Despite these changes, Microsoft remains committed to investing $80 billion in infrastructure during its 2025 fiscal year, having added more capacity last year than ever before. The company plans to strategically adjust its infrastructure while continuing to grow in all regions. This strategy allows Microsoft to allocate resources to growth areas and transition its existing compute capacity, including Hopper GPUs, from OpenAI training to its own inferencing workloads or renting them to customers.
Key takeaways:
- Microsoft has walked away from datacenter leases due to the inability of many facilities to handle the power and cooling demands of high-end AI hardware.
- Retrofitting existing datacenters for liquid cooling is seen as a more viable option than new leases, especially as AI workloads shift from training to inferencing.
- Microsoft is reallocating its existing Hopper GPU capacity, previously used for OpenAI training, to its own inferencing workloads or renting it to customers.
- Despite strategic adjustments, Microsoft remains committed to significant infrastructure investments, planning to spend $80 billion during its 2025 fiscal year.