An increasingly urgent challenge facing data centers is the staggering volume of hard drives reaching end-of-life.

While headlines celebrate soaring AI investments and ever-faster GPUs, most “dark side” coverage stops at energy consumption and waste-heat reuse. Almost no one talks about the other elephant in the server room: storage media simply doesn’t last as long as the data we keep. Much of what enterprises store must live for decades - far longer than the five-to-ten-year lifespan of the media that hold it.

In practice, that means continuous migration. After roughly five to 10 years, data must be moved to fresh HDDs, SSDs, or tape - and the old media must be retired – some are resold after data sanitization, others are incinerated or shredded. In the best-case scenario, the resulting electronic waste is separated and recycled. There are no definitive public statistics on annual disposals across all media types. But a simple look at historical shipments is sobering. Assume an average seven-year service life: the HDDs shipped in 2018 are aging out now. With roughly 375 million HDDs shipped that year, that’s on the order of one million drives hitting end-of-life every day.

Even though annual HDD unit shipments have fallen since their 2010 peak (~650 million), per-drive capacity has climbed, especially for data center and enterprise applications – so the industry has shipped more EB with fewer hard disk drives. This trend is likely to continue with the current growth in HDD areal densities. However, the number of HDDs shipped for data center applications will continue to grow, due to storage demand. From 2019 through 2024, about 1.2 billion HDDs were shipped, which eventually need to be replaced.

Analysts expect the volume of stored data to grow somewhat over five times per decade. Much of this data will likely be stored on HDDs in data centers. HDD unit shipments for data center and enterprise applications in 2024 were estimated to be about 5.6 percent higher than in 2023 and could be about 27 percent greater by 2030. This growth will add to the number of HDDs that eventually will need to be replaced.

Performance-optimized media in a longevity world

Today’s storage technologies were engineered first for compute performance, not for century-scale preservation. That mismatch is becoming a sustainability problem as datasets swell. Beyond the physical waste and materials intensity of device manufacturing, there’s the energy cost of simply keeping data alive: data only survives over time when the power stays on.

Data migration and maintenance are becoming a significant data-center workload because throughput isn’t scaling anywhere near as fast as data volumes (all of which must eventually be moved). As a result, the balance of data use could shift: more time and energy may go to keeping data alive and less to productive computing – an inevitable and worsening trade-off.

Extending device lifecycles helps – but only delays the inevitable. And large-scale “delete by policy” is unlikely. Data is too valuable today, and AI only increases the incentive to retain it in hopes of extracting future insights.

HDDs remain the capacity storage workhorses

For the foreseeable future, hard drives will continue to service capacity storage needs in data centers. Demand for performant, economical capacity will keep rising.

But if we keep solving all long-term data retention with short-lived media, the refresh treadmill – and its environmental and operational costs – will keep accelerating.

AI amplifies this further: it generates vast new datasets and compels retention of model outputs, training artifacts, and decision audit trails. Early market signals — such as lengthening lead times and upbeat moves in storage vendors’ order books and stock prices — point to an even faster ramp-up ahead.

Emerging approaches aim to break the trade-off between long-term durability and practical access. For example, technologies that target virtually indefinite retention with seconds-scale latency and high-bandwidth retrieval, bringing archival-class longevity closer to online usability. Crucial, the media is using basic materials, such as sand, and a comparatively simple process yielding a smaller carbon footprint versus other media technologies that require higher grade materials and significant more complex processes. The separation of the media from write/read electronics (at the cost of a latency) is a further step towards a deployment adaptable to the service level requirements of a wide range of use cases all the way to a stack of media in the form of a time capsule.

This isn’t tomorrow-morning deployment; scaling any new medium to data-center volumes takes time. But the direction is clear: to make data retention sustainable under exponential growth, we need media designed for longevity first, without giving up practical access.

A call to build – and fund – the future of storage

There’s good news: Technologies with long term media life are being developed. These need to be pursued and scaled. Historically, storage has been “unsexy” from a value creation perspective for venture investment compared to Compute or AI. That will change. Solving the storage sustainability gap will require not just technical imagination within the industry, but visionary investors willing to back platforms that deliver orders-of-magnitude improvements in lifespan, manufacturability, and total cost of ownership.

If we want AI’s promise without drowning in our own data exhaust, we must rethink how we store information for the long haul. One million drives a day is a stark metric. It should also be a rallying cry.