AWS quietly doubled EBS throughput and IOPS ceilings on its latest-generation EC2 instances this year as part of a broader block-storage push shaped by AI workloads, with no change to per-GB storage pricing. The catch, per independent analysis, is that capturing the gain means moving to newer, more expensive instance families — making this more of an upsell lever dressed as a free upgrade than a straightforward price improvement for existing fleets.

Direct VPC Network Peering between AWS and Google Cloud has been generally available since April 16, letting multicloud storage and data traffic move between the two providers without the public internet or full egress pricing. Microsoft has said Azure will join the open interoperability effort later this year, which would make direct low-cost interconnects the norm across all three hyperscalers rather than a two-provider exception.

While Azure and Google spent 2026 rolling out AI-optimized tiers and architecture rewrites, AWS’s core gp3 economics haven’t moved: still 20% cheaper per GB than gp2, with 3,000 IOPS and 125 MB/s included free. It remains the cheapest general-purpose baseline of the three clouds and the reference point worth checking any new AI-tier pricing against before assuming it’s a better deal.

New legacy blob storage account creation is already disabled as of this month, and any legacy or GPv1 account still standing on October 13 gets automatically converted to general-purpose v2 on Microsoft’s terms — including its default tiering and redundancy choices — which Microsoft’s own documentation warns can raise the bill. Databricks-managed DBFS accounts are handled automatically, but every other legacy account needs proactive migration through the Azure Policy or portal path in the next five weeks to keep control of cost and timing.

The new fully managed service gives enterprises a single view across their Azure Blob Storage estate — useful for exactly the kind of account inventory the legacy-account retirement above now requires, and a sign Azure is investing in governance tooling alongside raw capacity. Combined with Azure’s continued build-out of per-blob tiering and lifecycle automation, the emphasis this year is as much on managing sprawl as on adding new storage tiers.

Microsoft is positioning Premium Blob Storage’s low-latency, high-throughput profile explicitly around AI workloads, pairing it with tools like the LangChain Azure Blob Loader, which the company says runs up to 5x faster than prior community implementations. It’s a continuation of the same theme as Azure’s Foundry integration from prior weeks: Blob is being marketed as AI infrastructure, not a commodity object store competing purely on $/GB.

The new Hyperdisk tier upgrades raise IOPS ceilings and throughput specifically for AI training and inference workloads, extending Google’s block-storage line further into the same AI-capacity race AWS and Azure are running with their own 2026 announcements. Combined with the AI-optimized object storage tiers shown at the same event, Google is making the case that GCP’s storage stack, not just its TPUs, is a first-class part of its AI pitch.

Enhanced storage in Google Distributed Cloud now supports 6PB of object storage per zone — six times the prior limit — and 30 IOPS per GB per zone, a 10x performance increase, extending Google’s AI-capacity buildout into its on-prem and edge distributed offering rather than public cloud alone. For enterprises running GDC for data-residency or latency reasons, this closes much of the capacity gap that used to push AI-heavy workloads back to public GCP.

Alongside the AWS-Google Cross-Cloud Interconnect going GA, Google introduced caching that stores cross-cloud data locally to reduce egress charges and speed up queries that touch AWS or Azure data — a direct answer to one of the more persistent complaints about multicloud analytics cost. It’s a incremental feature next to the interconnect itself, but a useful sign that the egress-fee side of multicloud economics, not just network path, is now getting product investment.

Amazon is guiding to roughly $200B in 2026 capex (versus $125B in 2025), Google to $175–185B (versus $91B), and Microsoft to $110–120B (versus $90B), with the combined four-hyperscaler total climbing to $720–745B after the latest earnings calls. Roughly $450B of that is tied directly to AI infrastructure — servers, GPUs, datacenters — rather than traditional cloud capacity, and hyperscalers are increasingly using debt financing to bridge the gap between AI capex and free cash flow.

None of AWS’s, Azure’s, or Google’s 2026 block-storage moves — Hyperdisk upgrades, forced Unmanaged-to-Managed Disk migration, new Blob tiering — carry a headline price hike, but the argument is that migration complexity and default-tier mismatches quietly push customers toward overprovisioning. With enterprise storage waste already estimated near 70% of provisioned capacity, the practical advice is to treat any forced 2026 storage migration as a cost-review trigger rather than a routine technical task.

The cloud storage market is projected to grow from roughly $179.26B in 2026 to $513.86B by 2031, with AWS holding about 31% share, Azure 25%, and Google 11% of the broader infrastructure market. That growth rate is the backdrop for this year’s tier repricing and fee restructuring across all three providers: demand isn’t close to plateauing, so providers have room to keep experimenting with pricing structure rather than compete purely on headline $/GB.