This Stargate is expected to be the largest of the ones Microsoft and OpenAI are considering building, but it has been pointed out that the supply of AI chips may be a bottleneck.

In recent years, the importance of AI chips has expanded, and major semiconductor manufacturer NVIDIA announced a new GPU architecture for generative AI, 'Blackwell,' and a new AI chip that adopts this, 'B200,' in March 2024. In addition, NVIDIA CEO Jensen Huang told CNBC in March 2024 that the selling price of 'B200' will be $30,000 to $40,000 (approximately 4.5 million to 6 million yen).

NVIDIA announces GPU architecture 'Blackwell' and new GPU 'B200' to realize AI models with trillions of parameters - GIGAZINE

Microsoft also announced its own AI-specialized chip, the Maia 100, in November 2023. The chip will be deployed in Microsoft data centers in 2024 and will be used to enhance Microsoft Copilot and Azure OpenAI Services.

Microsoft announces its own AI-specialized chip 'Maia 100' and Arm processor 'Cobalt 100' - GIGAZINE

In addition, OpenAI is developing the next-generation large-scale language model GPT-5, which is expected to be released in the summer of 2024, so it is unlikely that Stargate will be used to train GPT-5, SiliconANGLE points out.

OpenAI's next-generation large-scale language model 'GPT-5' will be released in the summer of 2024 - GIGAZINE

in Software,   Hardware, Posted by logu_ii