Blog Post

Azure Storage Blog
6 MIN READ

Blob Storage Turbo Tier: High performance with object storage scale and economics

BenedictBerger's avatar
Nov 17, 2026

High-performance object storage for AI, analytics, and HPC workloads: SSD-backed speed for active data and HDD-backed economics for the rest, all in a single Blob Storage account.

At Microsoft Ignite, we are introducing the preview of Turbo tier for Azure Blob Storage, built for AI and other data-intensive workloads that increasingly put storage on the critical path of compute. Training pipelines need to continuously feed data to GPUs. Inference services need fast access to models and application data as they start and scale. Analytics and HPC pipelines need high throughput across large datasets. 

Meeting those performance demands can require customers to choose between SSD-backed performance at high TCO and the scale and economics of HDD-backed object storage or incur operational overhead across separate storage systems to get both.

Customers no longer need to choose between high-performance storage and cost-efficient, scalable object storage. Turbo tier removes that tradeoff by combining higher performance for performance-demanding data with cost optimization for the rest, within one Blob Storage account. Applications continue using the same account, namespace, and Blob APIs.

Turbo tier combines these product capabilities:

  • Higher throughput, transactions/second, and lower latency. Turbo tier accelerates demanding I/O patterns, including highly concurrent reads and large transfers.
  • One Blob Storage account and namespace across performance tiers. Use Turbo alongside Hot, Cool, and Cold tiers without managing a separate high-performance storage account or moving objects between storage accounts.
  • Automated placement across tiers. Smart placement helps place performance-sensitive data in SSD-based Turbo tier as access patterns change, while less-active data can tier down to HDD-based Hot, Cool and Cold tiers.
  • Same API surface. Continue using the same Blob APIs, SDKs, and tooling your applications use today as eligible data moves between performance tiers.

The result is a different approach to object storage performance: bring high performance to the subset of data inside your Blob Storage account that requires it rather than moving an entire dataset into a separate SSD storage architecture.

One storage account, multiple performance tiers.

Traditionally, cloud object storage has required customers to choose upfront between performance-optimized storage and cost-optimized storage. That choice becomes limiting as application needs and access patterns change.

Azure Blob Storage Turbo tier removes that friction. Customers can apply higher performance to the data that needs it while retaining cost-efficient storage for the rest, all within one account, namespace, and Blob API surface. Your storage architecture stays familiar. The performance class of your data can change as workload demands change.

An AI training pipeline might repeatedly access a subset of images or model data across many epochs. Analytics pipelines may repeatedly process recent partitions of a much larger data lake. Inference applications can require fast access to active models and application data. In each case, only part of the overall dataset may need SSD-backed performance at any given time.

Turbo tier brings SSD-backed performance directly into a standard Azure Storage account. Applications continue to address blobs through the same account endpoints and namespace as data changes tiers. Existing Blob APIs remain the application interface. Turbo is an access tier within the storage account, rather than a cache or a separately managed high-performance storage account.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Smart placement: Performance follows workload demand

Smart placement removes the burden of continuously identifying, promoting, and cooling individual objects as access patterns change. Instead, you configure the size of your high-performance working set while Smart placement manages eligible data movement between Turbo and HDD-backed tiers.

You configure a Turbo capacity target directly on your storage account through the Azure Portal, Azure CLI, Azure PowerShell, or the management plane REST APIs. The capacity target helps govern how Smart placement manages your high-performance working set.

 

With Turbo tier, the data lifecycle becomes:

  • Land fast. When configured for Turbo tier, new data in your storage account will land directly in Turbo so performance-sensitive processing starts against SSD-backed storage.
  • Cool as demand changes. When Turbo performance is no longer needed, Smart placement moves eligible data back to HDD-backed access tiers, helping maintain the configured Turbo working set and reducing SSD-backed capacity spend.
  • Accelerate when workloads demand it. As workload access patterns change, Smart placement promotes eligible data from HDD-backed access tiers into Turbo tier.

If you know exactly which data requires Turbo, you can skip smart placement and directly orchestrate movement through explicit Blob tiering operations.

Performance across the I/O patterns of demanding workloads

Modern AI and data workloads stress storage in different ways, sometimes within the same pipeline.

  • Small, concurrent reads. Training workers, inference services, and analytics systems can generate large numbers of concurrent requests. Low latency and high transaction rates help keep storage from becoming the pacing layer.
  • Large transfers. Model loading, dataset movement, and checkpointing depend on sustained bandwidth, making storage throughput critical to how quickly data reaches compute.
  • Bursty demand. Worker startup, model loading, and checkpoints can rapidly increase storage demand, requiring storage performance to scale with concurrency.

