AI hype and data-center capex have been main market drivers previously yr however there’s some angst currently about how a lot is being spent and whether or not there’s a actual enterprise case for all of it. That actually kicked off with DeepSeek undercutting value assumptions and now there’s a TD notice doing the rounds suggesting that Microsoft is pulling again on spending.
I’ve printed the textual content of the notice under.
Our channel checks point out that $MSFT has 1) canceled leases within the US, totaling ‘a few hundred MWs’ with a minimum of two personal information middle operators, 2) has pulled again on the conversion of 500’s to leases, and three) has re-allocated a substantial portion of its worldwide spend to the US. When coupled with our prior channel checks, it factors to a possible oversupply place for MSFT.
Our Channel Checks Point out Microsoft Has Cancelled Choose US Knowledge Heart Leases and Has Pulled Again on SQQ To Lease Conversion; Pulls Again on Worldwide Market Enlargement.
Our current channel checks point out that Microsoft has terminated choose leases with a minimum of two personal information middle operators throughout a number of U.S. markets, totaling a few hundred MW. Our checks point out that in some conditions, Microsoft is utilizing facility/energy delays as a justification for the termination. Recall, as we highlighted in our 2022 Takeaways from PTC, this is identical tactic that Meta used to cancel a number of information middle leases within the U.S. after we discovered in our checks that Meta had then canceled a $48B capex program associated to the metaverse (Meta subsequently lower its capex steering by $5.4B two weeks later).
Individually, our channel checks counsel that Microsoft has additionally pulled again on changing negotiated and signed Assertion of {Qualifications} (500’s) (the precursor to an information middle lease) into signed leases. Thus far, it’s at present unclear to us if that is merely a delay in 500-to-lease conversion or whether it is an outright termination of the five hundred with no conversion to leasing anticipated. For context, based mostly on our checks, a 500 units forth the phrases and situations for the lease and doesn’t represent a lease settlement. Nevertheless, the conversion charge of 500’s right into a signed lease is near 100%, with information middle operators utilizing this because the sign to begin information middle building. As well as, our channel checks point out that Microsoft can be re-allocating a substantial portion of its projected worldwide spend to the U.S., which suggests to us a cloth slowdown in worldwide leasing.
Why is that this taking place?
Whereas we’ve got but to get the extent of colour through our channel checks that we want into why that is occurring, our preliminary response is that that is tied to Microsoft probably being in an oversupply place. As we highlighted in our current takeaways from PTC, we discovered through our channel checks that Microsoft:
1. Walked away from a number of 100MW offers in a number of markets that had been in early/mid-stages of negotiations.
2. Let +1GW of LOIs on bigger footprint websites expire.
3. Walked away from a minimum of 5 land parcels that it had beneath contract in a number of Tier I markets.
On the time, we additionally highlighted that the magnitude of each potential information middle capability it walked away from and the choice to tug again on land acquisition (which helps core long-term capability development) in our view signifies the lack of a serious demand sign that Microsoft was initially responding to and that we believed the shift of their urge for food for capability is tied to OpenAI, which current press experiences seem to substantiate.
To that time, contemplate this: Microsoft was probably the most lively lessee of capability in 2023 and 1H24, at which era it was procuring capability relative to a capability forecast that contemplated incremental OpenAI workloads. Nevertheless, as we imagine is indicated by its choice to pause building on a knowledge middle in Wisconsin—which our prior channel checks indicated was to assist OpenAI—there’s capability that it has seemingly procured, notably in areas the place capability just isn’t fungible to cloud, the place the corporate could have extra information middle capability relative to its new forecast.
That is our interpretation of the present state of affairs, nevertheless, our view is topic to alter as we conduct incremental channel checks.”
There’s has lengthy been some angst about AI spend slowing down and indicators the ‘bubble’ is deflating. Nevertheless distinction this with what CEO Satya Nadalla mentioned in a podcast this week.
So I feel there are two locations the place I can say with some confidence. One is the hyperscalers
that do properly, as a result of the basic factor is in case you type of return
to even how Sam and others describe it, if intelligence is log of
compute, whoever can do a number of compute is a giant winner.
So the truth is it is manna from heaven to have these AI workloads
as a result of guess what? They’re extra hungry for extra compute, not only for
coaching, however we now know, for check time. If you consider an AI agent,
it seems the AI agent goes to exponentially enhance compute
utilization since you’re not even sure by only one human invoking a
program. It is one human invoking applications that invoke heaps extra
applications. That is going to create huge, huge demand and scale for
compute infrastructure. So our hyperscale enterprise, Azure enterprise, and different hyperscalers, I feel that’s a giant factor.
There are additionally some indications round a breakup of the Microsoft and OpenAI partnership.
It is robust to inform what’s true and what’s true and what’s common churn right here however there’s sufficient noise round this so as to add to the angst early this week.
