How real-time content velocity and modern storage reshape intent quality, AI scalability, governance, and performance across the data lifecycle.
Jeronimo, you’ve built a career at the intersection of AI and product management. What initially drew you into this space and ultimately to your role at Backblaze?
I was first drawn to AI while working on IBM Watson projects, where I learned about neural networks and how they mimic the way brain neurons work. That experience sparked my interest in how data shapes intelligence and showed me that managing and cleaning data is critical to AI accuracy. Over time I saw how many companies struggled not because of the AI itself but because their data was inaccessible, fragmented or inconsistent, which led me to focus on data quality and storage. That perspective ultimately brought me to Backblaze, where the mission to make cloud storage simple, affordable, and reliable aligns with my view that AI starts with storage and that effective data management enables companies to unlock insights and innovation.
AI innovation depends heavily on data infrastructure. Where do you see cloud storage making the biggest difference across the AI lifecycle, from ingest to inference?
Cloud storage is one of the most critical aspects of the AI lifecycle, acting as the foundation for speed, scale, and continuous improvement. It enables systematic aggregation, cataloging, and securing of data archives to accelerate new projects and simplify the testing of new models.
During training, scalable cloud storage ensures high throughput access to massive datasets and reliable storage for model checkpoints and weights. At the inference stage, it serves models and captures generative outputs along with logs and user feedback for evaluation and monitoring, fueling continuous iteration and improvement. Cloud storage turns data and models from static resources into actively managed artifacts that accelerate innovation throughout the AI lifecycle.
Scaling AI places unique demands on infrastructure. What are the most pressing storage challenges organizations encounter when they try to grow their AI initiatives?
As organizations try to scale their AI initiatives, some of the most challenging aspects of storage they encounter are cost, data management, and accessibility. It’s not enough to store large volumes of data; the data must be organized, easily retrievable, and governed with proper controls to be truly useful.
Having clean and well-structured data is vital for organizations looking to innovate in AI.
Another significant challenge is that data is frequently fragmented across silos and legacy systems. This fragmentation can create bottlenecks that slow AI growth. Successful organizations address these challenges by treating storage as a strategic enabler, building systems that scale in terms of cost efficiency, performance, and accessibility alongside their evolving AI maturity.
Backblaze emphasizes open, flexible access to data. Why is this openness so critical for AI teams compared to traditional cloud models?
Openness and flexibility in data access are crucial for AI teams, as they transform storage from a passive repository into an active enabler of innovation. Smart archiving and centralizing information into a structured, searchable archive that unifies diverse formats, normalizes and tags data for consistency, and enables fast indexing and querying. This level of openness is essential because it lays the groundwork for efficient analytics and modeling.
When data is easily accessible and well-organized, AI teams can move faster, experiment more freely, and reduce latency in both training and inference cycles. Backblaze’s emphasis on open, flexible access ensures that data is not only stored but also made immediately usable, which is precisely what today’s AI innovations demand.
Cost efficiency is a recurring concern for enterprises training and fine-tuning AI models. How does the right storage strategy help control expenses without slowing progress?
The correct storage strategy helps control expenses without slowing progress by focusing on long-term scalability and alignment with product evolution. Since collecting, processing, moving, and running inference on data are all core activities that impact both performance and cost, organizations need to design their storage early with these workflows in mind.
Failing to plan leads to compounding costs and growing infrastructure complexity as data volumes increase. By building storage solutions that balance high performance with cost efficiency from the start, organizations ensure their AI initiatives can scale smoothly. This approach prevents bottlenecks and rising expenses from slowing down innovation, allowing teams to maintain rapid development and deployment even as data demands grow.
Latency can make or break AI-driven applications. What approaches are most effective in ensuring storage systems keep pace with real-time processing needs?
To keep pace with the demands of real-time processing, storage systems must be designed with latency as a top priority. Among challenges such as cost, security, and compliance, latency is often the most critical barrier, especially during model inference, where even slight delays in serving predictions can negatively impact the user experience and slow adoption.