We tested a synthetic 4 KiB random-read workload against the same dataset, using 72 Standard D96s v6 worker VMs for Standard Blob with HDD-backed tiers and 256 for Turbo tier. The Standard Blob configuration used fewer workers to stay within the HDD-backed storage account limits.

In this benchmark, Turbo tier delivered 16.8× the transaction rate, reaching 10.1 million transactions per second compared with 601,000 for Standard Blob. Results are specific to this test configuration. Performance varies by workload, object size, concurrency, network conditions, and account configuration.

Read bandwidth is another critical dimension of storage performance. During checkpoint recovery, large GPU clusters must reload model weights ranging from several GiB to tens of TiB. High read throughput helps reduce GPU idle time after a job interruption. It also helps reduce data-loading delays in analytics workloads and when scaling out inference endpoints. In this separate read-bandwidth benchmark, Turbo tier sustained more than 20 Tbps:

 

Where Turbo tier fits: Turning storage performance into workload performance

Higher storage performance can help reduce data-loading delays and checkpoint pauses when storage is a bottleneck. The benchmarks above measure storage performance, not end-to-end workload speedups. Turbo tier is designed for workloads in which storage latency, throughput, or transaction rates directly affect application performance and compute efficiency.

  • AI training and fine-tuning. Feed large training clusters with active datasets while supporting high-bandwidth checkpoint and intermediate-state writes.
  • AI inference. Provide fast access to models and other latency-sensitive application data as inference services start, scale, and serve requests.
  • Analytics and HPC pipelines. Accelerate repeated processing against the active portion of large datasets and data lakes.
  • High throughput reads, writes, and checkpoints. Sustain large output and checkpoint reads and writes so compute resources can return to productive work sooner.

Across these scenarios, the objective is the same: apply higher storage performance where it creates workload value without requiring every object in your dataset to carry the cost of SSD-backed storage.

Align storage spend and workload economics with performance demand

The benchmarks above demonstrate storage performance. Turbo also changes the economics of high-performance Blob Storage: You pay for SSD-backed capacity only for the data that needs it, rather than for the entire dataset.

In the customer workloads we have analyzed across diverse use cases, only a fraction of the overall dataset typically requires the highest level of performance. As an illustration, consider a 1 PiB dataset with an approximately 20% active, high-performance working set. Rounded to whole TiB, the example places:

  • 205 TiB of active data in Turbo tier
  • 819 TiB of remaining data in Hot tier

Data placement

Illustrative monthly capacity cost (in USD)

1 PiB all-SSD baseline

$115,343

20% Turbo tier + 80% Hot tier

$37,500

Illustrative capacity-cost reduction

67%

* The example is based on East US 2 and assumes a full month at the stated placement. The illustrated savings apply to capacity costs in this scenario and do not represent total workload-cost savings or a pricing or savings guarantee.

In this illustrative scenario, the mixed Turbo and Hot configuration results in approximately 67% lower capacity cost than the all-SSD configuration used for this comparison (1 PiB x $0.11 per GiB = $115,343).

This example highlights the core economic model behind Turbo: align SSD capacity spend with the part of the dataset where performance creates value, while continuing to use cost-efficient HDD-backed access tiers for the rest.

The capacity comparison captures only one part of the economic value. At the workload level, Turbo may also reduce costs associated with idle compute, delayed processing, data movement, and operating separate storage architectures. (Actual workload TCO depends on each application’s compute, transaction, networking, and operational profile.)

High performance where it matters, Blob economics for the rest.

AI and data systems increasingly need high storage performance for data actively feeding compute, while retaining much larger datasets economically over time.

Turbo tier brings those requirements together in Azure Blob Storage. SSD-backed performance accelerates performance-demanding data, Smart placement adapts placement to changing access patterns and workload needs, and Blob Storage tiering keeps less performance-sensitive data on cost-efficient HDD-backed access tiers.

For supported configurations, Turbo runs within an eligible Standard Storage account, using the same namespace and Blob APIs your applications already use. During the preview, eligibility is limited to GPv2 accounts that use locally redundant storage (LRS) in selected regions, subject to regional capacity. 

Request access to the Turbo tier preview today and evaluate Turbo tier with your own workload to see how much of your data needs SSD-backed performance.

Updated Nov 14, 2026
Version 1.0