Microsoft is reportedly cancelling some U.S. data center leases and slowing down the execution of others. Altogether, these projects would have represented a few hundred megawatts in total power if completed. The news has led many to think the GenAI hardware boom is finally peaking. While that’s possible, I think honing in on which pieces of the hardware stack this involves and why this news is coming today is quite important… so let’s do it.

On the last call, the company had already spoken about shifting CapEx from long-lived assets like data centers to short-lived assets like chips. They've been "short on power & space" over the last few years & have aggressively tried to address this. Here's what they said about resolving this supply bottleneck on the last call:

"We have been short on power and space. As you see those investments land that we've made over the past 3 years, we get closer to that balance by the end of this year."

CFO Amy Hood last earnings

The end of Microsoft’s fiscal year is in June.

"We expect quarterly spend in Q3 & Q4 to remain at similar levels as our Q2 spend. In FY '26, we expect to continue investing against strong demand signals, including customer contracted backlog... However, the growth rate will be lower than FY '25 and the mix of spend will begin to shift back to short-lived assets."

CFO Amy Hood last earnings

So what does this mean? At a minimum, it means that Microsoft’s capacity "balance" will be coming sooner & data center growth needs will be lower than expected. It hints at the shift back to short-lived assets (like GPUs) coming more quickly. If that’s the case, this will be a negative for data center builders, but not a material negative for the GPU vendors like Nvidia. It also won’t be a negative for those renting out GPU capacity like Azure. In a best-case scenario, this could even unlock more budget for chip vendors by lowering spend needs elsewhere in the infrastructure stack.

If lower data center needs lead to lower GPU and compute capacity needs, that would clearly hurt GPU vendors and resellers. But? The relationship between these two assets is not as clear as one may think. There are confounding variables to consider. Microsoft’s move could also be a matter of data centers becoming increasingly dense & efficient. This has been a consistent theme, with all builders fixated on optimal productivity. If that’s the source, more GPUs can be packed into the same amount of real estate to offset the lower data center needs.

This could be a second-order reaction to DeepSeek making models more efficient to train, thus driving down space and GPU needs. I don’t think that’s the reason for the news. I strongly believe DeepSeek just means the same level of GPU demand leads to building way cooler & more powerful apps. Not less demand for that asset. As soon as one player decides to keep spending the same on GPUs to build the best possible end product… everyone will be forced to follow suit. Game theory. That’s intuitive. Everyone in the space is racing to make GPU deployment within data centers higher utilization & more efficient. There could be a DeepSeek component to causation here.

This news is related to lower data center footprint needs for OpenAI workloads specifically. Recall that Microsoft and OpenAI recently amended their investment agreement as part of OpenAI’s Stargate project. Azure lost cloud infrastructure exclusivity as part of the amended deal and freed OpenAI to explore other data center partnerships. It is VERY easy to see how that change could have led to subsequent cancellations like this. I think that’s highly probable.

While a faster-than-expected shift back to short-lived CapEx and more efficient data centers could both be contributing, I do think this is largely related to Stargate. The timing of this announcement and that contract amendment is likely not a coincidence. The first sign of Microsoft pausing data center plans in Wisconsin came days before Stargate was announced. If I’m right, this would directly lower overall demand for Azure specifically, but not anyone else. It’s not a demand risk for any of the other hyperscalers, as none of them are dealing with diminishing demand from a massive client. It’s also not a risk for GPU demand; the sources of that demand would shift to whoever OpenAI is working with instead of shrink.

Some are citing Meta pulling back from its own data center plans as a reason for why this isn’t Azure-specific. But? Meta’s move was in response to a rare bee species discovered on site that slowed things down. Meta is also building a massive 2-gigawatt facility for 2025 as the social media titan, Alphabet and Amazon all gear up for massive CapEx budgets. No changes to that as of today. Just like everyone else, I am speculating about the true cause of this headline. It will be very interesting to hear what Nvidia has to say on its earnings call Wednesday.

PS — for enterprise & consumer-facing software, any cost deflation related to any piece of the infrastructure budget (stemming from better supply dynamics) is good. Cheaper to build.