The story of Backblaze B2 has always been the same: fraction-of-S3 pricing, S3-compatible API, no surprises on the bill. That pitch worked well enough to build a real business. But in 2026, Backblaze is making a harder argument: B2 isn't just where you park cold data to save money. It's where you run AI pipelines, media archives, and application storage at scale, without your cloud bill becoming a negotiation.

The evidence is stacking up. B2 is priced at approximately $0.007/GB/month for storage, compared to AWS S3's standard rate of roughly $0.023/GB/month. That's not a rounding error. That's roughly one-third the cost for compatible object storage semantics. Combined with a pricing refresh that effectively eliminates or drastically reduces API transaction costs, and explicit positioning around AI/ML workloads, the message is clear: Backblaze is done being the budget option and is positioning B2 as the default object store for data-intensive teams who don't want to fund Amazon's infrastructure margins.

This isn't just marketing. The financial signals back it up.

And the enterprise wins are materializing too:

An eight-figure TCV deal from a company most engineers still associate with personal backup is the signal worth paying attention to.

The most underreported part of Backblaze's current positioning is the transaction fee structure. AWS S3 charges per 1,000 requests: PUT, COPY, POST, and LIST operations run at $0.005 per 1,000; GET and other operations at $0.0004 per 1,000. That sounds trivial until you're running a feature store that reads millions of small objects, a media pipeline ingesting thousands of files per hour, or an AI training loop checkpointing to object storage every few minutes. At that point, API costs stop being a rounding error and start being a line item that finance asks about.

Backblaze has moved to make many Class A and Class B API operations free or effectively free up to a multiple of stored data. The practical effect: teams that are API-call-heavy, not just storage-heavy, can model their costs with confidence. No surprise bills from a burst ingestion job. No penalty for building systems that make aggressive use of the storage API. Cloudflare R2, for context, also eliminates egress fees and offers zero-cost API operations, but its storage rate is higher than B2. Wasabi sits in a similar cost band to B2 but has historically imposed minimum storage duration requirements that create their own cost traps for short-lived data. Backblaze's combination of low per-GB pricing, reduced API costs, and S3 compatibility puts it in a distinct position: cheaper to store, cheaper to operate, and compatible with existing tooling.

The table above illustrates the tradeoffs. Backblaze isn't the cheapest on every dimension in every scenario. But for teams already paying egress through a Cloudflare or Fastly partnership, or running workloads where API volume is high and data doesn't move out constantly, B2's overall TCO lands lower than almost any alternative that also supports S3-compatible tooling.

Backblaze is explicitly marketing B2 for AI and ML workloads, including the ability to process exabyte-scale datasets and move data into any GPU environment. Skeptics might read that as a marketing team chasing buzzwords. But the underlying logic is sound, and most coverage is missing why. When GPU compute is expensive and increasingly spot-priced across multiple clouds, storage becomes the coordination layer. Your model checkpoints, training datasets, feature parquet files, and embedding snapshots need to live somewhere that isn't coupled to a specific GPU vendor or cloud region. If that storage is on AWS, you pay egress every time you run a training job on Lambda Labs, CoreWeave, or a spot instance elsewhere. If it's on B2, the economic and architectural incentives shift. The pattern this enables: decouple your durable object store from your compute environment entirely. Run GPUs wherever they're cheapest on any given day. Keep training artifacts, intermediate outputs, and model weights in a single B2 bucket. Use S3-compatible SDKs (Boto3, the AWS SDK, Terraform's S3 backend) with nothing more than an endpoint override. This isn't a theoretical pattern. It's the architecture that cost-conscious AI infrastructure teams are already converging on. Backblaze's pricing refresh and explicit AI positioning mean they're ready to support it at scale, not just as an afterthought. What teams need to think carefully about: this architecture shifts complexity into data orchestration, consistency guarantees, and cross-cloud observability. Cheaper storage doesn't simplify security. You still need encryption at rest and in transit, IAM-equivalent access controls, audit logging, and multi-cloud monitoring that captures object storage latency and failure events. Teams that swap S3 for B2 and assume everything else stays the same will run into operational surprises. The storage bill drops; the governance investment shouldn't.

The hyperscalers are not going to match B2's economics. That's not a prediction; it's a structural observation. Egress fees and per-request charges are core to how AWS, Google Cloud, and Azure monetize data that moves between services and out to customers. Lowering those fees across the board would cannibalize meaningful margin. Instead, they'll compete on lock-in: tighter integration with Bedrock, Vertex AI, and Azure AI means that if your model training, inference, and storage all live inside one cloud, switching any piece of it carries real friction.

That's the actual competitive battle in 2026. It's not storage cost per GB. It's whether the convenience and integration of staying inside one hyperscaler's ecosystem is worth the premium. For startups, independent AI labs, and engineering teams that want to keep optionality, the answer is increasingly no. Backblaze's value in that context isn't just cheap bytes. It's a credible, enterprise-ready storage layer that doesn't require you to choose a cloud. The eight-figure TCV neocloud deal announced in Q4 is a proof point that enterprises are reaching the same conclusion, not just indie developers looking to cut costs.

If you're running any of these workloads, run a pilot now:

The migration path is low-friction for most teams. S3-compatible SDKs work with an endpoint change. Terraform modules for S3 buckets require minimal modification. Tools like Rclone support B2 natively for data transfer and sync operations. If you're already using Boto3 or the AWS CLI, testing against B2 is a one-afternoon exercise. What you should not skip:

Backblaze is not trying to out-feature AWS. That's not the bet they're making and it's not the one engineering leaders should evaluate them on. The bet is simpler and more durable: if you need S3-compatible object storage and you're not getting value from deep hyperscaler integration, you're probably overpaying by a factor of two to three, and you're accepting pricing complexity that makes cost forecasting harder than it should be. The pricing refresh that eliminates most API transaction costs, combined with sub-cent storage rates and genuine AI/ML positioning, pushes B2 past the "cheap backup" category for good. The 24% YoY revenue growth and the first enterprise eight-figure deal suggest the market is catching up to what infrastructure-savvy teams have known for a while. The forward-looking question isn't whether B2 can replace S3 for every workload. It can't, and Backblaze isn't claiming otherwise. The question is whether you've done the honest TCO analysis for your data-intensive workloads and whether your current storage architecture is as cloud-agnostic as your engineering strategy claims to be. For most teams, the answer to both is no. That's the gap B2 is built to close.