Direct answer: at Azure Managed Lustre public list prices on 11 November 2026 (eastus, USD, excluding tax and discounts), the pricing page currently lists only Durable SSD in four tiers, billed per GiB/hour: 40 MB/s per TiB at $0.000114/GiB/h ≈ $85.22/TiB/month, 125 MB/s per TiB at $0.000198 ≈ $148.01, 250 MB/s per TiB at $0.000287 ≈ $214.54, and 500 MB/s per TiB at $0.000466 ≈ $348.34 (1 TiB=1,024 GiB, 730 hours). By capacity: 10 TiB = $852.17 / $1,480.09 / $2,145.38 / $3,483.44; 100 TiB = $8,521.73 / $14,800.90 / $21,453.82 / $34,834.43. The key is not "which tier is cheapest per TiB" but cost per MB/s: 40 tier ≈ $2.13/MB/s, 125 ≈ $1.18, 250 ≈ $0.86, 500 ≈ $0.70. Choose 40 for capacity, 500 for throughput; 10 TiB on 500 delivers about 5,000 MB/s, while 40 delivers 400 MB/s.

This guide is for architects, storage/SRE, HPC platform, finance and procurement selecting a parallel file system for AI training/HPC/rendering/EDA, comparing Managed Lustre with Blob/NetApp Files, or budgeting Lustre capacity and throughput. It separates official prices, editorial calculations and exclusions, covering the four tiers, per-TiB/per-MB/s costs, 10/50/100 TiB TCO, Durable SSD/LRS, Blob integration, AKS CSI, selection and a 90-day checklist. Prices come from the Managed Lustre pricing page, Azure Retail Prices API and Microsoft Learn overview (11 November 2026). This is not procurement advice.

Three rules to remember

(1) The four tiers do not sell the same capacity at different prices—they sell MB/s per TiB: 40 is cheapest per TiB ($85.22) but most expensive per MB/s ($2.13); 500 is most expensive per TiB ($348.34) but cheapest per MB/s ($0.70). (2) Throughput scales linearly with capacity: 10 TiB on 500 is about 5,000 MB/s; 100 TiB is about 50,000 MB/s; start from required MB/s, then derive capacity/tier. (3) The pricing page does not list a minimum capacity; this article calculates linearly per TiB—confirm the minimum/step in the create portal/docs.

Four tiers: GiB/hour, TiB/month and throughput

Managed Lustre tiers are named by MB/s per TiB: 40, 125, 250, 500. Billing is per GiB/hour; converted to TiB/month (1 TiB=1,024 GiB, 730h), the tiers are about $85.22, $148.01, $214.54 and $348.34 per TiB/month. Each TiB provides the tier's throughput: 40/125/250/500 MB/s; total throughput ≈ capacity (TiB) × tier. All four are Durable, SSD; the pricing page does not list a scratch tier.

TierGiB/hourTiB/monthThroughput/TiBCost per MB/s
40 MB/s per TiB$0.000114$85.2240 MB/s$2.13
125 MB/s per TiB$0.000198$148.01125 MB/s$1.18
250 MB/s per TiB$0.000287$214.54250 MB/s$0.86
500 MB/s per TiB$0.000466$348.34500 MB/s$0.70

Sources: Managed Lustre pricing page and Retail API (eastus, 11 November 2026). Editorial calculation: 1 TiB=1,024 GiB, 730h; TiB/month = GiB price×1,024×730; per MB/s = TiB monthly ÷ tier MB/s; excludes tax and discounts.

Capacity TCO: 10/50/100 TiB

Multiply by capacity: 10 TiB = $852.17 / $1,480.09 / $2,145.38 / $3,483.44; 50 TiB = $4,260.86 / $7,400.45 / $10,726.91 / $17,417.22; 100 TiB = $8,521.73 / $14,800.90 / $21,453.82 / $34,834.43. If you only need capacity with low throughput (e.g., pre-archive staging), 40 is cheapest; for AI training/rendering with high concurrent reads, 500 costs 4× per TiB but about 3× less per MB/s. Derive minimum capacity from peak throughput ÷ tier, then compare total cost.