Organizations can reduce latency by locating storage close to compute resources, using high-performance storage optimized for rapid read/write access, and streamlining data pipelines to eliminate bottlenecks. It’s also important to choose storage providers that offer predictable, low-latency access without imposing high egress costs.
While startups typically focus on reducing latency and managing costs, enterprises must also navigate governance and regulatory requirements. The most effective storage strategies strike a balance between low-latency performance, cost efficiency, and compliance, enabling AI systems to scale and respond in real-time.
Security and compliance remain non-negotiable in today’s landscape. How should organizations think about balancing regulatory requirements with innovation in AI workflows?
Balancing regulatory requirements with innovation in AI workflows starts with treating governance as a core element of storage architecture. A strong foundation for managing, securing, and auditing data is essential, not just to meet compliance standards but also to build trust in AI systems.
Modern cloud storage is evolving to support this balance by offering built-in controls like encryption by default, fine-grained permissions, audit trails, and data residency options. Equally important is data lineage, which involves knowing where data originated, how it was processed, and how it is used to inform AI models. This transparency is crucial for ensuring regulatory compliance and responsible AI practices.
Today’s storage platforms are focused on improving usability, enabling teams to move quickly without compromising control. When governance, data lineage, and accessibility are integrated, organizations can meet their compliance obligations while still accelerating AI innovation at scale.
You’ve worked closely with customers like Decart AI and Wynd Labs. What are some of the most compelling use cases you’ve seen that highlight storage as an enabler for AI?
For Decart AI, the challenge was all about efficient model training. They needed to move massive datasets quickly to utilize their computing resources. With Backblaze B2, they scaled to 16PB in just 90 days, trained across multiple GPU clusters, and achieved this with zero egress costs. They achieved 10 times the efficiency of their competitors, freeing their team to focus on pushing the boundaries of their models instead of wrestling with infrastructure.
Additionally, Wynd Labs utilized Backblaze’s high performance and free egress, enabling them to meet enterprise-scale demands while seamlessly reinvesting savings into product development. That level of scalable, cost-effective data access unlocked entirely new opportunities for their platform. In both cases, cloud storage wasn’t just a backend component; it was a strategic enabler that turned infrastructure from a bottleneck into a launchpad for innovation.
Many enterprises are still early in aligning storage strategies with MLOps architectures. What steps do you recommend they prioritize to avoid bottlenecks down the line?
The key is to start with compatibility and scalability in mind. One of the most effective first steps is selecting storage that integrates seamlessly with existing MLOps and compute stacks, as Backblaze B2 does through its S3 compatibility, which eliminates the need for a full re-architecture. We recommend starting with a proof of concept to validate key factors, such as ease of migration, performance, and integration.
This helps identify and resolve potential friction points early. Once that foundation is in place, the focus should shift to optimizing throughput, data movement, and orchestration. These are the elements that enable teams to train across clusters, run inference, and iterate quickly, without being hindered by storage-related bottlenecks. Prioritizing these steps early helps organizations build a storage layer that scales efficiently alongside their AI and MLOps maturity.
Looking ahead, how do you see the relationship between cloud storage and AI evolving, and what excites you most about the possibilities that are emerging?
We are in the age of large language models today, but I believe in the future video will become the new baseline for AI. As models shift from text to richer forms of content, the amount of data generated and stored will grow exponentially, since video combines images, audio, motion, and context in ways that demand far greater scale. Each new wave of generation produces more material that can be stored, reused, and retrained, creating a cycle where storage is central to advancing what models can do. What excites me most is seeing how this transition from text to video will expand the boundaries of AI, world simulations and make storage even more essential to the innovation loop.
A quote or advice from the author:- The real advantage in AI will come from understanding the value of your data and preserving it so future models can learn from it.
Jeronimo De Leon
Senior Product Manager of AI, Backblaze
Jeronimo De Leon is a seasoned product management leader with over 10 years of experience driving AI-driven innovation across enterprise and startup environments. Currently serving as Senior Product Manager, AI at Backblaze, he leads the development of AI/ML features, focuses on how Backblaze enhances the AI data lifecycle for customers’ MLOps architectures, and implements AI tools and agents to optimize internal operations.