While the data center industry continues to experience unprecedented growth, some have begun to question the size and staying power of the central new demand driver: AI.
As is typical, unknowns surrounding the ultimate size of the AI market are creating conversations about whether the data center industry has become overblown. Industry leaders like Alibaba’s Joe Tsai have made waves with comments about potential bubble formation, and reporters at the Wall Street Journal have followed up with comparisons to the infamous dotcom bust of the 2000s.
CBRE's North America Data Center Trends report has shown vacancy rates for primary North American data center markets falling to a record low of 2.8% in early 2024 from 3.3% a year earlier. Notably, this drop in vacancies persists despite the largest annual increase (by number of MW) in data center supply. In fact, North American data center inventory grew by 24.4% year-over-year in Q1 2024, adding 807.5 MW across key markets. In CBRE’s recent 2025 report, they note that demand continues to outpace new supply across both core and emerging hubs, with the global weighted average data center vacancy rate falling by 2.1% YoY in Q1 2025. North America exhibited a 43% YoY increase in inventory in Q1 across the four largest markets (Northern Virginia, Chicago, Atlanta and Phoenix) while still maintaining a relatively low vacancy rate.
Meanwhile, experts at the Stanford Institute for Human-Centered Artificial Intelligence (HAI), note in their 2025 AI Index Report that AI is becoming increasingly embedded in everyday life. We’re seeing record investment and usage — and optimism is rising. In 2024, U.S. private AI investment grew to $109 billion, with generative AI attracting $33.9 billion in private investment globally.
But the billion (trillion... zillion?) dollar question remains: Is all this a reflection of rock-solid expansion built on strong demand fundamentals, or is it indicative of a bubble growing on the back of a still-evolving tech revolution?
Stream specializes in technical infrastructure and innovation, but at our core, we are real estate developers with 26 years of disciplined investment within the data center sector. Unlike previous data center demand-driving trends (IoT, edge data centers, etc.) the demand from AI is well supported by traditional real estate metrics (in addition to the prognostications from industry experts).
Still, when it comes to development and investment, we’re firm believers in finding what’s real and creating transparency. Nobody can predict the exact future, but let’s talk about what we can see from where we are today.
Real vs. Fake Supply: Do We Have an Overprovisioning Problem?
Our data center development experts (and other specialists) have talked a lot about the danger of developers that tie up land and power without the time, resources or expertise to ultimately deliver real capacity. Often, these developers aren’t strategic investors capitalizing on ample demand with the actual skill to supply — they’re gamblers trying to win a hand just by pulling up a seat at a ‘hot’ table. With those developers involved, it’s true that the game can appear to be less stable or logic-driven than it really is.
Tsai noted, “I’m still astounded by the type of numbers that are being thrown around in the United States about investing into AI,” and he’s got a point — we’ve seen those numbers too. But from where we stand, we’re not too worried yet.
While most of us are not excited about getting older, there are advantages that come with age and experience. As a company established in 1999, Stream Data Centers has witnessed prior periods of ‘unprecedented growth’ and seen the nay-sayers asserting risk with claims like “The northern Virginia market is massively overbuilt”, most of which did not prove to be true over time.
As we look at some of the more ‘astounding’ numbers that are bandied about, we continue to expect that there is a fair amount of over-inflation of demand by developers who are looking to make easy money on a hot market.
Look at it this way: It’s not news that a few "all hat, no cattle” developers are aggressively relaying their requirements to utilities, swelling queues far past what is realistic and necessary to avoid bottlenecks. But without the capital, experience and customer history needed to build, deliver and operate a campus at the hyperscale quality level, much of this prospective power demand won’t be realized.
Taking these highly speculative projects out of the mix, true capacity demand is likely far lower than current utility queues indicate. It’s not unreasonable to think the same phenomenon is paralleled in the larger development landscape; where there’s immediate unrealistic resource demand, there’s bound to be assumptions of eventual unrealistic data center supply.
Even if we assume that demand and supply are only somewhat overstated, think about the current landscape of data center development and the hurdles providers face: a competitive power market, rapid absorption of prime real estate, still-shaky post-COVID supply chains, etc. Delivering hyperscale facilities is no easy feat, and fears around not having enough capacity thanks to competitive power and real estate landscapes still haven't been completely cleared.
When conditions are unpredictable, a slightly inflated project pipeline can help ensure the industry can deliver enough capacity without falling short — it accounts for natural selection. In some ways, it’s better for the industry to overprovision at the outset with the knowledge that some projects won't succeed than it is to temper our momentum and risk losing necessary ground we won’t be able to make up.
So, will the complexities of the data center industry (and the normal course of time) eventually rein in the non-viable speculative developers? Absolutely. Do current development forecasts indicate a bubble ready to explode? No. Should a slowdown be in the cards? We don't think so.
