March 12, 2025 • 25 mins
In the first of this week's two-part series, Ed Zitron walks you through how Microsoft pulling back from over a gigawatt of compute capacity is a dark omen for the so-called AI Revolution.
---
LINKS: https://www.tinyurl.com/betterofflinelinks
Newsletter: https://www.wheresyoured.at/
Reddit: https://www.reddit.com/r/BetterOffline/
Discord: chat.wheresyoured.at
Ed's Socials:
https://www.instagram.com/edzitron
https://bsky.app/profile/edzitron.com
https://www.threads.net/@edzitron
See omnystudio.com/listener for privacy information.
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Speaker 1 (00:02):
As media.
Speaker 2 (00:04):
Hello you, it's better offline. I'm your host ed Zeitron. Now,
I've talked about the pay all horses of the AI
apocalypse in the past, and these are the events that
signify that the general VAI era is coming to an end.
(00:26):
And the subject to this episode and its second part, well,
I believe represents the biggest paylest horses of the entire flock.
So on February twenty first analyst TD Cohen revealed that
Microsoft had canceled leases and I quote totaling a couple
hundred megawatts with at least two private data center operators
across multiple US markets canceled. The report also detailed how
Microsoft pulled back on converting negotiated and signed statements of
(00:50):
qualifications sqq's, which are added was precursors to a data
center lease, so effectively the first step before you really
agreed to something, well the last step, I guess. Although
the analyst, which is part of the TD Bank company,
added it was unclear whether Microsoft might convert these sqqs
in the future. These generally close close to one hundred
percent of the time. According to them, canceling these was
(01:13):
effectively rather unusual. Now TD Cohen also added that Microsoft
was reallocating a considerable portion of its projected international spend
to the US, which suggests to DD Cohen that there
was a material slow down in international leasing for Microsoft,
But one crucial, teeny tiny part of the report was
missed by just about everybody. I'm going to read directly
(01:36):
from the report, and you'll probably be able to tell
from the tone of my voice which the most pertinent
part is. Ah. As we highlighted in our recent takeaway
from PTC, the Pacific Telecommunications Council Conference, we learned via
our channel checks that Microsoft one walked away from multiple
one hundred plus make what deals in multiple markets that
were in early to mid stages and negotiations to let
(01:57):
one gigawatt of loys on larger footoprint sites expire, and
three walked away from at least five land parcels that
it had under contract in multiple Tier one markets. What
TD Cohen is saying is not just that Microsoft canceled
some data centers, but the Microsoft also effectively canceled over
a gigawatt of data center operations on top of the
previously reported multiple one hundred plus what megawat deals. If
(02:21):
we add in the land under contract, which is indeterminate.
Based on what tdcens has said and the deals that
were in flight, the total capacity likely amounts to even
more than a gigawat. For some context, data sent to
Dynamics reported that Microsoft had five gigawats of data center
capacity in April twenty twenty four, saying that Microsoft had
also planned to add one gigawat of capacity by the
October twenty twenty four and another one and a half
(02:42):
gigawats of capacity by the first half of twenty twenty five.
Based on this reporting, one can estimate the Microsoft as
somewhere between six and seven and a half gigawatts of
capacity at this time. As a result, based on td
Cohen's analyst analysis, even Microsoft has thro recombination of canceled leases,
pullbacks and statements of qualifications, cancelations of land parcels, and
deliberate expiration of letters of intent. That's those lois. They've
(03:06):
effectively abandoned data center expansion equivalent to over fourteen percent
of their current capacity. It's completely bloody insane, and this
story just kind of sat there. It kind of not
the market's confidence. But I don't know, I'm a lot
more worried about this than I think people are. And
really I don't like telling people how to feel, but
this kind of worries me. Okay, now I've thrown a
(03:28):
lot of new terminology at you. But before we move on,
let's explain some terms, because they're essential to understanding why
this is all such a big deal. First of all,
letter of intent or LOI in this context is a
statement that an entity intends to lease or buy land
or power from a data center. These can be binding
on non binding. A letter of intent is serious, though,
and walking away from one is not something you do idly.
(03:50):
It's not like not answering an email. Now we'll go
on to SQQS statements of qualifications. These set the terms
and conditions of a lease. While they do not themselves constitutally,
they convert into sign leases, as I mentioned, at an
almost one hundred percent rate according to TD Cohen, and
are generally used as a signal to the landowner to
start construction. Basically, they're the green light before the green lane.
(04:12):
As for Tier one markets, these are markets for hyperscale growth,
helped by favorable conditions like power, land and cabling. From
what I can tell, there's no fixed list of which
cities are Tier one in which aren't, but they include
obvious candidates like London, Singapore, as well as Northern Virginia,
which is the largest tube of data centers in the world. Finally,
we're talking power, megawatt and gigawe. This one is really important,
(04:33):
but it's also really confusing. Data center capacity is measured
not by the amount of computations the facility can handle,
but rather by power capacity, and that makes sense because
power capacity is directly linked to the capabilities of the facility,
with more power capacity allowing for more servers or more
power hungry chips, of course, and because chips themselves are
constantly getting faster and more power efficient. Power makes a
(04:55):
little more sense. If you mentioned in terms of computations
per second, you'd likely have a number that fluck you
as hardware as upgraded and decommissioned. When you hear megawat
or gigawatt in this episode, assume that we're talking about
capacity and not power generation unless I say otherwise. Now,
with that out of the way, let's talk more about
this report.
Speaker 1 (05:24):
Now.
Speaker 2 (05:24):
The numbers in the TD Cohen Report, which I'll link
in the episode spreadsheet, of course, heavily suggest that Microsoft,
the biggest purchaser of Nvidio GPUs and, according to Ted
Cohen and I quote, the most active data center LESE
of capacity in twenty twenty three and the first half
of twenty twenty four, does not believe that there's future
growth in generative AI, nor does it have faith in,
(05:44):
nor does it want responsibility for the future of open AI.
Data center buildouts take three to six years to complete,
and the largest hyperscalar facilities can easily cost several billion dollars,
meaning that these moves are extremely forward looking. You don't
just build a data center for the demand GI now,
but for the demand you expect further down the line.
