This post is based on a talk James Patrignelli, Director of Sales at Liquid Technology, delivered at DCD>Connect New York. In it, James broke down what’s happening in the IT hardware supply chain right now, and what it means for data center operators and IT leaders trying to make sense of the market.
The AI Demand Shock: What’s Happening Right Now
What we’re seeing regarding AI infrastructure right now isn’t just normal, incremental growth; it’s a step-function change. We’ve compressed what would normally be a decade of buildout into two or three years, and the numbers reflect that reality.
Cloud capital expenditure is projected to exceed $600 billion in 2026, with roughly $450 billion of that directed toward AI infrastructure — a 36% year-over-year increase from 2025. The AI server market reached $143 billion in 2024 and is projected to grow to $838 billion by 2030 at a 34.3% compound annual growth rate.
For people running and managing data centers or IT assets, these figures are the reason lead times on new equipment have stretched, specific components have become harder to source, and new hardware pricing has spiked across the board. The supply chain was not built for this pace.
That creates a specific opportunity. When existing supply can’t keep up with demand, the secondary market for existing hardware becomes dramatically more valuable.
As James put it during his DCD Talks session: “Every organization in this room is either building AI infrastructure, planning to build it, or competing with someone who is. And that reality has fundamentally changed what your existing hardware is worth.”
How the Supply Chain Broke
The disruption in the supply chain is being driven by three distinct pressure points that interact and cascade into each other in ways that amplify the effect.
The GPU Bottleneck
As you probably already know, NVIDIA controls roughly 90% of AI accelerator spending, and demand for its chips has consistently outpaced what the company can produce. That bottleneck reverberates through the entire ecosystem.
The secondary market data illustrates the pressure clearly. CoreWeave H100 GPUs contracts were recently rebooked at 95% of their original pricing. NVIDIA A100s (hardware built on 2020 architecture) can still command between $8,000 and $18,000 on the secondary market.
Understanding why requires a brief distinction between two types of AI workloads: training and inference. Training is the resource-intensive process of building an AI model; it can take weeks or months and genuinely benefits from the latest, most powerful hardware. Inference is what happens when the model is in use: every time someone submits a question to an AI tool and gets a response, that’s inference. Inference workloads represent the majority of production AI compute, and they run effectively on previous-generation hardware. That’s why older GPUs still have substantial, active buyers.
The Memory Crisis
The GPU shortage doesn’t operate in isolation. AI’s appetite for high-bandwidth memory (HBM) has triggered what IDC describes as an unprecedented memory chip shortage, and the numbers behind it explain why it won’t resolve quickly.
AI applications require roughly three times the manufacturing capacity of standard DRAM. Supply growth is projected at 16% in 2026 against 35% demand growth. IDC expects the shortage to persist into late 2027. NVIDIA alone has become a memory customer on the scale of a major smartphone manufacturer, and when that volume of demand gets redirected to AI, it starves every other segment.
The practical effect for IT asset holders: complete systems with memory already installed become more valuable because the individual components are harder and more expensive to source on their own.
The Cascade Effect
GPU and memory constraints don’t stop there. Standard rack servers are seeing extended lead times. Networking equipment faces similar sourcing pressure. Power distribution and cooling components have their own backlogs. Layered on top of all of this, 2025 tariffs on server hardware, components, and cooling equipment (particularly goods with supply chain exposure to China) have added cost and uncertainty that new-equipment buyers are absorbing directly
When new hardware costs more and takes longer to get, the secondary market becomes a significantly more attractive option for organizations that need capacity now.
There’s also a broader signal worth noting from the depreciation side: Three big hyperscalers (Apple, Microsoft, and Alphabet) have extended hardware depreciation schedules from three or four years to six — a change that reduced collective depreciation by an estimated $18 billion annually. Whether that reflects genuine confidence in extended hardware utility or financial engineering is debated, but either way, it tells you these assets have a longer productive life than the industry assumed even a few years ago.
