Microsoft and its partner, the AI powerhouse OpenAI, are gearing up for a colossal project in the realm of artificial intelligence with a jaw-dropping price tag of up to $100 billion.
According to reports on March 30th, Microsoft and OpenAI are in the works to launch a special data center project, including the "Stargate," an AI supercomputer set to kick off as early as 2028, aimed at providing OpenAI with enhanced computational power.
The total cost of the project could surpass $115 billion, with Microsoft bearing the majority of the financial investment. This staggering figure amounts to over three times Microsoft's total capital expenditure in 2023 on servers, buildings, and other equipment, and it's a hundredfold more than the construction costs of some of the largest data centers globally today.
The surge in AI interest is propelling numerous corporate giants to establish new AI data centers to support the development of generative AI. Sources reveal that Microsoft and OpenAI's project is segmented into five phases, with the "Stargate" slated to emerge in the final phase, expanding further by 2030, potentially requiring up to 5,000 megawatts of electricity to operate.
Currently, Microsoft is in the third phase of the project, aiming to launch a smaller-scale supercomputer for OpenAI around 2026. The bulk of the project's costs will be allocated in the subsequent two phases, including the acquisition of a significant number of AI chips.
In recent years, the prices of AI chips have soared alongside the AI boom. For instance, NVIDIA, a leading AI chip manufacturer, recently unveiled its first Blackwell chip, the B200, slated for release later this year. NVIDIA CEO Jensen Huang touted the B200 as "the most successful product launch in our history" and revealed in an interview that each B200 chip would be priced between $30,000 to $40,000.
It's reported that Microsoft and OpenAI will collaborate with multiple chip suppliers to provision chips for the project. Analysis suggests that chip costs typically account for half of the total investment in AI-centric data centers under construction. However, in November last year, Microsoft also introduced its in-house AI chip, Azure Maia 100, tailored for cloud training and inference, prioritizing support for Microsoft's own cloud services.
Although the project is still pending approval from both companies and may evolve over time, if the "Stargate" comes to fruition, it would mark the most significant collaboration between Microsoft and OpenAI in the next decade, showcasing their robust confidence in the future of AI and in each other.
In recent years, Microsoft's total investment in OpenAI has reached approximately $13 billion, allowing OpenAI to leverage Microsoft's data centers to support models like ChatGPT and GPT. However, internally at Microsoft, some employees have expressed discontent about such resource allocation. In mid-March, employees from Microsoft's AI platform team complained to the media, stating that the company's AI strategy excessively focused on its partnership with OpenAI, leading to a reduction in Microsoft's budget for original AI research.
On the other hand, for OpenAI, computational power is the most precious resource for developing large models. Previously, OpenAI's founder and CEO, Sam Altman, has repeatedly stated publicly that the lack of server resources has hindered further AI development, emphasizing the need for increased investment in computational power.
Microsoft and OpenAI have yet to provide an official response regarding the project. A Microsoft spokesperson told the media, "We have been planning the next generation of infrastructure innovations to continue driving the forefront of AI capabilities."
