Microsoft appears to have a significant gap between its ambitious AI expansion announcements and the actual number of advanced chips it has deployed in its datacenters worldwide. According to internal documents reviewed by the Guardian, the company has 2.2 million AI chips installed globally, despite targeting 1.8 million by the end of 2024 and investing $280 billion in datacentre infrastructure since 2022. This discrepancy raises fundamental questions about the pace of the global AI arms race and whether the world's largest technology companies are moving faster than they actually are.

The shortage highlights a critical bottleneck in AI development: the chips themselves. Nearly all advanced AI chips are manufactured by Nvidia, one of the world's two most valuable companies. Nvidia's supply chain remains one of the industry's most closely guarded secrets, with the company rarely disclosing how many chips it sells or to whom. Its major customers, including Microsoft, Amazon, Google, and Oracle, similarly keep their chip inventories confidential. Without this transparency, experts and investors struggle to assess whether AI infrastructure is truly booming or merely expanding at a measured pace.

Why Is Microsoft's Chip Count So Much Lower Than Expected?

The gap between Microsoft's public statements and its actual chip deployment stems from several interconnected challenges. Microsoft's chief executive, Satya Nadella, has publicly stated that the company's biggest constraint is not chip availability but rather electrical power and the physical infrastructure needed to house and cool the equipment. In a podcast interview, Nadella explained the core problem: "If you can't do that, you may actually have a bunch of chips sitting in inventory that I can't plug in. In fact, that is my problem today. It's not a supply issue of chips. It's actually the fact that I don't have warm shells to plug into".

Nadella, has publicly

This statement reveals a paradox at the heart of the AI infrastructure buildout. Even though Nvidia has manufactured and sold enormous quantities of chips, generating $215.9 billion in revenue in February 2026, many of those chips remain in warehouses or incomplete datacenters waiting for the supporting infrastructure to be ready. Building a datacentre is not simply a matter of purchasing equipment; it requires securing land, constructing buildings, installing electrical systems, and ensuring adequate cooling capacity. These physical constraints often take far longer than acquiring the chips themselves.

Shaolei Ren, a professor at the University of California, Riverside, who studies AI infrastructure, noted that Microsoft's sustainability reports, which are audited by third parties, paint a different picture than the company's investor presentations. According to Ren, Microsoft's actual AI capacity in 2024 was likely closer to 1.2 gigawatts, not the 5 gigawatts the company claimed in some announcements. Even using the lower figure, Microsoft should theoretically have roughly 4 million AI chips installed if it truly added 5 gigawatts of capacity over two years.

"According to their own metrics, Microsoft could be correct. But it isn't clear what they mean when they say they have added datacentre capacity. They are giving insufficient context. The sustainability reports are audited by a third party. They have more credibility than announcements," said Shaolei Ren, professor at the University of California, Riverside.

Shaolei Ren, Professor at University of California, Riverside

One major Microsoft project illustrates the gap between announcement and reality. The company's largest US AI development, called Fairwater, consists of datacenters in Wisconsin and Georgia. In April 2026, Nadella announced that the Wisconsin facility "is going live." However, satellite imagery from Epoch AI suggested only a portion of the building was operational, and Microsoft later admitted to a Wisconsin newspaper that Fairwater was not yet online as of May 2026. This pattern is common in large-scale datacentre construction; projects announced as multibillion-dollar, multi-gigawatt investments often take years to reach full operational capacity.

How Are Nvidia's Blackwell Chips Distributed Among Tech Giants?

The discrepancy becomes even more pronounced when examining Microsoft's allocation of Nvidia's newest chip model, the Blackwell. In March 2026, Jensen Huang, Nvidia's chief executive, announced that orders for Blackwell chips from Nvidia's top four customers, widely believed to be Amazon, Oracle, Microsoft, and Google, totaled 3.6 million units. Without a breakdown of this figure, industry analysts expected Microsoft, as one of Nvidia's largest historical customers, to hold close to 1 million Blackwell chips. Instead, internal documents indicate Microsoft has less than half that amount installed.

This gap raises a critical question: where are the remaining chips? Several possibilities exist. Microsoft may have ordered Blackwell chips but not yet installed them due to infrastructure constraints. The company may have prioritized other Nvidia models for its current deployments. Or, given Microsoft's partnership with OpenAI, some of its datacentre capacity may be dedicated to that partnership in ways not reflected in the documents reviewed by the Guardian. The company has not provided clarity on which interpretation is correct.

Steps to Understanding AI Infrastructure Bottlenecks

• Power Constraints: Datacenters require massive amounts of electricity, and securing reliable power sources near suitable locations is often the limiting factor, not chip availability itself.
• Construction Timelines: Building physical datacentre facilities takes significantly longer than manufacturing chips; projects announced as multibillion-dollar investments often require years to reach full capacity.
• Supply Chain Opacity: Nvidia, the dominant chip manufacturer, does not publicly disclose sales volumes or customer allocations, making it difficult for investors and analysts to assess the true pace of AI infrastructure growth.
• Inventory Gaps: Companies may purchase chips well in advance of installation, creating a lag between revenue recognition and actual deployment in operational systems.

The broader implication of Microsoft's chip situation is that the global AI arms race may be progressing more slowly than public announcements suggest. Tech companies have invested hundreds of billions of dollars in AI infrastructure, but the actual deployment of that infrastructure faces real-world constraints that cannot be overcome simply by spending more money. Power grids, construction capacity, and the physical laws of cooling large numbers of processors all impose limits on how quickly the AI buildout can proceed.

An analyst specializing in Nvidia commented on the discrepancy, stating that Microsoft's chip count appeared lower than expected given the company's public statements about its infrastructure expansion. This assessment aligns with Ren's analysis of Microsoft's sustainability reports, suggesting that the company's actual operational capacity lags significantly behind its capital expenditure announcements.

Microsoft has disputed the Guardian's calculations, insisting they were based on incorrect information, but the company has not specified which figures were wrong or provided alternative data. Sources within Microsoft have indicated that the company's total number of AI chips has "barely moved" over the past year, a statement that contradicts the narrative of rapid, continuous expansion presented in quarterly earnings reports and investor presentations.

As the AI industry matures, this gap between announcement and reality may become increasingly important for investors, regulators, and competitors trying to understand the true state of AI infrastructure development. Without greater transparency from Nvidia and its major customers about chip sales, inventory, and deployment timelines, the world will continue to rely on incomplete information to assess whether the AI boom is real or merely a story told by companies with billions of dollars at stake.

" }