The cloud storage market crossed roughly $173 billion in 2026. Object storage is growing faster than any other storage type in that market. And underneath both numbers, the ground is shifting: egress fees are facing a regulatory deadline, AI demand is still mostly ahead of us, and the open-source project a generation of engineers built their storage stacks around just went dark.
The cloud storage market crossed roughly $173 billion in 2026, growing at a 17.1 percent compound annual rate toward an estimated $380 billion by 2031, according to MarketsandMarkets. Object storage specifically is projected to grow faster than any other storage type in that market, at a 19.1 percent CAGR, driven by AI training data, backup and archive volume, and the steady migration away from on-premises file and block storage.
Those are the headline numbers. They're also the least interesting part of the story. Underneath a market that looks like simple, steady growth, five real shifts are reshaping how storage actually gets bought, priced, and architected in 2026. This report pulls them together.
Egress fees, long the industry's most reliable lock-in mechanism, are facing their first real regulatory reckoning. The EU Data Act, in force since January 2024 and enforceable since September 2025, will prohibit cloud providers from charging any switching fees at all, including egress charges incurred during a provider switch, starting January 12, 2027. AWS, Azure, and Google Cloud have already partially moved in this direction, waiving egress fees for full account exits since 2024, though ongoing multi-cloud egress remains outside the scope of the ban.
The pricing gap the ban is responding to remains wide. AWS charges $0.09 per GB for standard internet egress, Azure charges $0.087, and Google Cloud charges $0.12, the highest of the three. Cloudflare has documented that these rates can represent a markup of thousands of percent over wholesale transit costs. Meanwhile, Flexera's 2026 State of the Cloud data puts wasted IaaS and PaaS spend at 29 percent industry-wide, and egress and API fees are consistently among the largest unbudgeted line items driving that waste. This connects directly to what your cloud provider isn't telling you about hidden billing costs.
The practical effect in 2026: buyers increasingly treat egress exposure as a first-order architectural decision rather than a footnote, and zero-egress providers like Backblaze, Wasabi, and Cloudflare R2 have moved from niche alternative to default consideration for any workload with meaningful data movement.
The single most misunderstood fact in cloud storage right now: the AI-driven demand everyone is already straining under represents a fraction of what's coming. McKinsey's Global Survey on AI found only 28 percent of enterprises have deployed AI in production at scale, meaning across multiple functions with measurable impact. The other 72 percent are still in pilot or proof-of-concept. Forrester and Anaconda separately found 88 percent of enterprise AI agent pilots never reach production at all.
What's already straining storage and GPU capacity in 2026 is a preview, not the main event. Production AI workloads consume dramatically more compute and storage than pilots, particularly agentic systems, which can make dozens of model calls and memory lookups to complete a single task. A production agent's storage footprint, spanning short-term working memory, episodic history, semantic knowledge, and procedural instructions, grows continuously in a way pilot-stage systems never do. Vector database costs compound the same way: real production bills are running 2.5 to 4 times higher than pricing-page estimates once egress, index rebuild compute, and volume growth nobody budgeted for are counted.
Data gravity, the tendency of large datasets to resist being moved, has become the connective constraint across all of this. Gartner estimates organizations already spend 10 to 15 percent of their total cloud bill on egress alone, a number that scales directly with how much AI infrastructure sprawls across multiple storage systems and compute providers chasing available GPU capacity. This is the same dynamic covered in why storage functions as the anchor of the AI infrastructure stack.
Two forces are pulling the storage vendor landscape in opposite directions at once.
In February 2026, MinIO's open-source Community Edition was archived on GitHub, no further patches, no new features, the company having shifted focus toward a commercial product for AI workloads. MinIO had been one of the most complete third-party S3-compatible implementations. Its exit reinforced the exact lesson it created: build against the S3 API standard, not against any single vendor's implementation, because compatibility is a spectrum and even category leaders can change direction without warning.
Neocloud revenue is projected to reach roughly $20 billion in 2026 and approach $180 billion by 2030, and the category now spans everything from CoreWeave, an IPO'd company with tens of billions in hyperscaler contracts, down to brokers reselling capacity they don't own. Ownership, whether a provider actually owns the GPUs and storage it rents out, has become the single highest-leverage vetting question for buyers navigating this landscape, since it directly predicts support quality, pricing transparency, and exposure to a provider's own financing risk.
Cloud repatriation hit its highest recorded rate in 2026: 86 percent of CIOs now plan to move at least some workloads off public cloud, per the Barclays CIO Survey. The nuance matters as much as the headline: only 8 to 9 percent plan a full exit, according to IDC. This isn't a retreat from cloud, it's the end of cloud-first as an unexamined default. Gartner projects 90 percent hybrid infrastructure adoption by 2027, and the workloads actually moving back on-premises follow a consistent pattern: steady-state compute where cloud elasticity goes unused, continuous AI training and inference running at the most expensive available instance tiers, and data with tightening sovereignty requirements.
Public cloud spending is still growing overall even as this rebalancing accelerates, since new AI-driven workloads are being created faster than existing ones are being repatriated. Both trends are symptoms of the same underlying shift: buyers making deliberate, workload-by-workload placement decisions instead of defaulting to whichever platform they started on.
The market is growing, the rules are changing, and the biggest wave of demand still hasn't arrived. Mid-2026 is not the peak of this story, it's the setup.
Compare AWS, Google Cloud, Azure, and alternatives like Backblaze B2 Discover how much you could save in seconds
