WD says inference leaves far more data than users see
Original title: AI’s hidden storage bill: What inference leaves behind
An 8-second 3.97MB clip left more than 500MB of logs, caches, renders, backups, and safety records.
A MIT Technology Review piece, drawing on Western Digital measurements and a WD-sponsored IDC survey, argues that AI budgets undercount storage created by inference. WD says one short generated video left about 130 times more back-end data than the file returned to the user. McKinsey is cited projecting inference demand growing at a 35% CAGR through 2030 versus 22% for training, and more than 40% of data-center demand by decade’s end. The survey of 763 decision makers found nearly 95% reporting more data from generative AI.
Unconfirmed: The storage multiplier is WD’s internal measurement, and the survey was WD-sponsored.
- About 61% of respondents reported data-volume growth of 25% or more in 12 months.
- 96% anticipate needing faster archive retrieval for inference or RAG.
- The piece says retained logs and outputs can later feed RAG workloads.
- It argues HDD tiers suit large warm, cool, and cold AI data volumes.
- Vendors
- Western Digital
From the original article
4 min read at technologyreview.com
Ask most executives to picture the cost of AI and they’ll likely see GPUs—racks of accelerators with gigawatts of power—and a waitlist for silicon. But that picture leaves something out: Data storage costs. What happens to data after a model responds?
By Tim Rausch · technologyreview.com
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.