How AI Is Accelerating Part Obsolescence—and How Companies Can Adapt

We know demand from AI is driving tighter supply and skyrocketing prices. But did you know that it’s also playing an increasingly large role in EOL events?

How AI Is Accelerating Part Obsolescence—and How Companies Can Adapt

Article Highlights:

  • Component obsolescence used to follow how mature a technology was. A part was retired once something better had replaced it and demand had consequently faded. Now, however, EOL events increasingly follow critical decisions around capacity allocation.

  • Memory is the clearest example of how the AI industry is driving component obsolescence across the electronics manufacturing ecosystem. Micron, for example, issued end-of-life notices for DDR4 and LPDDR4 in June 2025, with DDR4 shipments expected to end within two to three months.

  • Memory gets the headlines, but the same pressure is reaching other semiconductor commodities as well. In early 2026, TSMC said it would streamline mature-node production, including 6-inch and 8-inch wafers, to free up capacity for AI products.

Over 600,000 electronic components were discontinued or declared obsolete in 2025. According to Z2’s lifecycle data, this is roughly one-third more than in 2023. The only year with more obsolescence events was 2022, when well over 700,000 components reached end-of-life (EOL) in the aftermath of the pandemic. What changed between 2022 and 2025, though, was the primary obsolescence drivers. The 2022 spike came from a sudden global shock to supply chains. The current wave of part obsolescence, on the other hand, comes from a deliberate, long-term choice across the semiconductor industry to prioritize AI and all the hardware that makes it possible.

In the age of AI, chipmakers are allocating fab capacity, engineering time, and capital to the artificial intelligence companies determined to build out their infrastructure and willing to pay premium prices to do it. Products for industrial, automotive, medical, and defense customers are often older parts built on mature processes. Because of this new competition, they’re now getting fewer wafer starts, longer lead times, and earlier EOL notices. For procurement and engineering teams operating in those sectors, this requires careful planning, and accepting the reality that preexisting assumptions about part lifecycles no longer hold.

Why AI Demand Is Shortening Component Lifecycles

A fab has a fixed amount of capacity. Every wafer start that goes to high-bandwidth memory (HBM) or advanced logic is one that can’t go to another semiconductor product. In addition, HBM uses an unusual amount of capacity: it requires roughly three times the wafer area to produce the same capacity as conventional DRAM, along with complex back-end steps such as through-silicon vias. When AI demand takes up that much production capacity, manufacturers have to decide what to stop making. Perhaps unsurprisingly, older, lower-margin lines are the ones that are getting cut.

Analysts increasingly see this as a structural change, not a cyclical one. In December 2025, IDC described the memory shortage as “a potentially permanent, strategic reallocation of the world’s silicon wafer capacity.”

The practical result is a change in why parts go obsolete. Component obsolescence used to follow how mature a technology was. A part was retired once something better had replaced it and demand had consequently faded. Now, however, EOL events increasingly follow critical decisions around capacity allocation. A component can now receive an EOL notice while it still has healthy demand and a large installed base, simply because its production line is worth more if it’s dedicated to manufacturing something more profitable.

The Memory Market Shows What’s Coming

Memory is the clearest example of how the AI industry is driving component obsolescence across the electronics manufacturing ecosystem. Micron issued end-of-life notices for DDR4 and LPDDR4 in June 2025, with DDR4 shipments expected to end within two to three months. Later that year, the company took those measures a step further still. Micron announced it would exit its consumer business to focus on advanced memory for AI data centers, which ended the 29-year-old Crucial brand.

A recent announcement from Samsung is just as telling, only in the opposite direction. The company delayed its planned DDR4 obsolescence after signing a long-term non-cancellable, non-returnable contract with a key customer, with that supply earmarked for server clients. Lifecycle extensions still happen, but they go to the largest buyers with the strongest contracts. Mid-sized original equipment manufacturers (OEMs) and industrial manufacturers are usually not in that group.

Availability also tends to disappear before the official deadline. Suppliers have published last-time-buy windows for DDR4, but parts are getting harder to find well before those dates. Pricing points the same way: DDR4 and even DDR3 contract prices are rising faster than newer memory types as capacity moves elsewhere, which is the opposite of the discounting buyers used to expect near EOL. Secondary suppliers will help over time, but Winbond and Nanya aren’t expected to have enough capacity to fully meet DDR4 demand until the end of 2026.

