Subscribe
Share
Share
Embed
TCW is a leading global asset management firm with over 50 years of investment experience and a broad range of products across fixed income, equities, emerging markets, and alternative investments. In each episode of TCW Investment Perspectives, professionals from the firm share their insights on global trends and events impacting markets and the investment landscape.
Welcome to the TCW Investment Perspectives Podcast, where
our investment professionals share their insights and
expertise on how to make the most of your portfolio.
Today, we examine how the explosive growth in cloud computing and AI is
contributing to record demand for data centers and the electricity that
powers them, and what that means for asset-backed
securities and fixed income markets generally.
I'm David Vick, a Managing Director on the Fixed Income team at TCW.
I have two guests with me today, each with a somewhat
different perspective on the issues created by TCW.
by the growing demand for computing and power.
Nikhil Chopra is a Senior Vice President and a Credit Analyst on
TCW's Fixed Income team, where he focuses on the power sector.
Dominic Bea is also a Senior Vice President in Fixed
Income, where he specializes in asset-backed securities.
Welcome to both of you.
And Dom, let's get started with you.
We know we've seen a ton of demand for data centers.
Can you talk a little bit about where that demand comes from, how
sustainable we expect it to be, and how have markets responded to that?
Yeah, absolutely.
So demand for data centers is rude simply in increased usage of the internet.
Many years ago, we would have talked about surfing the web, but whether
it's mobile, streaming, e-commerce, internet things, work from home,
or artificial intelligence, it's the everyday demand for the usage of
the internet that's driving a demand for data center storage, right?
Because all that data has to live somewhere.
And so whether that data lives on a server that's in a large
purpose-built multi-megawatt facility, that data needs
to be in a data center, needs to be in a physical place.
There's a few ways to think about the sustainability of this trend.
A lot of folks will point to Silicon Valley, smartest, most well-resourced
companies in the world, and they're investing heavily in data centers, right?
Because they know what the forward curve is going
to look like for the need for this data storage.
And they're buying and their long-term leasing data centers hand over fist.
But I also think the man on the street quiz is helpful too.
Like how sustainable do you think the demand
is for a supercomputer in your pocket?
I think it's pretty strong.
So whether sort of, yes, the smartest people in the room or the man
on the street, I think the demand for usage of the internet and the
consequential need to store data is really strong for years to come.
The market has responded a number of ways to serve this need.
So there's sort of three ways to host your data.
There's on-prem, co-location, and cloud.
On-prem is the kind of the old legacy version that's dying,
but it's your data on your servers in your building.
That can be expensive, not efficient, not reliable.
The second way is co-location.
That's your data on your servers in somebody else's building.
This business had some issues several years ago, but some
migrations to different types of hosting have played out.
And this is now sort of a mid-single-digit
growth type business that has a purpose.
The third variety of data hosting, which
is by far the fastest growing, is cloud.
That's your data on third-party servers in third-party business.
This is data as a service.
And this is growing incredibly strongly.
So revenue for, for instance, Microsoft, one of the large cloud service
providers in their web services division, is up 30% year on year.
And this is both due to the organic growth of
data and conversion primarily from on-prem.
Ultimately, most folks will use a combination
of these types of data hosting mechanisms.
Generally speaking, it makes a co-location in cloud.
But each of them have their own characteristics, and ultimately, and we'll talk
more about later, kind of their own credit qualities that you can lend against.
Great.
So if cloud computing is sort of the data centers or the fastest
growth, to build new data centers, what are the biggest challenges?
What's complicated about building data centers these days?
Yeah.
So building new data centers is really tough.
To step back data centers, it's space, connectivity, an operator, and power.
And power is the short supply right now.
It's the chief bottleneck in delivering more capacity to the market.
For instance, in Northern Virginia, it can take
up to seven years to get new data center power.
And in some secondary markets, up to 11 years.
So there's a real bottleneck.
We have enough space.
We have a fair number of cable operators.
We have enough fiber routes.
What we don't have enough of is delivered power.
So maybe that's a good segue to you, Nikhil.
With power being such a critical and potentially limiting factor,
how have electricity providers stepped up to meet the demand?
And what are some other things that they might think about
doing going forward, assuming the demand remains robust?
So I think for the purpose of this discussion, I
would like to divide the market into two segments.
One is the regulated utilities.
The other one is the unregulated power producers.
So for the regulated utilities, which are on fixed return on
their investment, these are utilities which own generation
plans, transmission distribution lines, and the like.
Again, they own a fixed return on their investment.
So there we have seen a response where they're going to build some
new gas plans, some renewables to help support this load growth.
So if you step away from that side of the market into the unregulated power
producer side of the market, here I think the response is somewhat muted.
So the way this works is for the unregulated
power producers, they rely on power prices.
Sustained power prices provide them signals to build new power plants.
And we haven't seen that sustained part yet, and therefore
they're a little gun shy on putting out new power plants.
So there's other limiting factors there as well.