This suggests that Microsoft believes its current infrastructure and its
(06:06):
likely scaled back plans for expansion will be sufficient for
a movement that Sachynadella once called a golden age for systems.
He did that less than a year ago. To quote
TD Cohen again, the magnitude of both potential data center
capacity Microsoft walked away from the decision to pull back
on land acquisition, which supports core long term capacity growth,
in our view, indicates the loss of a major demand
(06:27):
signal that Microsoft was originally responding to, and that we
believe the shift is in their appetite for capacity is
tied to open AI. To explain here, TD Cowen is
effectively saying that Microsoft is responding to a major demand signal,
and said major demand signal is saying you do not
need more data centers. Said demand signal that Microsoft was
responding to, in TD Cohen's words, is its appetite for
(06:48):
capacity to provide servers to open AI. And it seems
that said appetite is waning and Microsoft no longer wants
to build out data centers for America's most swagged out
AI guy. Now, I say that kind of as a joke,
and I do think that he's more swagged out than
Mark Zuckerberg. Mark Zuckerberg's trying too hard. However, Sam Moltman
is more damp than Mark Zuckerberg, so ultimately Zuckerberg wins. Now,
(07:10):
I want to make it clear that Microsoft is effectively
cutting its data center expansion by over a gigawatt of capacity,
if not more, and it's impossible to reconcile these cuts
with the expectation that generative AI will be this massive,
transformative technological phenomenon. I believe that the reason that Microsoft
is coming back is that it does not have the
appetite to provide further data center expansion for open Ai,
(07:31):
and it's having doubts about the future of generative AI
as a whole. If Microsoft believed that there was a
massive opportunity in supporting open AI's further growth, or that
it had massive demand for generative AI services, there'd be
no reason to cancel capacity, let alone cats or such
a significant amount. These moves also suggest that Microsoft is
walking away from building and training further large frontier models
(07:53):
like chat GPT's GPT four point five now and from
supporting doing so for others. Remember, Microsoft has significantly more
insight into the current health and growth of GENERATIVEAI than
any other company. Remember they have full access to all
of open AI's tech, probably the future stuff too, not
that there's much they know. All their research too, Microsoft
(08:13):
knows something we don't. As open AI's largest backer and
infrastructural partner and the owners of the server architecture where
they train them ultra expensive models, not to mention, the
largest shareholder in OpenAI, Microsoft can see exactly what is
or isn't coming down the pike on top of having
a view into both the sales of its own GENERATIVEAI
powered software such as Microsoft three sixty five copiler, and
(08:34):
sales of both model services and cloud compute for other
models run on Microsoft A zero, which is their cloud
platform in plain English, Microsoft, which arguably has more data
than anybody else about the health of the generative AI
industry and its potential for growth, has decided that it
needs to dramatically slow down its expansion.
Speaker 1 (08:51):
Now.
Speaker 2 (08:51):
To be clear, this expansion, I really am hammering this
home a lot, but I need you to understand this
is absolutely necessary for generative AI to continue evolving expanding,
even if it only does so in ways that kind
of do the same thing again and again. Now, before
we move on, I want to make it clear that
I'm not saying that Microsoft has stopped building data centers.
I've said it a few times, but these projects take
(09:11):
years three to six years to complete and are far
far in advance with their planning, and Microsoft does have
a few big projects in the works. One plan three
hundred and twenty four megawatt Microsoft Data center in Atlanta
is expected to cost one zero point eight billion dollars
and as far as I know, this deal is still
in flight, however, and this was cited separately by TD Cohen.
(09:32):
Microsoft has recently paused construction on parts of its three
point three billion dollar data center campus in Mount Pleasant, Wisconsin.
While Microsoft had tried to reassure locals that the first
phase of the project was on course to be complete
on time, its justification of delaying the rest of it
was well not brilliant, and it was to give Microsoft
an opportunity to evaluate and I quote the project's scope
(09:52):
and recent changes in technology, and consider how this might
impact the design of its facilities. Oh, bob, oh, buddy,
that's not good. You don't want it. No. One evaluates
the scope and then goes, oh, actually the scope is great.
I love it. Nor do they think about impacts and
go oh, let's do more. No, no, no, anyway, the
same registered article I'm citing here adds that and I
(10:15):
quote the review process may include the need to negotiate
some building permits, potentially placing another hurdle in the way
of the project. The register did add that Microsoft said
it expected to complete one hyperscaler data center in Mount
Pleasant as originally planned, though its capacity wasn't available. Arguably,
Microsoft would expand its data center infrastructure. Anyway, as more
stuff moves to the cloud and our dependence grows on it,
(10:37):
Microsoft and other providers need to build capacity. The organic
growth is natural and sadly inevitable. However, Microsoft's plans for
data center expansion were far far in excess of that
natural growth, and perhaps we're seeing a pullback from those
stated extravagances into something perhaps a little more reasonable.
Speaker 3 (10:55):
Now.
Speaker 2 (10:55):
I've talked a lot about megawards and gigawats, and you're
not in the data center business, and you should be.
The parties are an absolute laugh. This can all seem
a bit abstract, so let's put it into context. Without context,
it's hard to understand how big and one hundred megawat
data center is. These are some of the biggest. According
to the International Energy Agency, Small data centers can consume
(11:16):
anywhere between one and five megawats. These are, for the
most part, average size facilities, perhaps not for cloud compute giants,
but for other companies. It's kind of part of the course.
One hundred megawats, by comparison, is huge. It's the equivalent
of the annual energy consumption of between three hundred and
fifty thousand, four hundred thousand electric cars, and I know
some sort of pedant it's going to say, it's not
the same thing, shout out a fuck up. Go outside,
(11:38):
stop listen, go outside, go outside now, go do something anyway,
Although there are others that will likely dwarf what we
today considered to be a large facility, META is in
the process of constructing, for example, a ten billion dollar
data center campus in Louisiana with a proposed two gigawat capacity. Still,
whatever way you cut it, and one hundred megawatt facility
is big and it's a big long term investment. Cushman
(12:00):
and Wakefield's twenty twenty four global data center market comparison
gives some chilling context into how significant Microsoft's pullback is.
