AITech Interview with Jeronimo De Leon, Senior Product Manager of AI, Backblaze

Why It Matters

Effective storage transforms data from a bottleneck into a growth engine, directly influencing AI speed, cost, and compliance. Companies that align storage strategy with MLOps gain a competitive edge in the rapidly evolving AI market.

Key Takeaways

• AI success hinges on accessible, high‑performance storage.
• Cost, latency, and governance are top storage challenges.
• Backblaze B2 offers zero egress, scaling to petabytes.
• Openness enables rapid model iteration and reduces latency.
• Future AI will shift from text to video, demanding storage.

Pulse Analysis

Modern AI pipelines treat data as a living asset rather than a static dump. From raw ingestion to model inference, cloud storage determines throughput, training speed, and the ability to iterate quickly. Organizations that invest in high‑performance, low‑latency storage reduce bottlenecks that otherwise inflate compute costs and delay product releases. Moreover, unified, searchable archives simplify governance and enable consistent data quality, which directly improves model accuracy. As AI models grow in size and complexity, the storage layer becomes the primary lever for scaling both performance and cost efficiency.

Backblaze positions its B2 service as an open, S3‑compatible platform that removes traditional cloud friction. Zero egress fees and petabyte‑scale capacity let customers such as Decart AI move 16 PB in ninety days without paying for outbound traffic, delivering tenfold efficiency gains. The platform’s emphasis on fast indexing, metadata tagging, and fine‑grained permissions turns a passive bucket into an active data lake, accelerating training cycles and real‑time inference. By coupling cost transparency with predictable low latency, Backblaze enables enterprises to align storage spend with AI product roadmaps rather than reacting to surprise bills.

The next wave of AI will migrate from text‑only models to multimodal video engines, exploding data volumes by orders of magnitude. Video combines visual, audio, and temporal dimensions, requiring storage systems that can ingest terabytes per hour while preserving metadata for downstream training. Providers that deliver seamless scalability, instant retrieval, and built‑in compliance will become strategic partners rather than mere vendors. Companies that embed storage strategy early in their MLOps stack will capture richer datasets, shorten time‑to‑market for generative video applications, and safeguard the data assets that future models will continuously relearn.