Industry executives and experts share their predictions for 2026. Read them in this 18th annual VMblog.com series exclusive.
By Gleb Budman, CEO and co-founder of Backblaze
As we move toward 2026, the AI industry is approaching a critical inflection point at which raw horsepower alone will no longer guarantee a competitive edge. The industry is maturing into a more disciplined, architecture-forward phase where the success of a model depends less on the silicon it runs on and more on the infrastructure that feeds, protects, and audits it.
From the rise of specialized neoclouds to the mandatory adoption of immutable data pipelines, the next frontier of AI isn’t just about how fast we can compute; it’s about how intelligently we can manage the massive data flows that enable that computation. Below, I’ve shared six predictions for 2026 that will reshape how organizations build, scale, and secure their AI futures.
Prediction #1: Neoclouds stop being “GPU clouds” and become full AI workflow clouds.
In 2026, the fastest-growing cloud category won’t be hyperscalers or GPU clouds-it will be “neoclouds” that handle datasets, movement, feature stores, fine-tuning and inference. AI traffic is 10� more intense than traditional workloads and increasingly dominated by fewer, heavy endpoints. Providers that solve data ingest, caching, egress freedom, and training data pipelines (not just raw GPUs) will capture the next wave of AI spending.
Prediction #2: Immutable AI artifacts become a must-have selling point.
AI audits, safety reviews, and regulations require teams to store training corpora and production model checkpoints in immutable, WORM-protected storage, where training data and checkpoints are locked by default. As the EU AI Act and enterprise governance rules require reproducibility and provenance, object lock (WORM) becomes the easiest way to prove “this is exactly the data and model we deployed.” As a result, immutable AI data pipelines become a selling point for AI vendors to demonstrate the trustworthiness of their models.
Prediction #3: 50% of new marketing, training, and social content containing AI-generated scenes or actors.
2026 is the year when AI video moves from “demo” to “daily workflow.” In 2025, models finally hit quality thresholds, and creators shifted to storage-first media stacks. Creative teams will start looking for reliable egress and object storage pipelines to finally integrate AI video into rendering, editing, localization, and personalization at scale.
Prediction #4: Data movement will explode AI budgets.
The biggest AI cost surprise won’t be GPUs going forward, but the shift in datasets, prompting CFOs to require “egress clauses” in every AI contract. 95% of orgs already report surprise cloud fees; AI is only accelerating that. Flow-heavy AI pipelines (training refreshes, retrieval-augmented inference) create massive, expensive churn. Teams shift to egress-friendly, transparent storage providers in response.
Prediction #5: Multi-cloud AI pipelines become the default-with storage as the neutral control plane.
By late 2026, most serious AI teams run models across 2-4 GPU providers while keeping their data anchored in a neutral storage cloud they control. Crucial to maintaining this multi-vendor, best-of-breed approach will be terabit ingest and free egress architectures, making it rational to burst to whichever GPU fleet is cheapest or fastest. Storage becomes the home base, and compute becomes ephemeral and interchangeable.
Prediction #6: AI observability becomes an SRE mandate, including storage and data lineage.
In 2026, AI teams treat storage telemetry, network flow magnitudes, dataset versions, and feature drift as core SRE dashboards-not “nice to haves.” Customers are going to start demanding visibility into data freshness, dataset access patterns, I/O bottlenecks, and training data provenance. This becomes essential to ship reliable AI systems.
Why Data Infrastructure Will Decide AI Leaders
Bigger models and faster GPUs won’t define the future of AI. The real advantage lies in infrastructure, the systems that make AI reliable, auditable, and cost-effective at scale. Organizations that build storage-first architectures, immutable data pipelines, multi-cloud flexibility, and end-to-end observability won’t just control costs; they’ll also earn the trust of regulators and customers and set the industry standard.
At Backblaze, we see this shift clearly. The next wave of AI innovation will run on systems that seamlessly move, protect and track data. Infrastructure will be the ultimate differentiator in the AI era.
##
ABOUT THE AUTHOR
Gleb Budman is a co-founder and has served as our chief executive officer since 2007, guiding the business from its inception in a Palo Alto apartment to a company serving customers in more than 175 countries with over an exabyte of data under management. Gleb has served as a member of our board of directors since 2009 and as chairperson since January 2021. Prior to Backblaze, Gleb was the senior director of product management at SonicWall and the vice president of products at MailFrontier, which was acquired by SonicWall. Before that, he served in a senior position at Kendara, which was acquired by Excite@Home, and previously founded and successfully exited two other startup companies.