A gigawatt of data set to capacity is roughly the
entire operational IT load of Tokyo, which has a one
point to eight gigawatt capacity, or London nine hundred and
ninety six megawatts, or the Bay Area, which only has
eight hundred and forty two megawatts. These are actually very large.
(12:23):
It's just that Microsoft got rid of so much more.
And again, the total figure of canceled or abandoned capacity
is likely far higher than a gigawatt. That number only
accounts for the letters of intent that Microsoft allowed to expire.
It doesn't include everything else that the two data sentences
already killed, or the land pass was it abandoned, or
the deals that were in early to mid stages of negotiation.
(12:44):
Imagine walking away from two London's or two Tokyos of
capacity and it not being a massive deal. This is
a huge flipping deal. Microsoft is not simply walking back
some future plans. It's effectively canceling what it loudly insisted
was the future. If you think this sounds hyperbolic, consider this.
London and Tokyo are respectively the biggest data center markets
in Europe and Asia, according to the same Cushman and
(13:05):
Wakefield report. Canceling city's worth of capacity at a time when
artificial intelligence is supposedly revolutionizing everything certainly suggests that artificial
intelligence isn't really revolutionizing anything now. One other detail in
TD Cohen's report really stood out to me. While this
pullback in Microsoft's data center leasing, It's also seen a
commensurate rise in demand from Oracle related to the Stargate project,
(13:29):
a relatively new partnership of up to five hundred billion
dollars stopped saying it's five hundred billion dollars to build
massive new data centers for Ai, specifically for one company
led by soft Bank and of course open Ai with
investment from Oracle and MGX and one hundred billion dollar
investment fund backed by the United Arab Emirates. Open Ai
has committed eighteen to nineteen billion dollars to the Stargate project,
(13:51):
money it doesn't have, meaning that part of the twenty
five to forty billion dollars that they're raising at the
moment will be committed to funding these data centers, and
less as I'll get to later. Open Ai raises more
in debt. Leading the round is SoftBank, which is also
committing eighteen to nineteen billion dollars, as well as creating
a joint venture fund called sb open Ai Japan to
(14:11):
offer open Ai services to the Japanese market, something that
I thought was already happening, as well as spending three
billion dollars annually to use open AI's technology across its
group businesses. According to The Wall Street Journal. In simpler terms,
soft Bank is investing as much as I think it's
going to be like thirty billion dollars in open Ai,
then spending another three billion dollars a year on software
that only loses money and still hallucinates and shows no
(14:34):
side of getting meaningfully better or more reliable. Well a
soft bank actually sees value in open AI's tech, or
whether this purchase dealer is a subsidy by the back
door is open to debate. Given that three billion dollars
is equivalent to open AI's entire revenue from selling premium
access to Chatch GPT in twenty twenty four, which included
some major deals with the likes of Price Waterhouse Coopers,
I'm inclined to believe the latter. Even then, how is
(14:57):
it feasible that SoftBank can continue paying to get the
deal done? Microsoft changed the terms of its exclusive relationship
with open Ai to allow it to work with Oracle
to build out further data centers full of GPUs necessary
to power open AI's big, shitty, unprofitable and unsustainable models.
The open Ai Oracle stargate situation was a direct result,
according to reporting from the Information of open Ai becoming
(15:19):
frustrated with Microsoft and not providing it with service fast enough,
including an allotment of three hundred thousand of nvidious GBT
two hundred chips by the end of twenty twenty five.
For what it's worth, The Wall Street Journal reports that
Microsoft was getting increasingly frustrated with open AI's constant demands
for more compute. The relationship between the two entities had
start to fray, with both sides feeling kind of aggrieved. This,
(15:40):
combined with Microsoft's data center pullback, heavily suggests that Microsoft
is no longer interested in being open AI's infrastructure paypig
long term, at least after all it was if it was,
I mean it funded and support open AI's expansion rather
than doing the literal opposite.
Speaker 3 (15:56):
And you have to.
Speaker 2 (15:56):
Wonder if when that whole non exclusisive thing came along,
whether Microsoft was kind of like, no, no, you couldn't
possibly like the papers already at the pen they've already
got a stamp Ford sad woman's signature. No, don't sign
it. It would be so bad. No, I really don't. I
don't think that Microsoft's too cut up about that. And
(16:29):
here's a question for you. If general, if AI had
so much demand, why is Microsoft canceling data centrare contracts.
Why is Microsoft Oracle's largest customer as at the end
of twenty twenty three, allowing softmank and open Ai to
work with Oracle to build the future rather than Microsoft.
As mentioned previously, TD Cohen specifically noted in its report
that microsoft shifting appetite for capacity was tied to open Ai, which,
(16:53):
as I've said already, heavily suggests that Microsoft is at
best less invested in the future of the company, a
statement confirmed by the information which adds that Microsoft have
been trying to and I quote lessen its reliance and
open ai technology as they increasingly compete in selling AI
products well. At worse, this situation could suggest that Microsoft
is actively trying to dump open ai and it's having
(17:14):
questions about the fundamentals of this industry. Writ large in
very plain terms, Microsoft, despite its excitement around AI and
its dogged insistence that it's the future, has canceled data
center leases of over a gig or what of other
data center infrastructure. Doing so heavily suggests that they do
not intend to expand further, or at least to the
extraordinary levels that they'd initially promised. I know, I'm being
(17:37):
kind of repetitive. I know that I'm saying some of
these things repeatedly, and you might think why. I need
you to understand how significant this is because this did
not get covered enough. The coverage of this was dogshit.
I'm saying it, frankly, everyone missed this detail. I am
just one guy. I do a podcast and a newsletter
(17:58):
and I run a PR firm. Why am I the
person every not every time? But it just drives me
a little insane because this was seeing that how many
reporters actually read this report too. This is what drives
me insane with my work. But I really do enjoy
it was all right. While Microsoft has reiterated that it
intends to spend a ridiculous eighty billion dollars in capital
(18:20):
expenditures on AI in twenty twenty five, per CNBC article,
it's unclear how it intends to do so if it's
pulling back on data center expansion at such a large scale,
and the company has provided no tangible explanation or elaboration
as to how it might do so. While hardware upgrades
could account for some of those CAPEX, it would be
nowhere near the eighty billion dollar figure. Again, hyperscale data
(18:40):
centers aren't cheap. They're massive billion dollar or multi billion
dollar ventures. Microsoft, according to CNBC, also leases data center
capacity through core Weave and other providers, though at that
point the reporter has stopped being curious enough to ask
how much or who those other providers might be. Now, Luckily,
with the power of research, I found that the information
report that Microsoft plants spend about tenion dollars renting core
(19:00):
Weave service between twenty twenty three and twenty thirty, which
was also reported by CNBC and otherwise anyway planned past
tense being the operative word, because last week we learned
that Microsoft plans to scale back its purchase of capacity
from core Weave. Core Weave is, as you'll find out
in a future episode and newsletter probably before that. They're
a very odd company which is closely tied to another
(19:22):
company called core Scientific, which it actually rents service from,
and that company exited Chapter eleven bankruptcy only last year.
