According to a document seen by the media, Microsoft plans to have 1.8 million artificial intelligence chips by the end of 2024, which means Microsoft plans to double the number of graphics processing units (GPUs) it owns this year.
In addition, it is reported that Microsoft may purchase and use several different GPUs to handle different tasks.
The collaboration between Microsoft and OpenAI puts them at the forefront of the wave of generative artificial intelligence, but maintaining a cutting-edge position is expensive. Therefore, how to make generative artificial intelligence faster, better, and cheaper has become Microsoft's goal, which largely relies on the company's GPU chips, which mainly come from Nvidia.
According to analysts from DA Davidson, Microsoft spent $4.5 billion on Nvidia chips last year. A Microsoft executive said that this number roughly corresponds to Microsoft's actual expenses.
$100 billion plan
Meanwhile, according to two insiders, Microsoft is expected to spend approximately $100 billion on GPUs and data centers from the current fiscal year to the 2027 fiscal year.
The huge sum of $100 billion is reminiscent of an earlier report. At the end of March, it was reported that Microsoft and OpenAI were planning a massive data center project to build an AI supercomputer called Stargate, with an estimated cost of over $100 billion, which is 100 times the current cost of the largest data center.
Considering that Microsoft will announce its performance in the coming days and that listed companies must comply with the standard silent period before announcing their earnings, Microsoft has not responded to this news.
Nvidia's H100 is currently one of the most powerful GPUs on the market, estimated to cost around $30000. In addition, Nvidia also recently launched a Rocket product, called GPU Blackwell, which is 7 to 30 times faster than the existing H100 chip, while reducing power consumption by 25 times.
As artificial intelligence flourishes, other tech giants are also building massive GPU inventories. Earlier this year, Meta CEO Mark Zuckerberg stated that the social media company will purchase approximately 350000 Nvidia H100 GPUs in 2024, which means the company will have approximately 600000 such chips by the end of this year.
However, while competing with technology giants for high-performance Nvidia chips, Microsoft is also trying to reduce its reliance on Nvidia, and is working internally to design its own artificial intelligence chips.
However, some Microsoft insiders hold a skeptical attitude, stating that Microsoft has been lagging behind Nvidia for many years, and the current technological progress is so fast that even if Microsoft successfully develops its own chips, it cannot be guaranteed to be the most advanced.