Good morning,

As always, we hope you are well. Please see the attached PDF for our latest set of informational materials, as well as this link to some supplementary downloads on our website for those interested in an even deeper dive. Please let us know if you have any questions.

It has been difficult to miss over the past several weeks the growing concern of a massive “AI bubble,” with Wall Street commentators, industry executives, and media personalities alike all suddenly sounding the alarm far and wide. Google's CEO Sundar Pichai, Bill Gates, the Bank of England, and Bank of America have all provided commentary suggesting increased caution is warranted.

Nvidia's blowout earnings report yesterday has seemingly dampened those bubble fears and it may well be a good day for investors all throughout the stock markets given how significant is Nvidia’s influence. As we write this in the pre-market hours, names throughout the AI sector are ripping higher.

That said, it’s hard to argue against the need for just a little conservatism in light of the enormous multi-year expectations still priced into this market sector, and the risks it now poses to the broader economy. Yet at the same time, we genuinely believe AI is going to be a foundational technological innovation that changes the world as much (or perhaps more) than the internet did. It just may take a while longer than most expect. Not for a lack of need or demand, but for a lack of physical infrastructure.

Looking back at the late ‘90s / early 2000s internet bubble, the fallout of that implosion was felt throughout the economy and resulted in countless proclamations that the internet itself was just a novel, dying fad.  Looking back some 25 years later, it’s clear that wasn’t the case and “the internet” was a real and lasting thing.  And not just “real,” but far more transformational than anyone realistically expected in November of 2000. In practically every aspect of day-to-day human life, the internet is now a pervasive and virtually unavoidable force which has permanently changed the trajectory of human civilization. We created the technology, and then the technology changed us forever. So, was it a bubble?

Most technology stocks certainly got ahead of themselves. A lot of people made and lost fortunes. A number of “promising” and well-funded enterprises turned out to be nothing but an idea and a dream. Entire sub-sectors and pseudo-technologies rightfully suffered the fatal consequences of too much malinvestment, disappearing completely. And the hype did die for years (replaced quickly, I’m sure many of you will recall, with massive real estate speculation). But the internet didn’t dissolve.  Instead, a global technology renaissance sprang forth from the burst remnants of the so-called “bubble.”

Again, it’s easy enough to see parallels today if you look. And thus, some elevated degree of caution is certainly justified.

But, it's also vital to note that this AI-centric ecosystem is far from uniform or homogenous. There are a variety of sub-sectors focused on different elements of what we all think of as being “AI computing.” Nvidia, the world’s most valuable company with a ~$4.5T market cap, derives its value primarily from hardware sales (GPUs).  Other entities are entirely focused on software solutions that help to operate that hardware more efficiently. Meanwhile, enterprises and researchers are using that hardware to create the next great AI model, or tweak someone else’s existing model, or create new medicines, or new polymers, or a new traffic routing model, or a new [FILL IN THE BLANK]. In many cases, this may create absolutely massive economic efficiencies in businesses across almost every industry over the coming years.

Some companies are focused on cooling these data centers and computing systems using liquid direct-to-chip architecture, while others are planning ahead for even greater power densities through the advancement of immersion cooling technologies. We’ve met project developers building muti-gigawatt greenfield data center campuses in the middle of nowhere, while others are still pressing ahead with hyperscale projects in core markets (not an easy task these days).  Some are converting unused industrial or commercial properties into data centers, while still others, like our Greenlight subsidiary, are developing data center portfolios using smaller pockets of readily available power and substation infrastructure across a more distributed footprint. The point is, there are a lot of different businesses within the broader AI ecosystem and it’s unwise to paint them all with the same brush in evaluating a bubble narrative.

All that said, there are a couple of important but largely overlooked news items we wanted to highlight:

Microsoft’s CEO, Satya Nadella, recently noted that they now have GPUs sitting in boxes for lack of adequate rack space to operate them: Datacenter Dynamics Link.  He is quoted as saying:

“The biggest issue we are now having is not a compute glut, but it’s power – it’s sort of the ability to get the builds done fast enough close to power,” Nadella said. “So, if you can’t do that, you may actually have a bunch of chips sitting in inventory that I can’t plug in. In fact, that is my problem today. It’s not a supply issue of chips; it’s actually the fact that I don’t have warm shells to plug into.”

For those of you following our monthly updates for the past year or so, this should not come as a surprise and seemed to us to be an obvious consequence of not having enough powered rack space at the proper densities to accommodate everything Nvidia, AMD, and others are hoping to sell. This problem is not going to be quickly alleviated.