Core Scientific's financial statements, by the way, are very confusing.
They're a mess, and it mostly makes its money not
from selling high performance computer services, but from mining bitcoin.
It's a weird relationship, and it's weird to still that
Microsoft is even entangled with a company connected to Core Scientific.
(19:43):
But again, that's a future episode, future newsletter, future panic
attack that I'll give myself as I read s once
all day. Moving on, Microsoft also added in a comment
to CNBC in the same article that it continues to
grow at a record pace to meet customer demand. Okay,
excuse me. What customer demand? What customer demand? What is it?
(20:03):
What is the customer demand? What is going on? Microsoft
said back in April twenty twenty four that AI demand
was exceeding supply even after a seventy nine percent surgeon
capital expenditures, and Cfoamyhood send in their next quarterly earnings
in July twenty twenty four that demand remained higher than
microsoft available capacity. On its most recent January twenty twenty
five earning score, Cfoamyhood once again said, as your growth
(20:27):
included thirteen points from AI services, which grew one hundred
and fifty seven percent year of a year and was
ahead of expectations even as demand continued to be higher
than our available capacity. Riddle me this, batman, Why does
a company that keeps talking about having demand that exceeds
capacity decide to cancel multiple data centers which collectively account
for a significant chunk of its existing capacity. I don't know.
(20:50):
Let's see what Microsoft had to say when asked a
week or two ago. Ahem. Thanks to the significant investments
we've made up until this point, we are well positioned
to beat our current and increasing customer demand. Last year alone,
we added more capacity than any prior year in history.
What we may strategically pay just our infrastructure in some areas,
we will continue to grow strongly in all regions. This
allows us to invest in, allocate resources to grow areas
(21:12):
and for our future. The fuck are you talking about?
Sounds like Microsoft built too much capacity and in fact
as yet to see the customer demand that actually could
reach it. In fact, a couple weeks ago, Microsoft's CEO
Saturnadella said in a podcast interview that one of the
things is that there will be as a result of
data center expansion related to AI overbuild. Why is the
CEO of Microsoft saying that if nothing's changed. Microsoft also
(21:36):
in late January said it had thirteen billion dollars in
annual recurring revenue from AI. And by the way, that's
that's revenue. That's revenue. It's revenue, it's not profit and
On top of that, they don't. It's not like they've
made thirteen billion dollars. They're multiplying like a month's revenue
by twelve. And by the way, AI is also not
a line item on Microsoft's earnings, meaning that all of
(21:57):
this is just related revenue put into a hatamari and
rolled around by satching a Dell or picking up chairs
and shit in the office. Either way, this is a
piss poor amount that works out to about three point
twenty five billion dollars a quarter. These are mediocre numbers.
Their bush league and Microsoft's data center pullback suggests that
they're not going to improve. But wait, wait, or perhaps
(22:17):
Microsoft's pullback has something to do with Stargate open aiy's
big infrastructure project. In fact, I think that might have
something to do with a great deal. Let's take a look.
According to the information, open ai plans to have stargate
handle three quarters of its computing needs by twenty thirty.
Look they'll make it, which heavily suggests that Microsoft canceling
so much capacity is on some level linked to open
(22:39):
AI's future plans and not being part of them. Well,
the information reports that open ai is still forecasting to
spend thirteen billion dollars in twenty twenty five and as
much as twenty eight billion dollars in twenty twenty eight
on Microsoft's cloud compute in addition to whatever capacity it
gets from Stargate or Oracle. One has to wonder how
it intends to do so if it needs capacity that
Microsoft isn't building and doesn't have. And these are two
(23:02):
huge numbers, by the way. For context, thirteen billion dollars
is about ten percent of Microsoft's cloud revenue in the
twenty twenty four fiscal year, though it's unclear whether Microsoft
counts open AI's compute spenders its revenue. Nevertheless, I got
some real concerns about whether open ai is even capable
of expanding further, starting with a fairly obvious one. Open
AI's only source of money SoftBank, the world's worst tech
(23:24):
investor and the only company in the world better incinerating
huge piles of cash, and open Ai, well, they got
some money issues that I'll get into, and funnily enough,
that is going to be the topic of the next episode.
I'm not convinced that Microsoft's pullback is driven by Stargate
largly because I don't think Stargate at least with the
current spending goals and capacity targets is actually viable. The
(23:45):
real motivations, I believe have far more to do with
the fact that Microsoft is recognizing that maybe it got
generative AI and a large part of its future all wrong.
See you in the next episode. Thank you for listening
to Better Offline.
Speaker 3 (24:05):
The editor and composer of the Better Offline theme song
is Matasowski. You can check out more of his music
and audio projects at Matasowski dot com, M A T
T O S O W s ki dot com. You
can email me at easy at Better Offline dot com
or visit Better Offline dot com to find more podcast
links and of course my newsletter. I also really recommend
(24:27):
you go to chat dot Where's youreaed dot at to
visit the discord, and go to our slash Better.
Speaker 2 (24:31):
Offline to check out our reddit. Thank you so much
for listening.
Speaker 1 (24:36):
Better Offline is a production of cool Zone Media. For
more from cool Zone Media, visit our website cool Zonemedia
dot com, or check us out on the iHeartRadio app,
Apple Podcasts, or wherever you get your podcasts.
Speaker 2 (25:00):
SPO
As media.
Speaker 2 (00:04):
Hello you, it's better offline. I'm your host ed Zeitron. Now,
I've talked about the pay all horses of the AI
apocalypse in the past, and these are the events that
signify that the general VAI era is coming to an end.
