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 called into question, with internal documents revealing a discrepancy between its public statements and actual chip deployment. This issue is not just about numbers; it's about the very foundation of Microsoft's AI strategy and its ability to compete in the global AI arms race.
The discrepancy highlights a fundamental problem: the supply chain for AI chips, dominated by Nvidia, is a closely guarded secret. Nvidia does not disclose chip sales or clients, and tech giants like Microsoft do not reveal their chip holdings. This lack of transparency makes it difficult to assess the true state of AI development and competition. Personally, I find this opacity particularly intriguing, as it raises questions about the power dynamics and information asymmetries within the industry.
Microsoft's public claims about its AI infrastructure expansion are impressive. Since 2022, the company has invested over $280 billion in AI-focused datacentres, aiming to double its global footprint by 2027. However, the actual number of AI chips in operation is far lower than expected. This discrepancy suggests that Microsoft's datacentres may not be fully operational or may lack the necessary chips. It also raises questions about the company's ability to execute its ambitious plans.
One possible explanation is Microsoft's partnership with OpenAI, which may account for some of its datacentre deployments. However, the exact terms of this partnership are not public, leaving room for speculation. Additionally, some of Microsoft's major projects, like the Fairwater datacentres in Wisconsin and Georgia, appear to be far from operational, further complicating the situation.
The issue of chip availability is not just about numbers; it's about the practical challenges of building and operating AI datacentres. Microsoft's CEO, Satya Nadella, has highlighted the difficulty of securing electrical power and building datacentres close to power sources. This 'power vacuum' problem can lead to a situation where chips are sitting in inventory, unable to be used due to a lack of infrastructure.
The calculation of chip numbers from a company's 'AI capacity' in terms of power is a complex task. Dividing power usage by the power consumption of an AI chip, such as the H100, provides a broad approximation. However, this does not account for factors like cooling systems and other equipment, which also use electricity. Additionally, not all chips in a datacentre are AI chips, and server racks contain a mix of AI and non-AI chips.
In conclusion, Microsoft's AI plans are being held back by a shortage of chips, with internal documents revealing a discrepancy between its public statements and actual chip deployment. This issue highlights the challenges of building and operating AI datacentres, and the need for greater transparency in the supply chain. As the AI arms race intensifies, the ability to secure and deploy chips will be a critical factor in determining who wins the race. Personally, I believe that this shortage will force companies to reevaluate their strategies and potentially seek alternative solutions, such as partnerships or joint ventures, to overcome this critical bottleneck.
