Like the housing market, or airfare, prices for IT equipment have always gone up and down according to shifting marketplace dynamics or macroeconomic trends. And when prices go up, if there’s a need, IT still buys because that is what IT does. The job of IT is to keep the business running smoothly. There’s little tolerance, especially in large enterprises, for subpar performance, security breaches, or data and apps that don’t deliver what executives and departments need to produce great results and win customers.
But this year has been an anomaly, due to the outsized, long-term impact of the exploding AI industry on core IT infrastructure plumbing such as computer memory and flash storage. NAND flash contract pricing has spiked by 70-75 percent in Q2, according to TrendForce. Kingston has cited a 246 percent year-over-year increase in NAND wafer pricing, and many analysts are saying this pricing pressure has no near-term end in sight.
Demand for data center buildouts (primarily in the U.S.) to support the burgeoning AI infrastructure market worldwide has created the shortage and suppliers have struggled to keep pace. Storage vendors are making hay while the sun shines, raising prices by 20, 40, 60 percent and higher.
Enterprise IT is buying anyway, for now. IDC’s Q1 2026 tracker shows the worldwide external enterprise storage systems market hit $9.2 billion in vendor revenue, up 22.7 percent year over year, with all-flash arrays crossing 50 percent of that revenue for the first time. Cheaper HDDs are also not a fallback: Western Digital reports that it is essentially sold out for calendar 2026, with hyperscaler orders locked into 2027 and 2028.
See also: Why Storage is Becoming the Limiting Factor in AI Infrastructure
The Heavy Hand of AI
Sure, longstanding vendor relationships play into buying decisions and if the IT budget isn’t under extra oversight right now, it’s a simple decision to keep adding capacity. But there is so much more at play now. Regardless of where organizations are in the AI maturity phase, which is generally still early according to surveys and popular opinion, IT decision-makers are realizing that every model training pipeline and AI inference platform depends on fast access to large volumes of usable data. That means buying more flash for high-performance AI workloads. Nobody wants to get caught short when AI demand hits prime time at their business. Being late to AI reads as a bigger risk to the business than overpaying for flash.
At the same time, a refresh backlog is coming due. Many organizations delayed storage purchases in 2024 and 2025 as budgets shifted toward AI servers, and that delay has limits: systems age, contracts expire, data grows. This pent-up demand is hitting the market at the worst possible pricing moment because it cannot be deferred further. IDC attributes the more than 60% year-over-year growth in high-end systems specifically to this backlog effect. When the choice is refresh or risk data exposure on aging arrays, buyers refresh regardless of price.
You Can’t Simply Buy Your Way Out of This
But here’s the looming issue that calls for an alternative viewpoint. Buying at any price is a temporary fix. Inventory is becoming scarce for most buyers.
Micron shut down its 29-year-old Crucial consumer SSD brand in December 2025 to redirect all production capacity to AI and enterprise customers, and reports that it can only fill 55 to 60 percent of demand from its largest customers. Lead times on large enterprise orders have stretched past 40 weeks, pushing deployments into 2027 for buyers who don’t hold multi-year contracts.
Moving more data to the cloud is an option, depending upon an organization’s security and regulatory constraints. But how long until cloud prices, which have remained stable thus far, go up too?Across Alphabet, Amazon, Microsoft, and Meta, combined 2026 infrastructure capital spending is now projected near $760 billion, up 77 percent year over year. CEOs of these companies have publicly alluded to the same impact of supply chain constraints as contributors to higher spending.
The Data Conversation
Before buying anything else, IT leaders and their FinOps experts may want to take a pause and ask a simple question: How much flash do we actually need?
- Not every embedding needs to live on NVMe.
- Not every training checkpoint needs primary-tier performance six months after the run finished.
- Not every log file, contract, or medical image needs the most expensive medium available just because that’s where it landed by default.
Data profiles should determine tier, not old habits, and most enterprise environments have never done the classification work required to make more nuanced decisions on right-placing data across the many storage tiers now available.
Various industry estimates suggest that 30-50% of data occupying premium flash today is redundant, obsolete, or trivial, ROT data which consists of duplicate copies, expired projects, old logs, files nobody has opened in years. Then there is the much larger volume of cold data, the average 70% of information that’s still retained for compliance or occasional reference but hasn’t been touched in months or years.
Such low-value data is often still sitting on the most expensive tier of storage in the middle of a supply shortage, because nobody has ever separated it from what’s active. In a market where premium capacity costs roughly four times what it did a year ago, that gap has a real, measurable price tag attached to it for the first time.
Here’s the silver lining. Cleaning out, analyzing, and classifying unstructured data has double value to the business today. The immediate return is a smaller, cheaper flash footprint at a time when it matters the most. Secondly, the same classification exercise, identifying what a file is, how sensitive it is, how often it’s used, is the foundation that enterprise AI leaders need to find and trust the right unstructured data in the first place. This AI data quality gap is prevalent in many organizations and is an often-cited reason why AI projects are stalled in pilots and/or not delivering ROI.
Conclusion
Maybe your IT organization can delay a capacity refresh for now, or maybe you can still procure what you need now for the coming years, even at a drastically higher price. But at some point, the inevitable day will come and you don’t want to be stuck with limited options or supply chain delays that hurt your business. Data storage and infrastructure are strategic levers today in the enterprise like no time before. Will you take the chance that your core vendors will take care of you and that the budget will flex? Or will you choose a different path, one where data comes first and managing it strategically across storage silos with right-placing and intelligence sets your course?