What These Changes in the IT Supply Chain Mean for Your Organization
Understanding the macro dynamics is useful. Knowing what to do about them is more useful. For IT leaders, data center operators, and CFOs, three shifts in how you think about IT refresh and disposition are worth making now.
Your Refresh Cycle Is Now a Capital Event
The old model treated hardware decommissions as a cost line: write it off, pay for storage and logistics, accept whatever minimal recovery you could get. That no longer reflects the market.
Three, four, and five-year-old hardware has real buyers today, not scrap buyers, but enterprises, research institutions, international operators, and AI startups that need this equipment and can’t get new supply at reasonable lead times or prices. An infrastructure upgrade isn’t just a spend. Handled correctly, it’s a capital recovery opportunity that can partially offset the AI investment you’re already making.
For CFOs specifically: this is the conversation worth having with your IT and operations teams before the next refresh cycle. The hardware sitting in your data center right now may be generating more recovery value today than it will six months from now.
The Value Lifecycle Has Changed
Equipment isn’t falling off a depreciation cliff anymore; it’s stepping down through a sequence of use cases, each with its own buyer market.
It’s like player movement in professional sports. The top teams compete for the latest talent. But when those players move on, they don’t disappear; they go to other teams that are thrilled to have them. A starter on last year’s championship roster is still a huge upgrade for a team that’s building. That’s what’s happening with IT hardware right now.
Each stage has active demand. The implication is that a thoughtful disposition strategy — one that times exits against where hardware sits in that lifecycle — can generate meaningfully more recovery than a standard decommission schedule.
Your ITAD Partner Matters More Than Ever
We’ll acknowledge the obvious: we’re an ITAD provider making this case. That said, the argument holds, regardless of who’s making it, and the stakes are higher in this market than they were even two years ago.
Maximizing recovery on high-value hardware requires direct buyer relationships, not auction pipelines that pool your equipment with commodity inventory. It requires accurate market valuation before disposition decisions are made, not after. It requires verified data security practices (NAID AAA certification and documented chain of custody) because the regulatory exposure from a data breach on decommissioned hardware hasn’t gotten smaller as asset values have grown. It also requires downstream accountability: R2v3 certification and environmental compliance that actually holds up to audit.
The value in this market is real. Whether your organization captures it depends substantially on who’s handling the disposition.
Where This Is Headed And How Long the Window Stays Open
Two things are true simultaneously.
The structural demand driving secondary market values is real and long-term. AI infrastructure investment is backed by multi-year build contracts, sovereign compute initiatives, and deployment across every major industry vertical. The supply chain pressure, chip shortages, extended lead times, and tariff impacts aren’t resolving on a short timeline. The memory shortage alone is projected to persist into late 2027.
At the same time, the peak of the value window is finite. New chip generations are reaching general availability. Next-generation architectures are moving through the pipeline. New manufacturing capacity is coming online. When supply and demand find a new balance, the premium on previous-generation hardware will ease.
The takeaway is straightforward: assess what you’re sitting on now. Understand what your hardware is worth to an active buyer in today’s secondary market, and make disposition decisions with that information in hand. The window isn’t closing tomorrow, but it isn’t permanent either.
Turn Your Excess IT Equipment Into Working Capital with Liquid Technology
Liquid Technology has spent 25 years in IT asset disposition. Long enough to have navigated multiple market cycles and to recognize when conditions are genuinely different from the baseline.
The current environment is different. The organizations that recover the most from it won’t be the ones that decommission on autopilot. They’ll be the ones that treat their refresh cycle as the capital event it’s become, work with a partner who can accurately value and efficiently move high-demand hardware, and maintain the compliance standards — NAID AAA, R2v3, and ISO certifications — that protect them regardless of what the hardware is worth.
If you want to understand what your current IT inventory could recover in today’s market, we’re a straightforward conversation away. Learn more about our IT asset management services or explore our data destruction and disposition capabilities.