WD measurements show an 8-second AI video clip generates over 500MB of back-end logs, caches, and backups.
Most AI infrastructure budgets track GPUs and power, yet inference leaves a far larger data trail than the file a user receives. WD measured an 8-second 720p clip at 3.97MB and found more than 500MB of linked logs, cached context, renders, backups, and safety records behind it.
McKinsey projects inference demand will grow at a 35% compound annual rate through 2030, outpacing training at 22%, and will account for more than 40% of total data center demand by decade end. A WD-sponsored IDC white paper surveyed 763 IT and business decision makers and found nearly 95% saw data volumes rise from generative AI in the past year.
Training runs end, so teams can budget and retire them, while inference keeps generating retained data that does not shrink when compute demand dips. Storage growth of 25% a year quietly outpaces the compute line that gets the most attention, and idle GPUs waiting on data burn expensive hours fast.
Operators should map storage tiers to the AI data life cycle, separating active inference context from cold logs and synthetic outputs that organizations now hold longer. Watch for vendors and cloud providers to price retention and tiering explicitly as inference workloads scale into hundreds of gigabytes per clip.
What matters
- WD found that an 8-second 720p clip leaves over 500MB of back-end data, a 130x jump over the 3.97MB output.
- Inference runs continuously, so storage spend grows about 25% a year and outpaces planning assumptions.
- Watch whether hyperscalers tie storage tiers to inference logs as McKinsey’s 35% CAGR demand arrives.
Why it matters
Watch whether hyperscalers tie storage tiers to inference logs as McKinsey’s 35% CAGR demand arrives.
This GenAI News article was prepared in original wording using reporting and materials published by MIT Technology Review AI. Source reference: https://www.technologyreview.com/2026/10/22/1146948/ais-hidden-storage-bill-what-inference-leaves-behind/.
Drafted by the GenAI News review pipeline.