Abstract:
Microsoft is massively expanding its AI infrastructure. As demand for computing power from generative AI services continues to grow, the company expects full-year capital expenditures to reach approximately $175 billion in 2026, with plans to significantly expand data center capacity. However, what has attracted attention at the same time as this huge investment plan is Microsoft's re-evaluation of the service life of data centers and office building assets, and the resulting changes in capital expenditure statistics.
Microsoft previously expected full-year capital expenditures in 2026 to be approximately US$190 billion, but the latest forecast has been reduced to approximately US$175 billion, which on the surface is equivalent to a reduction of US$15 billion. But this does not mean that Microsoft has actually reduced investment in data center construction of a corresponding scale. The main reason for the decline in capital expenditures is that Microsoft has adjusted the accounting treatment of related assets.
Beginning in fiscal year 2027, Microsoft will extend the estimated service life of data centers and office buildings from the original 15 years to 25 years. Company finance chief Amy Hood said the adjustment was based on Microsoft's actual experience operating these assets in the past and its judgment on future usage. Microsoft believes that data centers and office buildings will last longer, so there is no need to continue to calculate their depreciation over a 15-year period.
This adjustment will first change the time distribution of asset depreciation. Microsoft expects the change to have only a small positive impact on operating profit in fiscal 2027, but the impact on the way capital expenditures are counted will be more obvious.
The reason is that Microsoft uses a large number of leasing methods to build data centers. Based on the new 25-year useful life estimate, more future data center leasing arrangements will shift from finance leases to operating leases. The two types of leases are treated differently in Microsoft's financial statements, with finance leases accounting for capital expenditures and operating leases not accounting for capital expenditures in the same way.
Therefore, even if the number of data centers actually built and used by Microsoft and the related investment plans do not decrease accordingly, the capital expenditure figures in the financial statements will still be reduced.
Microsoft has made it clear that apart from the accounting impact of this useful life adjustment, the company's judgment on the actual investment plan in 2026 has not changed. In other words, the adjustment from approximately US$190 billion to approximately US$175 billion does not mean that Microsoft suddenly reduced its US$15 billion in data centers, AI servers or other infrastructure construction.
This is especially important because Microsoft is currently at the peak of its AI infrastructure expansion. The company expects capital expenditures to reach approximately US$50 billion in the first quarter of fiscal 2027, and infrastructure investment throughout 2026 will remain at a very high level.
Microsoft is currently rapidly expanding the scale of its data centers to meet the growing computing needs of Azure cloud services, Copilot and other generative AI products. According to the latest disclosed information, Microsoft plans to expand data center capacity to approximately 38GW by 2032, more than tripling the current scale of approximately 12GW.
Among them, the proportion of AI-specific computing resources will also increase significantly. Currently, only about 2GW of Microsoft's approximately 12GW data center capacity is mainly used for AI chips. According to the current expansion plan, AI-related computing power may account for about one-third of the total capacity of about 38GW by 2032.
This means that Microsoft will still invest huge amounts of money in the next few years to build data centers, purchase GPUs and other AI hardware, and obtain power and land resources for these facilities. The decline in capital expenditure statistics cannot be directly understood as Microsoft is slowing down the construction of AI infrastructure.
In fact, Microsoft has accumulated sizable future lease commitments. As of the time of relevant financial disclosures, the company's outstanding lease commitments totaled approximately US$329.1 billion, the vast majority of which were related to data centers. These leasing arrangements are expected to commence between fiscal year 2027 and fiscal year 2033.
These leases that have not yet begun do not equate to the cash that Microsoft has paid, nor does it mean that all related data centers are currently in operation. Some agreements still need to meet corresponding contractual conditions, and the specific data centers, power scale, and future annual activations corresponding to these leases have not been fully disclosed.
But such a huge lease size still shows how huge Microsoft's demand for data center infrastructure will be in the next few years. Companies not only need to build their own data centers, but also sign long-term lease agreements with third-party data center operators to quickly obtain computing capacity.