Like all experienced developers, Stream remains focused on making smart moves in quality markets. Our experience tells us that if development is slowed down too much, it’s likely that we will find ourselves behind the curve with insufficient infrastructure to serve customers.
There is a well-known parable about homes built on different foundations: sand vs. stone. In the parable (and in our industry), structures and companies with strong foundations can weather the storm. Developers with strong foundations have the capital, knowledge and reputation required to continue to be successful, even as the winds invariably shift.
Keeping a firm grasp on market fundamentals with the discipline to suss out what’s valid demand and what’s overblown also helps ensure investment decisions are well informed so capital can make informed choices (and benefit from data center growth without falling for hype).
‘AI Data Centers’ Aren’t as Niche as You Think They Are
We get it, AI is different and a little scary. Even industry experts at places like IBM emphasize the differences in facilities, saying, “In contrast to AI data centers, typical data centers contain infrastructure that would quickly be overwhelmed by AI workloads. AI-ready infrastructure is specially designed for the cloud, AI and machine learning tasks.”
To many, this kind of description could convey that building an ‘AI data center’ means creating facilities that can't support any other kinds of workloads or IT equipment. But data center developers aren’t pigeonholing their infrastructure.
While data center design is changing to support denser networking and more physically compact GPU cluster deployments with technologies like Direct Liquid Cooling, these design and operational changes represent a leap forward for more than just AI.
In fact, because of their efficiency gains, many of these same design innovations would likely still be occurring even if AI’s growth was significantly reduced.
As researchers at JLL note, “While the rise of AI has accelerated development, exponential growth in data center infrastructure was already underway to support demand from legacy cloud-based applications and workloads.”
Traditional enterprise workloads (non-AI applications like file storage and sharing or transaction processing) represent the majority of data center power demand today and are predicted to account for around 55% of total demand in 2028.
Because of this, modern data center designs like Stream’s are AI Ready but built to support flexible cooling transitions from 100% air-cooled systems to a blend of air and liquid-cooling that will change with each customer’s technology refresh cycle. IT systems are rapidly evolving, which creates additional challenges for customers concerned about selecting a data center that can support these changes over 15+ years. Well-designed AI-ready data centers provide the best possible assurance of ongoing technical fit for both traditional and advanced workloads while meeting our hyperscale customers’ commitments to sustainable operations.
Put simply: Like most industries, we must continue to evolve to meet our customers’ requirements. The AI boom drives smart innovation, but not at the cost of safety, reliability, predictability or usability on any level. Foundational digital applications (plus the customers that use them and the organizations that create and manage them) will continue to benefit from AI-driven data center optimization.
If AI investment slows, data center resources won’t be stranded, just reallocated.
But What About the Past — And What About the Future?
In the rearview mirror, we have the dotcom bust. In front of us, we have companies like DeepSeek shaking up the AI world by purportedly challenging the level of investment and resources needed to capitalize on AI. Even if DeepSeek is right, history backs up the idea that decreases in the cost and complexity of AI deployments won’t cause a slowdown in usage. Just look at what NPR is reporting about AI and Jevons Paradox.
Some level of uncertainty is always a given. Boom and bust cycles are an ongoing part of our economic trajectory. However, parallels between today’s established digital economy (backed by a tech world with strong roots across many sectors) and the emerging internet landscape of the 2000s are not apples to apples — they may not even be apples to oranges. The biggest differences in today’s market and the dot-com boom/bust era are the companies at the heart of this AI push.
The leaders in this space are some of the biggest, best capitalized and most well-established companies on the planet. That’s in huge contrast with the 2000s-era companies that ‘dot-bombed’ in large part due to their unproven records and unprofitable business models. The other massive difference between the current AI-driven wave and the dot-com boom/bust is how integral today’s digital services are to almost every facet of modern life in a way they weren't a couple of decades ago.
Of course, regardless of the technology, business or market at hand, assuming ongoing growth at a staggering level is always risky. Like any other asset, smart data center investment requires understanding the changing variables at hand and being wary of overconfidence.
Is there a bubble? Probably not. Will Stream and other mature data center developers fail if there is one? No. Media excitement and inexperienced developers may be contributing to the impression that there is overblown growth, but the reality is different than the assumptions. Furthermore, AI-influenced designs are adaptable and better positioned to evade obsolescence, and the innovations from today’s data center developers deliver relevance far beyond AI.
After 26 years of serving the most demanding data center users on the planet, Stream stays dedicated to being here for 26 more years. This commitment to the future helps us keep our feet on the ground so that we never lose the trust we’ve developed as an expert that’s weathered market ups and downs and knows how to get to safety on the other side.
As disciplined real estate professionals, we’re confident that partnering with developers who’ve gained the trust of demanding end users via decades of discipline and real delivery (informed by previous ‘boom’ cycles) is still the best way to guard against the uncertainty that comes with this territory.