There is new EPA rules that were just introduced recently.
We'll see if the new administration comes in to change those rules.
There's also long backlog for gas turbines.
The interconnect queues for grid operators are very long.
So all these are limiting factors on the unregulated side of the market as well.
So that's why they're a little bit shy of going
out and building new power plants right now.
However, we have seen them step forward in different ways.
They have stepped into other options, for
example, add new megawatts to an existing site.
So if you have a combined cycle gas plant, 500 megawatts,
they're adding 10, 20, 30 megawatts to that plant at
a very cheap cost, which is a great use of capital.
So we are seeing that.
Also, what we are seeing is nuclear plants,
which are decommissioned, being restarted.
That is something we haven't seen in U.S. history.
So the first one was in Palisades, Michigan.
And now the three-mile island in Pennsylvania, where Microsoft stepped up to
sign a 20-year PPA, which is a power purchase agreement starting in 2028.
So this just goes to show the long nature of this demand.
And then lastly, I think the other aspect we are seeing
is some plant retirement schedules being pushed out.
In specific, we are seeing some coal plant
retirement schedules being pushed out.
And large utilities like Duke and Southern have talked about this.
So I think that's how I would characterize how we are seeing these
options and how electricity providers are responding to this.
And if I can make a comment there, I think what Nikhil is
talking about is power being responsive to data centers.
What we've seen increasingly now is data centers being responsive to power.
So now that there's a lack of power, you see data centers trying to co-locate
generation resources or just move to regions where power is available.
And what you're starting to see as well, we started
this podcast off talking about cloud and AI.
These are two uses that are very hungry for data center capacity.
They have different needs.
And so the cloud computing needs to be closer to its load center.
It needs to be close to the people that are calling and demanding the data.
So it needs what we call low latency.
AI, particularly AI training of models, is very hungry in
terms of capacity, but doesn't have a big latency requirement.
And so that can locate closer to sources of power, cheaper sources of power.
And so you see data centers now are being responsive to this problem of
decreased power and moving closer to the load center is part of that.
Got it.
Interesting.
So it sounds like there's a lot of new things in the
works, whether it's recommissioning an old nuclear power
plant or building new generation capacity overall.
It sounds like that requires a fair bit of capital.
How do we expect that to be handled by markets, the inevitable
capital raise that I assume will have to come down the road?
So overall, we feel the markets should be
conducive to finance these investments.
Ultimately, this depends on power prices on the unregulated
side, which we are going to see on a sustained basis.
On the regulated side, again, these are regulated
utilities with investment-grade credit ratings.
So they have access to the market on the debt
side and obviously on the equity side as well.
So we feel that there should be plenty of access
to the market to support these investments.
On the unregulated side, where we haven't seen new power plant
built just yet, and these are companies which are going to
have below investment-grade ratings, double B or below.
We feel that there as well, they will have access to market
once the price signals are very clear or they're able to sign
up for long power purchase agreements on the back of that.
And for data centers, I mean, the markets have already been responsive
All of this construction, and we're talking to put some numbers on this, right?
In 2021, it was $10 billion per random private data center construction.
Now it's a $30 billion pace
This is a big number, and markets have already been responsive.
And so you can finance a data center in the commercial mortgage market, in
the corporate debt market, but you can also finance in the ABS and CMBS.
We have about $30 billion outstanding between ABS and CMBS, and we expect to
see about $10 billion per year in issuance as more and more of this capacity
that's under construction completes and finds its way
from the bank market to the debt capital markets.
So when we're looking at these investments, whether it's on the power side
or in the data center side, what are factors that we think of as important?
You guys are looking at, what do you look at?
What do you think is important?
And what sort of opportunities would you
potentially see as we move forward from here?
So on the power generation side, I think we consider market specifics.
There is multiple unregulated markets in the US.
We, however, do not participate in every market.
There is supply-demand dynamics and other nuances with every market.
So we are selective in which markets we do participate in.
But we do see a lot of opportunities where we have participated in single asset
power plants, financings, or small portfolio financings and things like that.
We also like some tangential industrial-type businesses
which play onto this theme of higher power demand.
For example, businesses which provide infrastructure
like electrical transformers for the grid.
That's a good business to us.
We also like businesses which provide temporary power solutions.
That's, again, an example of a slightly different
business, which we also do tend to like.
On the data center side?
Yeah, so I think the big thing to understand here is
the thrust is coming from cloud service providers.
And cloud service providers are high-quality underlying tenants, right?
So this is Amazon, Microsoft, Google, Oracle.
These are investment-grade counterparts.
And because there's such a lack of data center capacity, these tenants
are willing to sign really long-term leases to lock up supply.
This creates really fantastic assets to lend against.
So we've got brand-new build that's highly efficient
to super high specifications where the payor is,
investment-grade counterparty, for 7, 12, 15 years.
This creates an ideal asset to finance and particularly to securitize.
And so that's why we've seen so much of the data center capacity come to the
asset-backed market where they can get a higher advance rate by pooling in
one of these master trust structures where they put all their assets in.
