Microsoft's AI ambitions are facing a critical challenge: a shortage of chips. The company's plans to install 1.8 million AI chips by the end of 2024 have been met with a significant hurdle, as internal documents reveal a stark discrepancy. With only 2.2 million chips installed, Microsoft's progress falls short of expectations, raising questions about the true state of its AI infrastructure.
This issue is not merely a numbers game. It highlights a fundamental problem in the AI industry: the closely guarded secrets of chip manufacturers like Nvidia. Nvidia, a powerhouse in the AI chip market, does not disclose its sales figures or client list, making it nearly impossible to gauge the true demand and supply dynamics. This lack of transparency casts a shadow of uncertainty over the entire AI landscape.
Microsoft's situation is further complicated by its ambitious expansion plans. The company has invested a staggering $280 billion in AI infrastructure, aiming to double its global datacentre footprint by 2027. However, estimating the exact number of datacentres built and their operational status is a complex task. Microsoft's annual reports and quarterly earnings provide glimpses into its progress, but the details are often insufficient.
The crux of the matter lies in the power consumption of datacentres. Microsoft claims to have added 5 gigawatts of datacentre capacity in the past two years, a figure that seems modest compared to the 10 gigawatts it could potentially have. This discrepancy suggests that Microsoft's datacentres may not be fully operational or may lack the necessary chips. The company's sustainability reports, audited by third parties, offer a more reliable picture of its AI capacity, but even these reports have limitations.
The Fairwater project in Wisconsin exemplifies the challenges Microsoft faces. Initially envisioned as a multi-gigawatt, multibillion-dollar investment, it has only partially operational after three years. This slow progress is not unique to Microsoft; it reflects the broader struggle of building massive AI datacentres.
The shortage of chips is a critical bottleneck. Microsoft's tie-up with OpenAI may explain some of the discrepancy, but the exact terms of their partnership remain undisclosed. Nvidia's balance sheets show impressive chip sales, but the question remains: where are the chips that Microsoft needs? The company's CEO, Satya Nadella, acknowledged the issue, attributing it to the lack of 'warm shells' to plug in the chips, rather than a supply shortage.
Calculating the number of chips required from a company's AI capacity is a complex task. Microsoft's internal documents indicate a mix of H100, A100, and Blackwell chips, each with different power requirements. A conservative estimate suggests that 10 gigawatts of capacity could support around 6.4 million chips. However, this calculation doesn't account for the energy used by cooling systems and other equipment, further complicating the picture.
In conclusion, Microsoft's AI plans are being hindered by a shortage of chips, a problem exacerbated by the lack of transparency in the chip industry. The company's ambitious expansion goals and the slow progress of its datacentres underscore the challenges of building a robust AI infrastructure. As the AI arms race intensifies, addressing this chip shortage is crucial for Microsoft and the entire tech industry to maintain its competitive edge.
