HawkInsight

  • Contact Us
  • App
  • English

Spend more than a hundred billion dollars!Microsoft and OpenAI plan to collaborate to create AI supercomputers

Microsoft and OpenAI executives are developing an ambitious data center project with plans to build multiple supercomputer devices over the next six years.

According to media reports, Microsoft and OpenAI are planning an ambitious data center project that could cost over $100 billion.

AI Supercomputer

The rapid deployment of generative artificial intelligence technology worldwide has led to a surge in demand for AI data centers that can handle more advanced tasks than traditional data centers.

To this end, Microsoft and OpenAI executives have developed a data center project plan, and it is expected that the project will provide impetus for OpenAI's artificial intelligence. OpenAI currently uses Microsoft data centers to support its generative artificial intelligence ChatGPT in exchange for Microsoft's exclusive right to resell OpenAI technology to its customers.

It is understood that the proposed data center project by Microsoft and OpenAI includes multiple supercomputer devices to be built over the next six years, which are expected to be completed in five stages.

微软、OpenAI

According to insiders, there will be an unprecedented supercomputer in these supercomputer devices, which will be temporarily named the "Stargate" internally.

Microsoft and OpenAI are currently in the middle of the third phase of the five phase plan, and Stargate will be the fifth phase of the aforementioned plan. The two companies hope that the Stargate can be launched as early as 2028, but the specific launch time also depends on whether the subsequent large model capabilities of OpenAI can be significantly improved.

It is reported that Microsoft is developing a fourth stage supercomputer smaller than Stargate for OpenAI, which may be launched as early as around 2026.

Regarding the relevant news, Microsoft did not respond positively. A spokesperson for the company stated in a statement, "We have been planning for next-generation infrastructure innovation to continue promoting the development of artificial intelligence capabilities."

The plan was grand and difficult

Although this plan sounds very exciting, the reality is that it really needs to overcome many difficulties before it can be implemented.

Firstly, there is a chip issue. The report states that the fourth and fifth phases of the above-mentioned plan require a large amount of artificial intelligence chips. However, artificial intelligence chips are still in short supply, and their prices are not low.Taking Nvidia, a global chip company, as an example, its latest "Blackwell" B200 artificial intelligence chip will be priced between $30,000 and $40,000. According to reports, the supercomputer proposed by Microsoft and OpenAI is expected to use millions of dedicated server chips.

However, OpenAI and Microsoft do not intend to hang themselves on the tree of Nvidia. According to insiders, the new project will purchase chips from different suppliers to be used together.The chip dilemma is just one of the several challenges that the above plan needs to address. According to people familiar with the project, in order to build a large-scale data center, Microsoft and OpenAI also need to solve the problem of "how to put more GPUs than currently in a single rack" to improve the efficiency and performance of chips. This also means that the two companies need to find new ways to solve the problem of chip overheating.

In addition, electricity is also an issue that both companies need to consider. To far exceed the computing power currently provided by Microsoft to OpenAI from its data centers across the country, Stargate would require at least a few gigawatts (GW) of electricity. To address the energy shortage issue, Microsoft and OpenAI are already considering alternative new energy sources, such as nuclear energy.

According to a job posting released by Microsoft last year, the company is recruiting a "Chief Project Manager for Nuclear Technology" to evaluate how nuclear energy can be used to power data centers hosting artificial intelligence models. There are reports that Microsoft hopes to focus on using microreactors and small modular reactors (with construction and operating costs far lower than large nuclear reactors) to "power the data centers where Microsoft Cloud and Artificial Intelligence are located."

Even if technical difficulties are solved, capital investment is still a big problem. Sources say that Microsoft may provide funding for the aforementioned data center project. It is expected that the cost of building the supercomputer in this project will be hundreds of times higher than the currently operating largest data center, and the initial cost may reach billions of dollars. Even if wealthy like Microsoft, this amount of money is not a small amount, and the expected expenditure of this plan is almost three times that of Microsoft's capital expenditure on servers, buildings, and other equipment last year.

Due to many uncertainties, including the location of the Stargate and whether it will be built in a single data center or adjacent multiple data centers, the overall project plan and related costs may change.

·Original

Disclaimer: The views in this article are from the original author and do not represent the views or position of Hawk Insight. The content of the article is for reference, communication and learning only, and does not constitute investment advice. If it involves copyright issues, please contact us for deletion.