Capacity40 tier/month125 tier/month250 tier/month500 tier/month
10 TiB$852.17$1,480.09$2,145.38$3,483.44
50 TiB$4,260.86$7,400.45$10,726.91$17,417.22
100 TiB$8,521.73$14,800.90$21,453.82$34,834.43
10 TiB total throughput400 MB/s1,250 MB/s2,500 MB/s5,000 MB/s
100 TiB total throughput4,000 MB/s12,500 MB/s25,000 MB/s50,000 MB/s

Editorial calculation: capacity × TiB monthly; total throughput = capacity (TiB) × tier MB/s (official tier naming); excludes minimum capacity/step, tax and discounts.

Throughput economics: why 500 can be "cheaper"

Divide "monthly per TiB" by "throughput per TiB" to get cost per MB/s: 40 = $2.13, 125 = $1.18, 250 = $0.86, 500 = $0.70. So 40 is cheapest per TiB but most expensive for throughput; 500 is most expensive per TiB but cheapest for throughput. If an app needs 12.5 GB/s (12,500 MB/s) of reads, 25 TiB on 500 (25×500=12,500 MB/s) costs about $8,708/month; on 40 you would need 312.5 TiB (about $26,630/month) to reach the same throughput. Conversely, if you need 2 GB/s and 100 TiB capacity, 40 already provides 4,000 MB/s—no need for 500.

Requirement40 tier125 tier250 tier500 tier
Capacity to reach 2,500 MB/s62.5 TiB20 TiB10 TiB5 TiB
Monthly cost$5,326.08$2,960.18$2,145.38$1,741.72
Capacity to reach 12,500 MB/s312.5 TiB100 TiB50 TiB25 TiB
Monthly cost$26,630.40$14,800.90$10,726.91$8,708.61

Editorial calculation: required capacity = target MB/s ÷ tier; monthly = capacity × TiB monthly; excludes minimum capacity/step, tax and discounts. 2,500/12,500 MB/s are example targets.

Durable SSD, LRS and Blob/AKS integration

The official overview says Managed Lustre uses Azure managed disks as OST data disks; Durable file systems use Premium SSD configured as LRS, replicating three copies within the local datacenter to survive disk/rack failures. For regional/global redundancy, integrate with Azure Blob Storage for data placement and cost management; AKS can mount it through the CSI driver. Data at rest is encrypted with Azure-managed keys by default, with customer-managed keys possible via Azure Disk server-side encryption. It is not a general SMB/NFS file share—it is a Lustre parallel file system for high-throughput/low-latency parallel access.

CapabilityOfficial noteCost/architecture impact
MediaDurable, Premium SSDAll tiers bill per GiB/h
RedundancyLRS, three local copiesRegional/global needs Blob integration
Blob integrationConnect Blob containers for placement/costBlob requests/storage separate
AKSCSI driverPods mount; nodes/network separate
EncryptionAzure-managed keys default, CMK optionalKey Vault/CMK may bill
ProtocolLustre parallel file systemNot general SMB/NFS

Sources: Managed Lustre overview and pricing page (11 November 2026). Blob/AKS/Key Vault bill as their own services; not estimated here.

Selection: Lustre, Blob and NetApp Files

Managed Lustre fits AI training/inference data loading, HPC simulation, rendering, EDA, genomics and other workloads needing high-concurrency/high-throughput/low-latency parallel access and accepting Lustre clients/protocol. Blob fits object storage, data lakes, cold/warm data and cost-first workloads without POSIX parallel-file-system semantics; cheaper but a different latency/throughput model. Azure NetApp Files fits NFS/SMB enterprise file sharing and low-latency scenarios with different protocol/throughput needs. Filter by protocol + throughput + capacity + latency first; don't compare only $/TiB—Lustre's 40 tier looks cheap at $85/TiB, but if throughput is insufficient you will upgrade tier or add capacity.

