Just 2.2 million chips: Microsoft’s AI fleet below expectations
A Guardian investigation finds Microsoft operates far fewer AI chips than the industry assumes — despite $280 billion in spending.
Illustration · AI-generated (AI IN LIFE)
At a glance
- Around 2.2 million AI chips in operation at Microsoft (internal documents, Guardian)
- Investment volume: roughly $280 billion
- Fewer than 500,000 Blackwell chips — about 1 million expected
- Sustainability reports suggest only ~1.2 gigawatts of AI capacity in 2024
- Announced: 5 GW build-out in two years; 10 GW would require ~6.4M chips
What does the investigation show? According to internal documents reviewed by the Guardian, Microsoft currently operates around 2.2 million AI chips — far fewer than experts expected given roughly $280 billion in investment. One Nvidia analyst comments anonymously: “They’re low to me. They’re less than I expected Microsoft would have.”
Where is the gap? Of Nvidia’s latest Blackwell generation, Microsoft reportedly owns fewer than 500,000 units — about one million had been expected. The internal target of 1.8 million chips for 2024 was also exceeded only with delay. Researcher Shaolei Ren of UC Riverside points to Microsoft’s audited sustainability reports, which suggest only about 1.2 gigawatts of AI capacity for 2024: “The sustainability reports are audited by a third party. They have more credibility than announcements.”
What does Microsoft say? CEO Satya Nadella has named the problem himself — the constraint is not chips but power hookups: “You may actually have a bunch of chips sitting in inventory that I can’t plug in.” Flagship projects like the Fairwater data centres in Wisconsin and Georgia are only partially operational so far.
Why does it matter? Microsoft has announced about 5 gigawatts of additional data-centre capacity within two years; a theoretical total of up to 10 gigawatts would require roughly 6.4 million chips. The gap between announced and actually connected capacity is becoming the AI boom’s key metric.
What does it mean for the market? If even the biggest AI spender cannot get its chips online, the industry’s bottleneck visibly shifts: from chip procurement to power and grid infrastructure. Capacity claims by cloud providers are likely to face harder questions.
FAQ
Did Microsoft buy too few chips?
Not necessarily — per CEO Satya Nadella, some chips sit unused in inventory because there is not enough power capacity to connect them.
Where do the numbers come from?
From internal documents reviewed by the Guardian, supplemented by audited sustainability reports and analyst estimates.
Why does the gap matter?
Because it shows the AI boom’s bottleneck is shifting to power and grid connections — no longer primarily chip supply.
