In the fast-paced world of finance, the buzz around AI and data-center capital expenditures has taken center stage over the past year. Lately, though, there’s been a notable undercurrent of concern. Are these investments justified, or is this a bubble waiting to burst? This conversation gained momentum when DeepSeek challenged prevailing cost assumptions, and now there’s a buzz about a TD note suggesting Microsoft might be curbing its spending.
I’ve outlined the details of the note for you.
According to our insider sources, Microsoft has made some significant changes: they’ve canceled several leases in the U.S., affecting ‘a few hundred megawatts’ with at least two private data center operators. They’ve also pulled back on converting these contracts into long-term leases and have redirected a significant portion of their international budget to their operations in the States. This realignment suggests that Microsoft might be anticipating an oversupply.
Based on our findings, Microsoft has terminated certain data center leases with two private operators in various U.S. regions, amounting to hundreds of megawatts. Interestingly, in some cases, delays in facilities or power supply are cited as reasons for these terminations. Previously, Meta used a similar strategy to cancel several U.S. data center leases, notably after slashing a $48 billion capex plan tied to its metaverse project.
Furthermore, we’ve observed that Microsoft has hesitated on converting preliminary agreements (known as 500s) into finalized leases. The reason for this delay remains unclear; it could be temporary or an outright halt. Typically, a 500 outlines lease terms but isn’t the lease itself. The conversion rate has historically been almost 100%, prompting operators to start building. Additionally, Microsoft is reallocating much of its projected international expenditure to the U.S., hinting at a slowdown in their international leasing pace.
So, what’s driving these shifts?
Our sources haven’t unearthed the full picture yet, but a plausible theory is that Microsoft might be dealing with oversupply issues. For example, our discussions revealed that Microsoft:
1. Withdrawn from several 100MW deals across various markets, during different negotiation stages.
2. Allowed over 1GW of Letters of Intent on larger sites to lapse.
3. Canceled contracts on at least five land plots in key markets.
This shift away from prior expansion plans—especially land purchases supporting future growth—suggests Microsoft might have lost a major demand signal fueled by projects like OpenAI. Recent reports seem to confirm this assessment.
To illustrate, Microsoft was quite aggressive in securing capacity throughout 2023 and early 2024, anticipating growth in demand tied to OpenAI tasks. Yet, decisions like halting construction of a Wisconsin data center, originally intended for OpenAI, imply they may now have surplus capacity, particularly where such capacity isn’t easily repurposed for other cloud needs.
This is our current interpretation though, and fresh intel could shape our understanding further.
For some time, there’s been apprehension that AI investment might ease off, feeding into fears that the ‘bubble’ is deflating. However, juxtapose this with remarks from CEO Satya Nadella on a podcast this week. He confidently stated that hyperscalers, like Azure and others, stand to gain from this AI evolution, due to the sheer computational demand these AI processes require. AI agents could drastically boost compute needs, creating unprecedented scale and demand for infrastructure, potentially benefiting major players in the hyperscale space.
Meanwhile, whispers of a split between Microsoft and OpenAI are swirling, adding another layer of complexity to the narrative. While it’s challenging to separate truth from normal business fluctuation, the chatter is enough to raise eyebrows early this week.