(00:26):
And the subject to this episode and its second part, well,
I believe represents the biggest paylest horses of the entire flock.
So on February twenty first analyst TD Cohen revealed that
Microsoft had canceled leases and I quote totaling a couple
hundred megawatts with at least two private data center operators
across multiple US markets canceled. The report also detailed how
Microsoft pulled back on converting negotiated and signed statements of
(00:50):
qualifications sqq's, which are added was precursors to a data
center lease, so effectively the first step before you really
agreed to something, well the last step, I guess. Although
the analyst, which is part of the TD Bank company,
added it was unclear whether Microsoft might convert these sqqs
in the future. These generally close close to one hundred
percent of the time. According to them, canceling these was
(01:13):
effectively rather unusual. Now TD Cohen also added that Microsoft
was reallocating a considerable portion of its projected international spend
to the US, which suggests to DD Cohen that there
was a material slow down in international leasing for Microsoft,
But one crucial, teeny tiny part of the report was
missed by just about everybody. I'm going to read directly
(01:36):
from the report, and you'll probably be able to tell
from the tone of my voice which the most pertinent
part is. Ah. As we highlighted in our recent takeaway
from PTC, the Pacific Telecommunications Council Conference, we learned via
our channel checks that Microsoft one walked away from multiple
one hundred plus make what deals in multiple markets that
were in early to mid stages and negotiations to let
(01:57):
one gigawatt of loys on larger footoprint sites expire, and
three walked away from at least five land parcels that
it had under contract in multiple Tier one markets. What
TD Cohen is saying is not just that Microsoft canceled
some data centers, but the Microsoft also effectively canceled over
a gigawatt of data center operations on top of the
previously reported multiple one hundred plus what megawat deals. If
(02:21):
we add in the land under contract, which is indeterminate.
Based on what tdcens has said and the deals that
were in flight, the total capacity likely amounts to even
more than a gigawat. For some context, data sent to
Dynamics reported that Microsoft had five gigawats of data center
capacity in April twenty twenty four, saying that Microsoft had
also planned to add one gigawat of capacity by the
October twenty twenty four and another one and a half
(02:42):
gigawats of capacity by the first half of twenty twenty five.
Based on this reporting, one can estimate the Microsoft as
somewhere between six and seven and a half gigawatts of
capacity at this time. As a result, based on td
Cohen's analyst analysis, even Microsoft has thro recombination of canceled leases,
pullbacks and statements of qualifications, cancelations of land parcels, and
deliberate expiration of letters of intent. That's those lois. They've
(03:06):
effectively abandoned data center expansion equivalent to over fourteen percent
of their current capacity. It's completely bloody insane, and this
story just kind of sat there. It kind of not
the market's confidence. But I don't know, I'm a lot
more worried about this than I think people are. And
really I don't like telling people how to feel, but
this kind of worries me. Okay, now I've thrown a
(03:28):
lot of new terminology at you. But before we move on,
let's explain some terms, because they're essential to understanding why
this is all such a big deal. First of all,
letter of intent or LOI in this context is a
statement that an entity intends to lease or buy land
or power from a data center. These can be binding
on non binding. A letter of intent is serious, though,
and walking away from one is not something you do idly.
(03:50):
It's not like not answering an email. Now we'll go
on to SQQS statements of qualifications. These set the terms
and conditions of a lease. While they do not themselves constitutally,
they convert into sign leases, as I mentioned, at an
almost one hundred percent rate according to TD Cohen, and
are generally used as a signal to the landowner to
start construction. Basically, they're the green light before the green lane.
(04:12):
As for Tier one markets, these are markets for hyperscale growth,
helped by favorable conditions like power, land and cabling. From
what I can tell, there's no fixed list of which
cities are Tier one in which aren't, but they include
obvious candidates like London, Singapore, as well as Northern Virginia,
which is the largest tube of data centers in the world. Finally,
we're talking power, megawatt and gigawe. This one is really important,
(04:33):
but it's also really confusing. Data center capacity is measured
not by the amount of computations the facility can handle,
but rather by power capacity, and that makes sense because
power capacity is directly linked to the capabilities of the facility,
with more power capacity allowing for more servers or more
power hungry chips, of course, and because chips themselves are
constantly getting faster and more power efficient. Power makes a
(04:55):
little more sense. If you mentioned in terms of computations
per second, you'd likely have a number that fluck you
as hardware as upgraded and decommissioned. When you hear megawat
or gigawatt in this episode, assume that we're talking about
capacity and not power generation unless I say otherwise. Now,
with that out of the way, let's talk more about
this report.
Speaker 1 (05:24):
Now.
Speaker 2 (05:24):
The numbers in the TD Cohen Report, which I'll link
in the episode spreadsheet, of course, heavily suggest that Microsoft,
the biggest purchaser of Nvidio GPUs and, according to Ted
Cohen and I quote, the most active data center LESE
of capacity in twenty twenty three and the first half
of twenty twenty four, does not believe that there's future
growth in generative AI, nor does it have faith in,
(05:44):
nor does it want responsibility for the future of open AI.
Data center buildouts take three to six years to complete,
and the largest hyperscalar facilities can easily cost several billion dollars,
meaning that these moves are extremely forward looking. You don't
just build a data center for the demand GI now,
but for the demand you expect further down the line.
This suggests that Microsoft believes its current infrastructure and its
(06:06):
likely scaled back plans for expansion will be sufficient for
a movement that Sachynadella once called a golden age for systems.