This adjustment to extend the service life from 15 years to 25 years is also related to the data center itself becoming more and more like a long-term infrastructure. Traditional data centers may need to quickly update equipment and carry out technical upgrades, but the main building, power facilities, cooling systems and other infrastructure do not necessarily need to be completely scrapped after 15 years.
For Microsoft, extending the expected service life of building assets to 25 years means that it can be more consistent with the actual life cycle of these large infrastructures. At the same time, this will also change the classification of future lease arrangements in the financial statements, affecting the way capital expenditure is presented.
It is important to note that the life cycle of servers, GPUs and other IT hardware is not the same as that of data center buildings. Just because Microsoft's data center buildings can be used for 25 years does not mean that the AI servers in them can also run for 25 years. A large number of GPUs, CPUs and network devices still need to be constantly replaced according to technological development and performance requirements.
This is why a large portion of Microsoft's current capital expenditures are still directed toward IT hardware. As the demand for AI computing grows, companies not only need to build computer rooms, but also need to continuously purchase large amounts of GPUs, CPUs, network equipment and other computing infrastructure.
Microsoft’s capital expenditures in fiscal year 2026 have reached approximately US$145.3 billion, a significant increase from before. About two-thirds of this is for IT hardware. By the first quarter of fiscal 2027, the company expects capital expenditures to further reach approximately $50 billion.
Such a huge investment scale also reflects Microsoft’s judgment on AI computing needs. In recent years, major technology companies, including Microsoft, have invested tens of billions of dollars in building data centers to provide computing power for generative AI services. ChatGPT, Copilot, and various enterprise AI applications consume a lot of GPU and data center resources.
Microsoft still believes that market demand is sufficient to support such a huge infrastructure investment. The company's cloud business continues to grow, and a large number of enterprise customers are purchasing databases, security services, office software and AI-related products.
Therefore, this change in the US$175 billion capital expenditure figure should be understood more as an adjustment in accounting standards rather than a re-judgment of Microsoft's AI infrastructure needs. The company is still building a large number of data centers and will continue to expand in the coming years.
From an investor's perspective, this change in accounting treatment is also worthy of attention. Capital expenditures are often seen as an important indicator of the scale of infrastructure investment by technology companies. If you just compare the two figures of US$190 billion and US$175 billion, it is easy to conclude that Microsoft is cutting back on AI investment.
But this is not the case. Because some data center leases will be converted from finance leases to operating leases, the corresponding expenses will not enter the capital expenditure statistics in the same way, but Microsoft will still need to bear the long-term costs incurred by these lease arrangements. In other words, a reduction in capital expenditure figures does not mean a simultaneous reduction in the actual economic costs of infrastructure construction.
Microsoft’s adjustment thus changes more about “how to record investment” rather than “whether to invest”. In the context of continued growth in AI computing demand, Microsoft still needs to lock in land, power, buildings and computing resources in advance for data center capacity in the next few years.
As data centers gradually transform from traditional IT facilities into long-term assets similar to power plants and communications infrastructure, the 25-year service life assumption may also become an important change in the development of the industry. For Microsoft, this will not only bring it closer to the actual life cycle of large data center buildings, but will also change how future capital expenditures, depreciation and lease costs appear in financial statements.
Therefore, Microsoft's adjustment of its 2026 capital expenditure forecast from approximately US$190 billion to approximately US$175 billion cannot simply be understood as the company cutting its investment in AI infrastructure by US$15 billion. On the contrary, in the face of a 38GW data center expansion plan, hundreds of billions of dollars in future lease commitments, and continued growth in AI computing needs, Microsoft is building an extremely large infrastructure system for many years to come, and the 25-year lifespan adjustment only changes the way this huge investment is presented on financial statements.
Related articles:
Microsoft promotes data center expansion and plans to triple computing power
Comments