Demand for computer chips is blazing hot. Investor sentiment is another matter entirely. Recent market pullbacks and mixed messages are signaling caution on capital-intensive bets, like, you know, the massive data center projects tied to AI.
In this crazy world of hot chips and cold feet, where does that leave CIOs? If AI projects get scaled back, paused or shelved, what happens to all that hardware and infrastructure being built today? Will the slowdown (or abandonment) create opportunities for CIO innovation — or deliver a gut punch to your already-stretched AI and budget strategies?
Following the money
Many CIOs find themselves at a crossroads — trying to decide whether their AI projects are tied to a rising star or destined to crash and burn before they deliver a decent use case or a glimmer of ROI.
On the one hand, Nvidia reported a jaw-dropping $57 billion in revenue for Q3 2025, up a whopping 62% year-over-year and mirrored by the booming data center business — together, underscoring skyrocketing demand for AI. Yet, a disconcerting pre-Thanksgiving broad blue-chip retreat — across major benchmark indexes and individual blue-chip names — quickly knocked the bloom off Nvidia’s earnings news, as fears of an AI bubble roared back to front of mind for executives and markets.
So where does that leave all the shiny new data centers — whether freshly built or under construction? Will market doubts see them ditched and forgotten, or will the industry’s enduring optimism about AI keep the boom going?
“Realistically, I don’t see an end of build coming,” said Michael Bergen, executive vice president of analytics and marketing at Industrial Info Resources (IIR), a market research organization that delivers critical global supply-side intelligence for the energy markets.
There are some technology advancements that “aren’t ones we can ever go back on,” Bergen said, likening AI cravings to that of Internet speeds. “Imagine going back to dial-up internet after having experienced broadband; that just isn’t the direction we’re moving in.”
Moreover, according to IIR’s tracking, AI data center projects are planned out over the next decade. “Really, the only things that could stop them are politics or the [lack of] availability of materials,” he said.
Well, maybe that’s not all that could rupture AI data center projects.
“It’s very difficult to identify asset bubbles before they burst. Sometimes they might just be balloons, with the ability to deflate via asset corrections,” said Shriram Bhashyam, COO of Sydecar, a special purpose vehicle and fund administration platform. “We are definitely seeing ‘bubbly indicia,’ he added, referring to early bubble-like signals in the market.
For one, startups are showing an especially unhealthy disregard for risk. “There are many telltale symptoms: overvaluation, investor FOMO and exuberance among retail investors being pitched on data center builds, and media frenzy,” Bhashyam said, pointing to Thinking Machines Lab raising the largest seed round ever: $2 billion at a $10 billion post-money valuation as a prime example. “This was done without a product and without disclosing what it was building.”
The public side is a bit foggier. Stock analysts are still debating whether Nvidia and the hypescalers are overpriced or not. But it appears that the big money is falling on the gloomier side, especially after Ray Dalio, billionaire founder of Bridgewater Associates, called the latest market boom a “big bubble with big wealth gaps poised for a politically explosive bust” in a CNBC interview.
Why the gloom and doom on the public investment side?
One big reason why the bubble question keeps surfacing is because AI spending and AI revenue are dramatically out of sync, Bhashyam explained. Industry estimates suggest that roughly $400 billion is being poured into infrastructure to build, train and operate AI models, compared with only about $45 billion in AI revenue last year.
“With a 3- to 4-year useful life of a chip or processor, and spending expected to multiply in the coming years, one has to squint to see the path to a return on investment,” Bhashyam said.
Even so, all may not be lost (or so investors and CIOs managing big, expensive AI projects hope).
“Even with pockets of speculation, this might be more of a transformative bubble. In that case, we might see some near-term corrections, but over the long term, the transformative power of AI might dwarf the dollars invested in it over the next few years,” Bhashyam said.
When infrastructure outpaces demand
Might, it should be noted, being the operative word. Now that everyone feels like they’re standing at a roulette table in Vegas trying to arbitrarily pick a winning number, it’s time for responsible CIOs everywhere to develop a backup strategy.
According to a McKinsey report, companies will invest almost $7 trillion in capital expenditures on data center infrastructure globally by 2030. The hyperscalers are not the only companies on an AI data center spending spree.
“There is a kind of mania right now to keep investing in high-density facilities, but whether or not the bubble bursts, there will eventually be a need for all this infrastructure,” said Joe Morgan, COO at Patmos, a technology provider specializing in digital infrastructure, with a focus on hosting, AI compute services, custom data centers and ISP solutions.
“There is an obvious parallel here with the dot-com boom, when questions were raised about the massive investment in fiber, and subsea, and domestic broadband, and the previous generation of data centers,” Morgan pointed out. “Did the bubble burst then? Yes. Do we all still benefit from those investments? Also, yes.”
There’s also the too-big-to-fail question of it all, he added. There’s probably too much momentum to stop the data center investment train before it runs out of track.
“The companies building gigawatt data centers are kind of too big to fail. These are hyperscale projects from the world’s biggest IT companies. The question is, when they all come online in two years’ time, will the expected demand actually be there? I honestly think that nobody knows,” Morgan said.
Whether or not the bubble bursts, there will eventually be a need for all this infrastructure.
Joe Morgan
COO, Patmos
The coming reset in AI data centers
Building a backup plan to survive and prosper in this scenario requires CIOs to consider alternative uses for these shiny new data centers, in case any are abandoned or underutilized.
“I wouldn’t expect widespread abandonment, but we will see delays, scope reductions and ownership changes,” said Shishir Shrivastava, practice director at TEKsystems Global Services. “The industry is consolidating and maturing, hyperscalers are acquiring smaller firms and adjusting capacity plans to better align with consumption. Some single-tenant AI builds will be converted into multitenant or colocation facilities, allowing operators to diversify usage and stabilize returns.”
