Backblaze and Dimensional Research have released new research pointing to surprise charges on enterprise cloud storage strategies. With rising egress and retrieval fees shaping how data is stored, accessed, and moved for AI workloads. The company previewed similar findings for media and entertainment earlier in October, where the study reported that 93 percent of organisations encountered unexpected cloud storage fees, often monthly, and that billing opacity forced workflow trade offs.
Backblaze positioned the work as part of a push for pricing transparency and open cloud architectures that reduce lock in. The broader survey underpinning the latest release drew on responses from hundreds of leaders managing large scale environments, suggesting the concern extends beyond a single vertical. Such economics now influence architecture as much as performance. This is a cost of data mobility, not just storage.
AI Workloads Amplify Data Movement
As AI adoption scales, enterprises shuttle vast datasets from object stores to GPU clusters and back, multiplying the number of billable data movements that can trigger provider fees. Each step in the pipeline, from feature generation to validation to model deployment, risks cost leakage if data crosses regions, clouds, or the public internet.
Providers are responding at the margins, particularly on exit. “Cloud switching just got easier,” said Amit Zavery, noting that Google Cloud now waives network data transfer when customers migrate off the platform. That change removes one barrier to portability. It does not resolve day to day egress exposure inside multi cloud AI workflows.
Hyperscalers Adjust Exit Charges
AWS has also updated its posture, offering 100 GB per month of data transfer out under its free tier and a process to waive data transfer fees for customers moving away, with additional provisions tied to the EU Data Act. These policies align with regulators’ emphasis on switching and interoperability while leaving intact most ongoing egress, inter-zone, and inter-region charges that accumulate during normal operations.
For architecture teams, the message is mixed, exit is cheaper, but the meter still runs when applications span zones, regions, and clouds. Procurement needs to price the run state, not just the switch. That requires mapping data paths and quantifying typical monthly movements, then contracting accordingly with either allowances or capped rates.
Procurement Levers To Restore Control
Pragmatically, buyers are leaning on three levers to contain the hidden cost of object storage, placement, peering, and policy.
First, align data locality with compute by favouring same region execution and private interconnects to suppress chargeable egress, a basic but often neglected design control.
Second, embed portability in the contract, mandating exit fee waivers, predictable data transfer tiers, and explicit allowances for bursty AI training periods, reflecting that data gravity moves with GPU scheduling.
Third, formalise switching procedures, including initiation and migration windows, so that negotiated credits apply during real cutovers rather than expiring mid move. Google has documented such exit programmes, detailing notice, initiation, and migration periods that trigger credits for leaving customers, a model others are mirroring. The survey’s through line is clear, pricing transparency is a competitive feature for storage.
Hyperscalers Adjust Exit Charges
Backblaze is using the moment to differentiate on predictable fees, pitching open cloud workflows and, for high throughput use cases, offerings designed to decouple performance from egress penalties.
The company has argued that opaque network charges throttle media and AI economics even when storage pricing looks low on paper. “Production and distribution workflows are being throttled not by technology, but by cloud economics,” said Gleb Budman, tying budget predictability directly to content availability and data reuse.
As data mobility becomes a first order cost, competitive dynamics will hinge on transparent, portable storage contracts as much as on raw performance claims.