The Squeeze Is Spreading to Mature-Node Parts

Memory gets the headlines, but the same pressure is reaching other semiconductor commodities as well. In early 2026, TSMC said it would streamline mature-node production, including 6-inch and 8-inch wafers, to free up capacity for AI products. The mature 28 to 90 nm processes used for microcontrollers, analog ICs, power devices, and passives are under similar strain, as foundries prioritize their most profitable capacity.

Another factor is making this transition even harder for traditional industries now forced to compete with artificial intelligence. AI data centers aren’t built only from GPUs and HBM. They also use large numbers of conventional components, so data center builders are buying many of the same legacy chips everyone else needs. Power management parts are a good example. Power management ICs and MOSFETs aren’t AI components, but AI-driven demand is making them increasingly more difficult to source. A medical device manufacturer buying a common voltage regulator may now be competing with hyperscalers for the same allocation.

Why Obsolescence Hits Some Industries Harder Than Others

Consumer electronics companies can often design around a discontinued chip in their next product cycle. But organizations whose products need to stay in service for decades don’t have that same level of flexibility. Aerospace and defense is the strongest example of this new vulnerability. Defense accounts for less than two percent of the U.S. semiconductor market, so it has little leverage over chipmakers. Meanwhile, an aircraft like the F-35 is expected to stay in service for 20 to 40 years. This creates a stark mismatch between the lifecycle of the product and the lifecycle of the components that go into it.

Automotive, industrial automation, and medical device manufacturers face the same incongruity to varying degrees. A redesign means requalification, retesting, and, in regulated industries, recertification, which can take longer than the remaining life of the part being replaced. Part obsolescence in these sectors directly affects margins, production schedules, and sometimes even regulatory standing.

How Companies Can Adapt to Accelerated Part Obsolescence

For businesses looking to mitigate the worst impacts stemming from part obsolescence, the fundamental risk management measures haven’t changed: redesign, last-time buy, or extended end-of-life production through an authorized aftermarket source. What has changed is how early teams need to decide between these options.

Treat lead-time changes as early warnings

An EOL notice is no longer the first sign of trouble. A part whose lead time suddenly jumps from eight weeks to twenty-six—or that keeps going into allocation—may be getting pulled toward a high-priority market like AI or quietly wound down ahead of a discontinuation. Either way, the buyer has a problem, and the earlier it gets flagged, the more options they’ll have to choose from.

Move lifecycle risk into design reviews

Lifecycle risk now needs to be part of the design phase. Before a part is approved for a new design, engineers should ask whether it shares a process or fab with products the manufacturer is prioritizing for AI. Market breadth matters as well. Parts that sell into many end markets, with no single dominant segment, tend to be more resilient than parts that depend heavily on a single one.

Scrutinize supplier behavior

Published longevity programs can help, but they are only commitments. A supplier’s track record on EOL notice periods, product change notifications (PCNs), and support for industrial customers says more about how it will act when capacity is tight than its marketing does.

Make last-time buys a data decision

Buying too much ties up working capital, risks write-downs, and raises storage and shelf-life problems. Buying too little, on the other hand, forces an expensive redesign or a trip to the broker market, where counterfeit risk goes up. Last-time-buy quantities should come from actual consumption forecasts, the remaining life of the product, and the realistic timeline for qualifying a replacement.

Qualify alternates before you need them

Form-fit-function alternates and second sources take time to validate. Teams that already have pre-qualified alternates for their highest-risk parts can treat an EOL notice as routine. Teams that don’t end up redesigning under a tight turnaround, with meaningful revenue at risk.

Manage Part Obsolescence With Z2

Each of the steps mentioned above depends on having current, accurate lifecycle data for every part in your bill of materials. Supply chain risk management (SCRM) platform Z2 provides component intelligence on more than one billion electronic parts, including lifecycle status, obsolescence forecasts, and last-time-buy windows. BOM Risk Grades show which assemblies have the highest concentration of at-risk components, too, allowing engineering and procurement to focus on the highest-risk parts first. Built-in alternate part recommendations help teams qualify replacements before an EOL notice arrives.

AI is changing how chipmakers decide which parts to keep making. Companies that track that change at the part level will have time to adapt. Those that don’t, meanwhile, will learn about it from EOL notices. A software solution with the market and data visibility of Z2 can mean the difference between one result and the other.

To learn how Z2 can help businesses navigate component obsolescence with actionable insights and industry-leading market intelligence, schedule a free trial with one of our product experts.