And we have a claim against a kind of cross-collateralized
pool of assets as opposed to sort of a commercial mortgage
loan where you're just lending against a single asset.
But this industry that's creating these sorts of assets that
have a long-term lease are really attractive for us to finance.
Just to provide some context to the conversation, looking
over the past decade, demand for power has been roughly flat.
But as we look out over the next few years, it's going to
grow at about 3% annualized on the back of data centers and
some other factors like on-shoring, electrification, etc.
So that is one aspect.
But in specific, from data centers, we expect demand to be about 55 to 80 gigs.
And again, just to put that into context, it is something
like adding 40 million new households in the U.S.
And that just provides some context as to what kind of demand numbers we're
looking at, which we haven't seen in a very long time in this country.
Those are some very large numbers, for sure.
Yeah.
And one thing to kind of bring that back to my market
is you have all this issuance that's coming to the ABS
market because it's the ideal sort of structure for this.
The ABS market is much smaller than the corporate market.
And so you sort of have this technical mismatch behind the amount
of construction and capacity that's being delivered and financing
that's needed versus the market where it's sort of the best home.
And right now, data center markets, credit spreads are very tight.
But over the next two, three, four, five years, this is going to apply a major
supply technical to the market, which is going to be favorable to ABS investors.
And Nikhil, you talked about the obviously growing demands for power.
What are some of the options in terms of how to generate that power?
Does it all have to be existing technology or is there things like solar or
smaller generation capabilities that might make sense in this environment?
We think a lot of the companies, the data center companies, they're
looking at new technologies like small modular nuclear reactors.
This is something they're just starting to explore now.
So we do feel that it will take some time.
It is not a near-term solution.
It may be a medium to a long-term solution.
So I think that's the most promising in terms of the load
that they're facing because this is going to be 24-7 load.
And a lot of the renewable technologies are not entirely suited for that.
So in terms of that, that is probably the
most suited technology as we look forward.
Got it.
And if we're looking at a sort of a long-term solution, what's
the risk of obsolescence or something on the data center side?
Like, can you do a long-term power solution?
Does that still work with data centers and potentially the
risk that they're, you know, no longer viable in a few years?
So we think that the demand for data centers
is very strong and very sustainable.
But of course, obsolescence risk is a big problem, especially when
you're talking about something that's so technology intensive.
And it's probably our most important underwriting
criteria when we look at a data center deal.
To simplify a complex topic, there's one ratio you can look
at with a data center called its power unit efficiency ratio.
And that is the total amount of power the data center consumes
divided by the total amount of power consumed by the servers.
So it's telling you how efficient the data center is in terms
of cooling itself because they throw off a lot of heat.
So to put some specific numbers on that, the average
PUE ratio now is a little over one and a half.
The newest and most dynamic data centers being
delivered have a PUE between, say, 1.15 and 1.2.
Anything around 1.3 is considered state-of-the-art.
You have a lot of data centers out there that are older, less efficient.
They weren't purpose-built, so they don't cool as efficiently.
Consuming power around a 2x PUE ratio.
We are really concerned about the obsolescence of those types of assets.
And so we tend to avoid those sorts of exposures.
Going forward, when you look at an asset that's at a 120 PUE currently,
right, there is a question of, you know, how much better can you get?
And what is the marginal impact to going from a 120 to 1.1?
And it's probably not worth building a new data center to replace
something that, say, at a 120, it is if it's a 2.0x, right?
Because that efficiency eventually feeds
through to the cost to operate the data center.
We're really focused on obsolescence, but it's something that you can
quantify, and you can underwrite, and you can make sure that you're
exposed to assets that have a pretty good long-term outlook.
Great.
Thanks, you both.
Some interesting topics, for sure.
Thanks, everyone listening, for tuning in to another
edition of TCW's Investment Perspectives Podcast.
I want to thank Nikhil and Dom for joining me today and
sharing their insights into navigating the markets.
For more information on TCW strategies, please visit our website at tcw.com.
Thanks for listening, and we'll pick up next time exploring
the trends and opportunities shaping global markets.
Thank you for joining us today on TCW Investment Insights.
For more insights from TCW, please visit tcw.com slash insights.
This material is for general information purposes only and does not constitute
an offer to sell or a solicitation of an offer to buy any security.
TCW, its officers, directors, employees, or clients may have
positions in securities or investments mentioned in this
publication, which positions may change at any time without notice.
While the information and statistical data contained herein are based on sources
believed to be reliable, we do not represent that it is accurate and should
not be relied on as such or be the basis for an investment decision.
The information contained herein may include preliminary
information and or, quote, forward-looking statements, end quote.
Due to numerous factors, actual events may
differ substantially from those presented.
TCW assumes no duty to update any forward-looking
statements or opinions in this document.
Any opinions expressed herein are current only as of
the time made and are subject to change without notice.
Past performance is no guarantee of future results.