The servers humming in vast data centers built for artificial intelligence training don’t just consume electricity. They swallow hard drives by the millions. And that hunger now ripples outward, hitting the organizations tasked with preserving what humanity has posted online.
Brewster Kahle noticed it first in the procurement orders. The founder of the Internet Archive and its Wayback Machine told 404 Media the preferred 28- to 30-terabyte drives have grown scarce or prohibitively expensive. “We have found that the preferred 28-30TB drives are just not available or at very high price,” he said. The Archive pulls in more than 100 terabytes of fresh material daily. It already sits on more than 210 petabytes. Upgrades never stop. “We gather over 100 terabytes of new materials each day, and we have over 210 Petabytes of materials already archived on machines that need continuous upgrades and maintenance, so we need to constantly get new hard drives.”
A very real issue. Costing time and money.
His words landed in early May 2026, but the pressure had built for months. Western Digital told investors in February that it had sold its entire 2026 production run to major enterprise customers. CEO Irving Tan put it plainly during the company’s fiscal second-quarter earnings call, as reported by PCMag: “We’re pretty much sold out for calendar 2026. We have firm POs with our top seven customers.” Those buyers run the hyperscale data centers feeding large language models. Smaller purchasers, including nonprofits and universities, stand at the back of a very long line.
Prices tell the story faster than any earnings transcript. Reports point to jumps of 30 to 50 percent for high-capacity enterprise drives since late 2024. Some accounts cite certain models tripling in cost. A 4TB drive that sold for around $100 last summer now lists near $190 on retail sites. Secondary markets have sprung up. Scalpers move SSDs on eBay at markups that would have seemed absurd two years ago. Tom’s Hardware noted large-capacity HDDs now cost up to three times more, with production capacity largely booked by hyperscalers.
The Wikimedia Foundation feels it too. With more than 65 million articles on Wikipedia and vast media repositories in Wikimedia Commons, storage forms the backbone of its mission. A spokesperson told 404 Media that price increases began to bite after the end of 2025. “We’ve certainly seen price increases since the end of 2025. These price increases are of concern to us, as with every other player in the industry. We see the primary impact in the purchase of memory and hard drives but also in terms of lead times on server deliveries and our capacity to place future orders.”
Nonprofit discipline kicks in. The foundation maintains its own data centers. It stretches existing hardware, shifts investment priorities, and searches for flexibility. Yet the pressure remains. Every extra dollar spent on drives pulls from other budgets. Every delay in procurement risks slower growth of the knowledge base millions rely on daily.
Academics and individual data hoarders face tighter constraints. University research labs that once bought drives in moderate volume now compete against trillion-dollar balance sheets. Hobbyists who filled racks with archival projects have slowed or stopped. Some report waiting lists stretching into 2027 or 2028 for the specific models optimized for long-term, high-density storage.
Micron made the economics explicit. The company exited its Crucial consumer memory and SSD business entirely. Its statement cited “AI-driven growth in the data center” that created a surge in demand. Resources shifted to the largest strategic customers. The message was clear: consumer and small-scale buyers no longer sit at the front of the allocation queue.
But storage forms only half the problem. Archivists also battle website owners who have tightened defenses against automated crawlers. Many operators now block bots indiscriminately, unable or unwilling to distinguish between those harvesting data for AI model training and those capturing snapshots for the historical record. The same measures that slow commercial scrapers also hobble the Internet Archive and similar projects. Preservation becomes harder even before the first byte reaches a hard drive.
Kahle remains pragmatic. The Archive benefits from community donations and continues to hunt for manufacturer partnerships. “We are fortunate to have an active community that donates to the Archive, and we are also looking for help from hard drive manufacturers in these difficult times,” he said. Workarounds exist for now. Yet the trend points in one direction. Costs climb. Availability shrinks. The pace of archiving slows.
Analysts tracking the supply chain saw this coming. Backorders for enterprise-grade HDDs stretched to two years in some cases as early as late 2025, according to reports from that period. AI training datasets, inference logs, synthetic data generation, and video output all demand massive nearline storage. High-capacity mechanical drives still offer the cheapest dollars-per-terabyte for cold and warm data. Until alternatives scale, the drives remain the bottleneck.
Recent coverage reinforces the point. Slashdot’s discussion of the 404 Media reporting drew comments from data hoarders and sysadmins watching their procurement budgets swell. Yahoo Tech and Gadget Review pieces from the past week described the situation as an existential threat to digital history, with hyperscalers locking up supply that once flowed more freely.
The irony sits heavy. Artificial intelligence systems train on vast scrapes of the public internet. Those same systems accelerate the scarcity of the hardware needed to keep copies of that internet intact for future generations. Delete a page today and the Wayback Machine might never have captured it. Pay three times as much for the drive that would have stored the snapshot and the economics of preservation shift.
Industry watchers expect no quick relief. New fabrication capacity for high-density HDDs takes years to build. NAND flash faces its own supply strains from AI workloads. Data centers continue to sign long-term purchase orders that stretch into 2027 and 2028. Nonprofits and public institutions cannot match those commitments.
So the Internet Archive keeps crawling. Wikimedia keeps serving. Researchers keep requesting grants that stretch further than before. And somewhere in a server rack, a 30-terabyte drive that once cost $550 now carries a sticker price closer to $600. If you can find one at all.
The record of our digital age grows more expensive to maintain. That fact alone should give pause to anyone who assumes the internet’s memory will persist without deliberate effort. Shortages have a way of forcing choices. The choice here pits the relentless expansion of AI computation against the quiet work of keeping humanity’s collective output from vanishing into bit rot.
WebProNews is an iEntry Publication