Two data centers in Nvidia’s hometown of Santa Clara, CA stand empty as a result of a lack of available power: Yahoo!Finance Link. 

The story summarizes the challenge.

“The fate of the two facilities highlights a major challenge for the US tech sector and indeed the wider economy. While demand for data centers has never been greater, driven by the boom in cloud computing and AI, access to electricity is emerging as the biggest constraint. That's largely because of aging power infrastructure, a slow build-out of new transmission lines and a variety of regulatory and permitting hurdles.”

Again, DPO’s view has long been that there is actually quite a lot of available power, it’s just not in the right places and it isn’t necessarily available around the clock.  Stories like this one are a great example of how hard it is to make these elements come together, and why the pace of this AI infrastructure build-out is likely going to be slower than many hope.

So, if overlooking Nvidia’s recent uplifting news, how does one play this theoretical “bubble” as an investor and/or as a participant in this space? We try to frame the industry as follows:

• There is more demand for AI computing today (and likely over the next 5 years) than can be served by existing power, data center, and cooling infrastructure.  Companies are indeed finding significant success with implementing AI tools, and unless we believe that demand has already peaked or is only temporary, we need more chips, GPUs, and power and cooling infrastructure.

• There are not yet enough chips and GPUs to serve today’s AI computing demand.  Nvidia, AMD, and others will need to build more, and their current sales projections and market forecasts already reflect that expectation.

• There are also not yet enough energized data center facilities to accommodate all of the chips and GPUs currently in existence, and certainly not enough to accommodate those units in these manufacturers’ sales forecasts.

• And there are not yet enough generation assets, transmission lines, storage systems, etc. to accommodate all of these data centers that will one day be built.

So, our questions for the industry at large are:

• If you can’t get the power in the right places, how can you build and energize the data centers?

• And if you can’t build and energize the data centers, how can you power the chips and GPUs?

• And if you can’t power the chips and GPUs, how can you manufacture and sell those chips and GPUs at the prices your investors expect?

• And if you can’t manufacture and power GPUs, how can all of these enterprise end-users benefit from AI computing in the timeframes their own investors now expect?

Our expectations are as follows:

• Companies which own powered data center footprints will become increasingly valuable as it becomes clear the powered rack space is the thing in short supply.

• Companies working on cooling infrastructure and/or software to help these facilities operate more efficiently and with less power will have plenty of opportunity to succeed, even in a “bubble.”

• Companies developing data center footprints or powered-land deals with an emphasis on speed-to-market will be seen as increasingly valuable as delays in the broader sector become more obvious.

• Powered land, even in sub-hyperscale sizes, is about to become an even greater gold-rush than most expect today.

• Just last week, Alex shared with our team a 3rd-party powered-land transaction which implied a per-megawatt value of roughly 4x Greenlight’s own internal forecast model.

• Suppliers of backup power, microgrids, switchgear, etc. should become more valuable as they sell more equipment, though it’s worth noting they are constrained by the same elongated approval, interconnection, and construction timelines as those larger projects themselves.

• Islanded generation and computing facilities, once unthinkable, may soon become the norm as the grids struggle to keep up (though this has negative implications for utilities which may not be easy to navigate).

• Companies making computing hardware with a large number of expensive units in their sales forecasts may have a tougher 2026 and 2027.

• Who is frantically buying computing equipment with a 3 to 4-year useful life that will sit in a cardboard box for 24 months before it can be utilized?

In short, the next year or two should be an incredibly interesting time in the computing and energy sectors. There are clearly some signs of froth and overly optimistic cost and timeline projections, even among some major names in the space, but we don’t think you can just label the entire sector (or the technology itself) a bubble.  In 25 years, we’ll be using a lot more AI, just like we’re using the internet a lot more than in the year 2000. Maybe that will be a huge strain on our grids, or maybe some mixture of quantum computing, nuclear power, and technological efficiencies will have largely solved that problem by then. But in the meanwhile, the next 2-5 years is going to be wildly tumultuous as society (and Wall Street) comes to grips with the realities of these endeavors. We continue to think small and nimble has some real advantages over massive and lumbering, and we’re optimistic about our strategy amid this absolutely enormous opportunity.

That’s it for November. As always, we appreciate you continuing to follow along and we wish you all the best as we head into the final month of 2025 and beyond!

Regards,

Andrew Webber, Chairman Greenlight Data Centers