Why Every Neocloud Needs a Cloud Storage Strategy

Marty Falaro
Executive Vice President and Chief Operating Officer

Most AI infrastructure conversations are about GPUs. Who has capacity? Who’s leasing it? Who’s building the next cluster? As a storage guy, I find that a bit annoying. But I get it. When it comes to scaling AI workloads, compute has been the predominant constraint, and it still is in a lot of places.

But what happens to the data once the training run ends, the inference pipeline is live, or the model needs to pull from a growing archive of unstructured data? That's a storage problem. And increasingly, neoclouds and other AI platform providers are seeking to partner with experienced, independent cloud storage providers like Wasabi, who have already delivered cloud object storage at scale globally, rather than trying to solve it themselves. 

The shift toward composable AI infrastructure

Over the past couple of years, we've watched a real shift in how the best infrastructure providers approach AI. Instead of trying to build every layer themselves, they're assembling best-of-breed technologies (network, compute, storage) into a single, integrated platform so their customers get one seamless experience instead of a pile of vendors to manage on their own.

This is the model many neoclouds are embracing: infrastructure purpose-built for AI workloads brought together by a provider willing to partner for the pieces that aren't their core strength.

We see this model play out across our own partner ecosystem. It’s not a trend we’re reacting to, rather a model Wasabi was built for from day one. We don’t sell compute, so we don’t compete with the neoclouds the way the hyperscalers do. We do one thing and one thing better than anyone else: S3 cloud object storage, priced simply and predictably, with no penalty for accessing or moving your data. We’ve announced and continue to build out new classes of storage to support these broad, AI-centric workloads.

The hidden cost of cloud object storage for AI workloads

Opaque billing practices and hard-to-calculate storage fees that extend beyond raw capacity are well-known problems in the industry. For several years running, our Cloud Storage Index surveying IT decision-makers has consistently found that fees for API requests, egress, and other transaction fees make up 50% of the average cloud storage bill. For AI workloads, I expect that percentage to run even higher.

That's because AI workloads move constantly. Training jobs checkpoint to storage over and over as they run. RAG applications go back to object storage on every single query, pulling context from a document store or vector database. Data gets replicated across regions, handed off between compute providers, and pulled into different tools throughout its life.

Every one of those steps generates its own API call. If you store your data with a hyperscaler, each one of those API calls comes with an API fee. And every time data moves out of their environment to another provider, it triggers egress fees on top of those API charges.

More movement means more fees, and AI workloads move more than almost anything else we store. For years, the hyperscalers have been the default answer for cloud object storage, but their fee structures are exactly what make moving data unaffordable at scale.

Megaport was up against this very problem for standard object storage, and they solved it with Wasabi. 

Wasabi powers Megaport’s standard object storage layer

Megaport is a leading global automated infrastructure platform. They've built a global, software-defined network and, through their acquisition of Latitude.sh, added GPU-powered compute—real building blocks for an AI infrastructure platform, and storage. What they needed was a standard object storage partner that shared their vision of an open, interconnected cloud ecosystem built for performance and affordable scalability that’s not designed to lock customers in.

That's what this partnership delivers. Megaport is extending high-speed private connectivity and compute services to Wasabi's 16 global storage regions, so customers can provision and access dedicated standard object storage in minutes seamlessly from Megaport's platform. They’ll experience the same affordability and cost predictability that Wasabi customers already expect, no matter how much data grows or how often it moves. 

The future of AI storage and neocloud infrastructure

Megaport is a great partnership, and one we are very happy to be able to share publicly. It’s also just a preview of where we think infrastructure is heading: more providers assembling best-of-breed stacks instead of building (or buying) everything themselves. We expect to see more of these partnerships take shape, and we intend to be the AI storage layer of choice when they do.

We've recently started describing Wasabi as the AI storage cloud. Partnerships like this one are why: not because we do everything AI infrastructure needs, but because we've focused on doing the one part that every AI stack needs—fast, predictably priced, and secure cloud object storage. And we do it better than anyone else.

 Check out the Megaport-Wasabi press release for more details.