Projects that continue successfully will be those designed with flexibility in mind, he said — for example, with modular layouts, scalable cooling and the ability to support mixed workloads.
“This moment is less about collapse and more about optimization,” Shrivastava added.
This could mean plenty of options for CIOs to reduce operational, computing and storage costs. But there can also be some complications with those deals and beyond.
Energy becomes the next major constraint
The AI boom is about to hit an energy wall, which is the next big bottleneck, Shrivastava said. Building new data centers quickly isn’t a problem, but creating new power generation overnight isn’t possible. “As LLM workloads continue to scale, energy shortages will become a defining challenge for hyperscalers and enterprise data centers alike,” he said.
If AI growth slows, it could be a temporary reprieve that eases grid strain in dense data center regions.
“But the longer-term challenge remains: how to power these facilities sustainably. Many next-generation AI data centers are already turning to renewable sources and liquid cooling, but that introduces new water demands,” Shrivastava added.
Still where there is loss, there is also gain, if your strategy and negotiation points are rooted in reality.
A potential glut — and real consequences
“First off, if AI infrastructure outpaces demand or the bubble pops — say, due to model efficiencies stalling massive training runs or enterprises pulling back on budgets — we’re likely looking at a glut by 2026-2027,” said Adnan Masood, chief AI architect at UST.
He noted that signs of this reversal already exist and points to several indicators: Microsoft has already halted planned data center projects, amounting to roughly 2GW of power capacity in the U.S. and Europe, and is reportedly looking at leasing out excess capacity by 2027-28; and AWS has paused leasing discussions in key spots. Plus, Masood noted that the user base is struggling: China’s already at 20-30% utilization on their AI compute, leading to the scrapping 100-plus AI projects.
“Yeah, some [new data centers] could get abandoned mid-build or right after — think half-finished shells in hot markets like Northern Virginia or Phoenix, where permitting delays or demand shifts hit hard. We’ve seen it with Microsoft’s Wisconsin site, where they halted after dropping $262 million,” Masood said. But total abandonment? Unlikely.
“More often, it’s mothballing or fire sales,” he said, offering an example: “Assets like Nvidia H100 GPUs, whose cloud rates dropped from $8 per hour in 2024 to $3 per hour now, using Thunder Compute, flood secondary markets, depreciating 35%-50% in a hard bust scenario.”
Adnan Masood, chief AI architect, UST
Savings and shortfalls for CIOs
Bottom line? In the short term, CIOs may take a hit from an AI bubble burst, and it’s not too soon to plan a just-in-case rebound strategy now.
“CIOs might face write-downs on recent buys, the economy could see a tech-sector slowdown echoing dot-com’s $5T wipeout, and vendors like Nvidia risk order cancellations, with REITs [Real Estate Investment Trusts] writing off empty facilities. Supply chains ease up, though, meaning less scramble for transformers or concrete,” Mahood said, ticking off scenarios.
But on the flip side, there could be some major bargains in that bust, too — indeed a veritable “‘goldmine’ for CIOs — if played right,” Mahood said.
“Imagine locking in 20%-plus discounts on colo leases or GPU rentals — AWS already slashed H100 instances 45% this year,” he said. According to Mahood, strategies include:
-
Burstable contracts (commit low, burst high at marginal cost).
-
ROFR on decommissioned hardware (grab those stranded GPUs cheap).
-
Snagging orphaned renewable PPAs for your own sustainability goals.
“Enterprises could experiment with AI projects that were too pricey before, like custom models for supply chain optimization,” Mahood said.
CIOs can probably snag more than a few hardware bargains after an AI bust too, according to Eric Ingebretsen, chief commercial officer at SK Tes, a global IT asset disposition company. His assessment:
“We expect demand for secondary market enterprise equipment to remain high and continue to see increases in decommissioning projects from hyperscale data centers, as uncertainty about tariffs and economic caution dissipates, resulting in a steady flow of high-quality enterprise equipment into the market. We are seeing surging demand within the enterprise and data center sectors, particularly for components such as HDDs, SSDs, memory and GPUs,” Ingebretsen said in an email.
Planning ahead for multiple scenarios will prevent any panic thinking and allow you time to map out the advantages you’ll want to seek and acquire. But don’t procrastinate for too long.
“We will eventually make use of the infrastructure being built, but getting there may require suppliers take a haircut to wait for downstream demand to catch up. And any temporary glut in computing capacity could ultimately benefit CIOs by lowering the cost of computing,” said Professor Andy Wu at Harvard Business School.
The wider fallout CIOs can’t ignore
CIOs might also want to consider offering some sort of aid or advice for the communities that their companies serve, or where they and other employees live. An AI bust will hurt these areas if data centers are in the vicinity.
“CIOs who anticipate this shift can benefit by acquiring computational capacity at lower cost, but energy grids and local ecosystems may bear the scars of overexpansion. The lesson for CIOs and investors alike is clear: Sustainable advantage will belong to firms that integrate AI strategically, not those merely chasing the hype,” said Professor Frédéric Fréry, Co-Director of the ESCPTech Institute, ESCP Business School.
Indeed, an AI bust will likely hurt everyone. But its continued growth may be harmful as well. There are tough problems ahead either way.
The huge investments taking place in the AI space today, including data centers, cloud computing and energy providers – indeed, the entire technology ecosystem — is connected with the rest of the economy, said Sumit Johar, CIO at BlackLine, a cloud-based financial platform.
“While the AI boom is raising the risk of climate change with exponential growth in energy use, a sudden downturn can lead to a significant downturn in the technology spending that may impact the overall economy significantly,” Johar said.
Hope for the best, plan for the worst. Strategy wins the day.