He did that less than a year ago. To quote
TD Cohen again, the magnitude of both potential data center
capacity Microsoft walked away from the decision to pull back
on land acquisition, which supports core long term capacity growth,
in our view, indicates the loss of a major demand
(06:27):
signal that Microsoft was originally responding to, and that we
believe the shift is in their appetite for capacity is
tied to open AI. To explain here, TD Cowen is
effectively saying that Microsoft is responding to a major demand signal,
and said major demand signal is saying you do not
need more data centers. Said demand signal that Microsoft was
responding to, in TD Cohen's words, is its appetite for
(06:48):
capacity to provide servers to open AI. And it seems
that said appetite is waning and Microsoft no longer wants
to build out data centers for America's most swagged out
AI guy. Now, I say that kind of as a joke,
and I do think that he's more swagged out than
Mark Zuckerberg. Mark Zuckerberg's trying too hard. However, Sam Moltman
is more damp than Mark Zuckerberg, so ultimately Zuckerberg wins. Now,
(07:10):
I want to make it clear that Microsoft is effectively
cutting its data center expansion by over a gigawatt of capacity,
if not more, and it's impossible to reconcile these cuts
with the expectation that generative AI will be this massive,
transformative technological phenomenon. I believe that the reason that Microsoft
is coming back is that it does not have the
appetite to provide further data center expansion for open Ai,
(07:31):
and it's having doubts about the future of generative AI
as a whole. If Microsoft believed that there was a
massive opportunity in supporting open AI's further growth, or that
it had massive demand for generative AI services, there'd be
no reason to cancel capacity, let alone cats or such
a significant amount. These moves also suggest that Microsoft is
walking away from building and training further large frontier models
(07:53):
like chat GPT's GPT four point five now and from
supporting doing so for others. Remember, Microsoft has significantly more
insight into the current health and growth of GENERATIVEAI than
any other company. Remember they have full access to all
of open AI's tech, probably the future stuff too, not
that there's much they know. All their research too, Microsoft
(08:13):
knows something we don't. As open AI's largest backer and
infrastructural partner and the owners of the server architecture where
they train them ultra expensive models, not to mention, the
largest shareholder in OpenAI, Microsoft can see exactly what is
or isn't coming down the pike on top of having
a view into both the sales of its own GENERATIVEAI
powered software such as Microsoft three sixty five copiler, and
(08:34):
sales of both model services and cloud compute for other
models run on Microsoft A zero, which is their cloud
platform in plain English, Microsoft, which arguably has more data
than anybody else about the health of the generative AI
industry and its potential for growth, has decided that it
needs to dramatically slow down its expansion.
Speaker 1 (08:51):
Now.
Speaker 2 (08:51):
To be clear, this expansion, I really am hammering this
home a lot, but I need you to understand this
is absolutely necessary for generative AI to continue evolving expanding,
even if it only does so in ways that kind
of do the same thing again and again. Now, before
we move on, I want to make it clear that
I'm not saying that Microsoft has stopped building data centers.
I've said it a few times, but these projects take
(09:11):
years three to six years to complete and are far
far in advance with their planning, and Microsoft does have
a few big projects in the works. One plan three
hundred and twenty four megawatt Microsoft Data center in Atlanta
is expected to cost one zero point eight billion dollars
and as far as I know, this deal is still
in flight, however, and this was cited separately by TD Cohen.
(09:32):
Microsoft has recently paused construction on parts of its three
point three billion dollar data center campus in Mount Pleasant, Wisconsin.
While Microsoft had tried to reassure locals that the first
phase of the project was on course to be complete
on time, its justification of delaying the rest of it
was well not brilliant, and it was to give Microsoft
an opportunity to evaluate and I quote the project's scope
(09:52):
and recent changes in technology, and consider how this might
impact the design of its facilities. Oh, bob, oh, buddy,
that's not good. You don't want it. No. One evaluates
the scope and then goes, oh, actually the scope is great.
I love it. Nor do they think about impacts and
go oh, let's do more. No, no, no, anyway, the
same registered article I'm citing here adds that and I
(10:15):
quote the review process may include the need to negotiate
some building permits, potentially placing another hurdle in the way
of the project. The register did add that Microsoft said
it expected to complete one hyperscaler data center in Mount
Pleasant as originally planned, though its capacity wasn't available. Arguably,
Microsoft would expand its data center infrastructure. Anyway, as more
stuff moves to the cloud and our dependence grows on it,
(10:37):
Microsoft and other providers need to build capacity. The organic
growth is natural and sadly inevitable. However, Microsoft's plans for
data center expansion were far far in excess of that
natural growth, and perhaps we're seeing a pullback from those
stated extravagances into something perhaps a little more reasonable.
Speaker 3 (10:55):
Now.
Speaker 2 (10:55):
I've talked a lot about megawards and gigawats, and you're
not in the data center business, and you should be.
The parties are an absolute laugh. This can all seem
a bit abstract, so let's put it into context. Without context,
it's hard to understand how big and one hundred megawat
data center is. These are some of the biggest. According
to the International Energy Agency, Small data centers can consume
(11:16):
anywhere between one and five megawats. These are, for the
most part, average size facilities, perhaps not for cloud compute giants,
but for other companies. It's kind of part of the course.
One hundred megawats, by comparison, is huge. It's the equivalent
of the annual energy consumption of between three hundred and
fifty thousand, four hundred thousand electric cars, and I know
some sort of pedant it's going to say, it's not
the same thing, shout out a fuck up. Go outside,
(11:38):
stop listen, go outside, go outside now, go do something anyway,
Although there are others that will likely dwarf what we
today considered to be a large facility, META is in
the process of constructing, for example, a ten billion dollar
data center campus in Louisiana with a proposed two gigawat capacity. Still,
whatever way you cut it, and one hundred megawatt facility
is big and it's a big long term investment. Cushman
(12:00):
and Wakefield's twenty twenty four global data center market comparison
gives some chilling context into how significant Microsoft's pullback is.
A gigawatt of data set to capacity is roughly the
entire operational IT load of Tokyo, which has a one
point to eight gigawatt capacity, or London nine hundred and
ninety six megawatts, or the Bay Area, which only has
eight hundred and forty two megawatts. These are actually very large.
(12:23):
It's just that Microsoft got rid of so much more.
And again, the total figure of canceled or abandoned capacity
is likely far higher than a gigawatt. That number only
accounts for the letters of intent that Microsoft allowed to expire.
It doesn't include everything else that the two data sentences
already killed, or the land pass was it abandoned, or
the deals that were in early to mid stages of negotiation.
(12:44):
Imagine walking away from two London's or two Tokyos of
capacity and it not being a massive deal. This is
a huge flipping deal. Microsoft is not simply walking back
some future plans. It's effectively canceling what it loudly insisted
was the future. If you think this sounds hyperbolic, consider this.
