Microsoft Dropped Some AI Data Center Leases, TD Cowen Says

Microsoft has canceled some leases for US data center capacity, according to TD Cowen, raising broader concerns over whether it's securing more AI computing capacity than it needs in the long term. From a report: OpenAI's biggest backer has voided leases in the US totaling "a couple of hundred megawatts" of capacity -- the equivalent of roughly two data centers -- canceling agreements with at least a couple of private operators, the US brokerage wrote Friday, citing "channel checks" or inquiries with supply chain providers. TD Cowen said its checks also suggest Microsoft has pulled back on converting so-called statements of qualifications, agreements that usually lead to formal leases.

Microsoft in a statement on Monday reiterated its spending target for the fiscal year ending June, but declined to comment on TD Cowen's note. Exactly why Microsoft may be pulling some leases is unclear. TD Cowen posited in a second report on Monday that OpenAI is shifting workloads from Microsoft to Oracle as part of a relatively new partnership. The tech giant is also among the largest owners and operators of data centers in its own right and is spending billions of dollars on its own capacity. TD Cowen separately suggested that Microsoft may be reallocating some of that in-house investment to the US from abroad.

it is the tariffs, plain and simple.

[williamyf]

That datacenter has to be filled with AMD processos made in taiwan, with nVIDIA cards also madein taiwan, with memory made in corea (samsung and SKHinix).

all that stuff is subject to tariffs. Better to cancel the USoA datacenters, and ramp up datacenters in mexico and canada to serve USoA companies that do not need their workloads/data to remain in USoA soil. After all, these are cloud workloads, they can be moved on a whim, and latencies would not affected too much.

TL;DR: is not lack of demand, is tariffs making it more expensive to acquire the computing power to fill up them datacenters

Quoting from TFA:

"or it may be that Redmond is concerned by the uncertainty that actions by the Trump administration are causing in the global marketplace."

"While we may strategically pace or adjust our infrastructure in some areas, we will continue to grow strongly in all regions."

Simply stated, MS re-alocated investment to the USoA BEFORE trump announced tariffs to semiconductors.

while trump announcing tariffs on "suff" was expected by all, the tariffs on semiconductors specifically were truly something out of left field...

And before someone says no one likes copilot, midjourney, grok, gemini, et al:

The article refers to demand for TRaINING AI, which can mostly is done in datacenters.

meanwhile, things like copilot rely on INFERENCE, which can be done either in the cloud (like microsoft's data centers), the edge (say, a shared AI inference server for the whole organization in their own datacenter, or device-side either on your NPU or GPU.

so. Copilot is not relevant to the article.

Re:it is the tariffs, plain and simple.

[laughingskeptic]

Agree. In general, an uncertain business climate causes corporations to review their options. When an executive committee is tasked with coming up with options, they always do. Microsoft was already investing heavily in the UAE where the cost of electricity is 1/4 the cost in the U.S. My guess is they are going to invest even more heavily in places where electricity is cheaper than here. Why wouldn't they?

I can see why

[CEC-P]

We were trying not to laugh as I absolutely roasted the Copilot demo and sales people that contacted us since we already have a huge o365 license count. Management wanted to look into it. It's not allowed anywhere near customers because it doesn't work so the entire sales department is out. It can take notes in meeting but they're not sure what will happen when the meeting video recording expires or if we delete it. I asked about every single problem I heard about copilot online and they didn't have an answer. So it's experimental garbage that doesn't do anything we need and is a glorified note-taker for now that makes everyone paranoid to say anything in a meeting and they want $30/mo for that. Nope. At least they're finally catching on.

Re:

[blue trane]

What if their actual costs make AI too cheap to meter, due to cheap Wa hydropower and new cheap training methods?

Re:

[gweihir]

Nothing in human history was ever "too cheap to meter". There are people that like to lie though and then there are shortages of meters.

Re:

[HiThere]

Air has been "too cheap to meter" most of my life. There are places where water is "too cheap to meter". (This doesn't mean it doesn't have a cost, it means there's a fixed cost that's included in something lese.) I could come up with others.

That said, expecting AI to be "too cheap to meter" isn't likely short of the AIs taking over management of society (including the setting of goals).

Re:

[blue trane]

If you look at the Sankey energy flow diagram for Washington state, will you too see that a third of electricity generated is rejected despite hydropower, nuclear, and wind able to satisfy the load with their high efficiencies? Why isn't this evidence that electricity is only metered because decoupling laws have explicitly separated retail pricing from energy supply and demand, thus politically administering rates rather than letting markets drive them lower and even negative (like wholesale prices)?