London and Tokyo are respectively the biggest data center markets
in Europe and Asia, according to the same Cushman and
(13:05):
Wakefield report. Canceling city's worth of capacity at a time when
artificial intelligence is supposedly revolutionizing everything certainly suggests that artificial
intelligence isn't really revolutionizing anything now. One other detail in
TD Cohen's report really stood out to me. While this
pullback in Microsoft's data center leasing, It's also seen a
commensurate rise in demand from Oracle related to the Stargate project,
(13:29):
a relatively new partnership of up to five hundred billion
dollars stopped saying it's five hundred billion dollars to build
massive new data centers for Ai, specifically for one company
led by soft Bank and of course open Ai with
investment from Oracle and MGX and one hundred billion dollar
investment fund backed by the United Arab Emirates. Open Ai
has committed eighteen to nineteen billion dollars to the Stargate project,
(13:51):
money it doesn't have, meaning that part of the twenty
five to forty billion dollars that they're raising at the
moment will be committed to funding these data centers, and
less as I'll get to later. Open Ai raises more
in debt. Leading the round is SoftBank, which is also
committing eighteen to nineteen billion dollars, as well as creating
a joint venture fund called sb open Ai Japan to
(14:11):
offer open Ai services to the Japanese market, something that
I thought was already happening, as well as spending three
billion dollars annually to use open AI's technology across its
group businesses. According to The Wall Street Journal. In simpler terms,
soft Bank is investing as much as I think it's
going to be like thirty billion dollars in open Ai,
then spending another three billion dollars a year on software
that only loses money and still hallucinates and shows no
(14:34):
side of getting meaningfully better or more reliable. Well a
soft bank actually sees value in open AI's tech, or
whether this purchase dealer is a subsidy by the back
door is open to debate. Given that three billion dollars
is equivalent to open AI's entire revenue from selling premium
access to Chatch GPT in twenty twenty four, which included
some major deals with the likes of Price Waterhouse Coopers,
I'm inclined to believe the latter. Even then, how is
(14:57):
it feasible that SoftBank can continue paying to get the
deal done? Microsoft changed the terms of its exclusive relationship
with open Ai to allow it to work with Oracle
to build out further data centers full of GPUs necessary
to power open AI's big, shitty, unprofitable and unsustainable models.
The open Ai Oracle stargate situation was a direct result,
according to reporting from the Information of open Ai becoming
(15:19):
frustrated with Microsoft and not providing it with service fast enough,
including an allotment of three hundred thousand of nvidious GBT
two hundred chips by the end of twenty twenty five.
For what it's worth, The Wall Street Journal reports that
Microsoft was getting increasingly frustrated with open AI's constant demands
for more compute. The relationship between the two entities had
start to fray, with both sides feeling kind of aggrieved. This,
(15:40):
combined with Microsoft's data center pullback, heavily suggests that Microsoft
is no longer interested in being open AI's infrastructure paypig
long term, at least after all it was if it was,
I mean it funded and support open AI's expansion rather
than doing the literal opposite.
Speaker 3 (15:56):
And you have to.
Speaker 2 (15:56):
Wonder if when that whole non exclusisive thing came along,
whether Microsoft was kind of like, no, no, you couldn't
possibly like the papers already at the pen they've already
got a stamp Ford sad woman's signature. No, don't sign
it. It would be so bad. No, I really don't. I
don't think that Microsoft's too cut up about that. And
(16:29):
here's a question for you. If general, if AI had
so much demand, why is Microsoft canceling data centrare contracts.
Why is Microsoft Oracle's largest customer as at the end
of twenty twenty three, allowing softmank and open Ai to
work with Oracle to build the future rather than Microsoft.
As mentioned previously, TD Cohen specifically noted in its report
that microsoft shifting appetite for capacity was tied to open Ai, which,
(16:53):
as I've said already, heavily suggests that Microsoft is at
best less invested in the future of the company, a
statement confirmed by the information which adds that Microsoft have
been trying to and I quote lessen its reliance and
open ai technology as they increasingly compete in selling AI
products well. At worse, this situation could suggest that Microsoft
is actively trying to dump open ai and it's having
(17:14):
questions about the fundamentals of this industry. Writ large in
very plain terms, Microsoft, despite its excitement around AI and
its dogged insistence that it's the future, has canceled data
center leases of over a gig or what of other
data center infrastructure. Doing so heavily suggests that they do
not intend to expand further, or at least to the
extraordinary levels that they'd initially promised. I know, I'm being
(17:37):
kind of repetitive. I know that I'm saying some of
these things repeatedly, and you might think why. I need
you to understand how significant this is because this did
not get covered enough. The coverage of this was dogshit.
I'm saying it, frankly, everyone missed this detail. I am
just one guy. I do a podcast and a newsletter
(17:58):
and I run a PR firm. Why am I the
person every not every time? But it just drives me
a little insane because this was seeing that how many
reporters actually read this report too. This is what drives
me insane with my work. But I really do enjoy
it was all right. While Microsoft has reiterated that it
intends to spend a ridiculous eighty billion dollars in capital
(18:20):
expenditures on AI in twenty twenty five, per CNBC article,
it's unclear how it intends to do so if it's
pulling back on data center expansion at such a large scale,
and the company has provided no tangible explanation or elaboration
as to how it might do so. While hardware upgrades
could account for some of those CAPEX, it would be
nowhere near the eighty billion dollar figure. Again, hyperscale data
(18:40):
centers aren't cheap. They're massive billion dollar or multi billion
dollar ventures. Microsoft, according to CNBC, also leases data center
capacity through core Weave and other providers, though at that
point the reporter has stopped being curious enough to ask
how much or who those other providers might be. Now, Luckily,
with the power of research, I found that the information
report that Microsoft plants spend about tenion dollars renting core
(19:00):
Weave service between twenty twenty three and twenty thirty, which
was also reported by CNBC and otherwise anyway planned past
tense being the operative word, because last week we learned
that Microsoft plans to scale back its purchase of capacity
from core Weave. Core Weave is, as you'll find out
in a future episode and newsletter probably before that. They're
a very odd company which is closely tied to another
(19:22):
company called core Scientific, which it actually rents service from,
and that company exited Chapter eleven bankruptcy only last year.
Core Scientific's financial statements, by the way, are very confusing.
They're a mess, and it mostly makes its money not
from selling high performance computer services, but from mining bitcoin.
It's a weird relationship, and it's weird to still that
Microsoft is even entangled with a company connected to Core Scientific.
(19:43):
But again, that's a future episode, future newsletter, future panic
attack that I'll give myself as I read s once
all day. Moving on, Microsoft also added in a comment
to CNBC in the same article that it continues to
grow at a record pace to meet customer demand. Okay,
excuse me. What customer demand? What customer demand? What is it?
(20:03):
What is the customer demand? What is going on? Microsoft
said back in April twenty twenty four that AI demand
was exceeding supply even after a seventy nine percent surgeon
capital expenditures, and Cfoamyhood send in their next quarterly earnings
in July twenty twenty four that demand remained higher than
microsoft available capacity. On its most recent January twenty twenty
five earning score, Cfoamyhood once again said, as your growth
(20:27):
included thirteen points from AI services, which grew one hundred
and fifty seven percent year of a year and was
ahead of expectations even as demand continued to be higher
than our available capacity. Riddle me this, batman, Why does
a company that keeps talking about having demand that exceeds
capacity decide to cancel multiple data centers which collectively account
for a significant chunk of its existing capacity. I don't know.
(20:50):
Let's see what Microsoft had to say when asked a
week or two ago. Ahem. Thanks to the significant investments
we've made up until this point, we are well positioned
to beat our current and increasing customer demand. Last year alone,
we added more capacity than any prior year in history.
What we may strategically pay just our infrastructure in some areas,
we will continue to grow strongly in all regions. This
allows us to invest in, allocate resources to grow areas
(21:12):
and for our future. The fuck are you talking about?
Sounds like Microsoft built too much capacity and in fact
as yet to see the customer demand that actually could
reach it. In fact, a couple weeks ago, Microsoft's CEO
Saturnadella said in a podcast interview that one of the
things is that there will be as a result of
data center expansion related to AI overbuild. Why is the
CEO of Microsoft saying that if nothing's changed. Microsoft also
(21:36):
in late January said it had thirteen billion dollars in
annual recurring revenue from AI. And by the way, that's
that's revenue. That's revenue. It's revenue, it's not profit and
On top of that, they don't. It's not like they've
made thirteen billion dollars. They're multiplying like a month's revenue
by twelve. And by the way, AI is also not
a line item on Microsoft's earnings, meaning that all of
(21:57):
this is just related revenue put into a hatamari and
rolled around by satching a Dell or picking up chairs
and shit in the office. Either way, this is a
piss poor amount that works out to about three point
twenty five billion dollars a quarter. These are mediocre numbers.
Their bush league and Microsoft's data center pullback suggests that
they're not going to improve. But wait, wait, or perhaps
(22:17):
Microsoft's pullback has something to do with Stargate open aiy's
big infrastructure project. In fact, I think that might have
something to do with a great deal. Let's take a look.
According to the information, open ai plans to have stargate
handle three quarters of its computing needs by twenty thirty.
Look they'll make it, which heavily suggests that Microsoft canceling
so much capacity is on some level linked to open
(22:39):
AI's future plans and not being part of them. Well,
the information reports that open ai is still forecasting to
spend thirteen billion dollars in twenty twenty five and as
much as twenty eight billion dollars in twenty twenty eight
on Microsoft's cloud compute in addition to whatever capacity it
gets from Stargate or Oracle. One has to wonder how
it intends to do so if it needs capacity that
Microsoft isn't building and doesn't have. And these are two
(23:02):
huge numbers, by the way. For context, thirteen billion dollars
is about ten percent of Microsoft's cloud revenue in the
twenty twenty four fiscal year, though it's unclear whether Microsoft
counts open AI's compute spenders its revenue. Nevertheless, I got
some real concerns about whether open ai is even capable
of expanding further, starting with a fairly obvious one. Open
AI's only source of money SoftBank, the world's worst tech
(23:24):
investor and the only company in the world better incinerating
huge piles of cash, and open Ai, well, they got
some money issues that I'll get into, and funnily enough,
that is going to be the topic of the next episode.
I'm not convinced that Microsoft's pullback is driven by Stargate
largly because I don't think Stargate at least with the
current spending goals and capacity targets is actually viable. The
(23:45):
real motivations, I believe have far more to do with
the fact that Microsoft is recognizing that maybe it got
generative AI and a large part of its future all wrong.
See you in the next episode. Thank you for listening
to Better Offline.
Speaker 3 (24:05):
The editor and composer of the Better Offline theme song
is Matasowski. You can check out more of his music
and audio projects at Matasowski dot com, M A T
T O S O W s ki dot com. You
can email me at easy at Better Offline dot com
or visit Better Offline dot com to find more podcast
links and of course my newsletter. I also really recommend
(24:27):
you go to chat dot Where's youreaed dot at to
visit the discord, and go to our slash Better.
Speaker 2 (24:31):
Offline to check out our reddit. Thank you so much
for listening.
Speaker 1 (24:36):
Better Offline is a production of cool Zone Media. For
more from cool Zone Media, visit our website cool Zonemedia
dot com, or check us out on the iHeartRadio app,
Apple Podcasts, or wherever you get your podcasts.
Speaker 2 (25:00):
SPO
Host
Ed Zitron
Popular Podcasts
Stuff You Should Know
If you've ever wanted to know about champagne, satanism, the Stonewall Uprising, chaos theory, LSD, El Nino, true crime and Rosa Parks, then look no further. Josh and Chuck have you covered.
Dateline NBC
Current and classic episodes, featuring compelling true-crime mysteries, powerful documentaries and in-depth investigations. Follow now to get the latest episodes of Dateline NBC completely free, or subscribe to Dateline Premium for ad-free listening and exclusive bonus content: DatelinePremium.com
Las Culturistas with Matt Rogers and Bowen Yang
Ding dong! Join your culture consultants, Matt Rogers and Bowen Yang, on an unforgettable journey into the beating heart of CULTURE. Alongside sizzling special guests, they GET INTO the hottest pop-culture moments of the day and the formative cultural experiences that turned them into Culturistas. Produced by the Big Money Players Network and iHeartRadio.