ScenarioFirst choiceCheck first
AI training data loadingLustre 250/500Peak MB/s, clients
HPC/rendering/EDALustreParallel IO, capacity
Large capacity, low throughputLustre 40Cost per MB/s
Objects/data lake/cold dataBlobProtocol/latency
NFS/SMB enterprise shareNetApp FilesProtocol/throughput
AKS persistent volumeLustre CSINodes/network
Regional/global redundancyLustre + BlobBlob requests/storage

Editorial summary, not a vendor recommendation; protocol/performance/prices follow official docs and bills; Blob/NetApp costs are covered in their own articles.

Hidden costs and 90-day checklist

Eight hidden costs: minimum capacity/step possibly above expectations, throughput scaling linearly with capacity, Blob requests/storage, AKS nodes/network, clients/drivers/operations, backup/snapshots (if done on the Blob side), cross-region/egress, and buying far more capacity than data just to reach throughput. 90 days: weeks 1–2 inventory data size, peak/average throughput, concurrent clients and protocol; weeks 2–3 derive capacity/tier from target MB/s; weeks 3–4 compare Blob/NetApp; weeks 4–6 pilot mount/benchmark; weeks 6–8 validate Blob integration/AKS CSI/encryption; weeks 8–10 optimize capacity/tier and avoid overbuying for throughput; weeks 10–12 reconcile the bill and set capacity/throughput alerts.

Cost itemTriggerControl
Lustre capacityTiB × tier monthlyDerive from peak throughput
Tier premium40→500Compare per MB/s
Minimum capacity/stepAt creationCheck portal/docs
Blob integrationRequests/storage/egressLifecycle/tiering
AKSNodes/network/CSIBudget with cluster
Clients/operationsDrivers/monitoring/labourStandard images
Cross-region/egressData movementKeep close to compute

Editorial summary: resources outside the price list bill as other Azure services; minimum capacity/step is not listed on the pricing page—check the create portal.

FAQ

How is Managed Lustre billed?

Per GiB/hour in four tiers: 40 MB/s/TiB $0.000114, 125 $0.000198, 250 $0.000287, 500 $0.000466; at 1 TiB=1,024 GiB and 730h, about $85.22/$148.01/$214.54/$348.34 per TiB/month.

What does 10 TiB cost per month?

40 tier $852.17, 125 $1,480.09, 250 $2,145.38, 500 $3,483.44; total throughput is about 400/1,250/2,500/5,000 MB/s.

Which tier is cheapest?

Depends on the basis: 40 is cheapest per TiB ($85.22), 500 is cheapest per MB/s ($0.70 vs $2.13). Choose 40 for capacity, 500 for throughput.

How do I estimate throughput?

Total throughput ≈ capacity (TiB) × tier MB/s. 10 TiB on 500 is about 5,000 MB/s; 100 TiB is about 50,000 MB/s. Set peak MB/s first, then derive capacity/tier.

Is there a minimum capacity?

The pricing page does not list a minimum capacity/step; this article calculates linearly per TiB. Check the create portal/official docs for the actual minimum and step, which can change small-capacity totals.

Lustre or Blob?

Choose Managed Lustre for POSIX parallel file system, high throughput/low latency and Lustre clients; choose Blob for object storage, data lakes, cold/warm data and cost-first workloads. Lustre can integrate with Blob for placement/redundancy.

Is LRS enough?

Durable file systems default to Premium SSD LRS with three local copies, surviving disk/rack failures; for regional/global redundancy integrate with Blob. Cross-region DR needs separate design.

How do I reduce cost?

Five steps: (1) derive capacity from peak throughput, avoid overbuying; (2) use 40 for large capacity/low throughput; (3) use 500 for high throughput and compare per MB/s; (4) use Blob tiering/lifecycle for long-term data; (5) verify minimum capacity/step and Blob/AKS add-ons.

Research method and limitations

Official facts (pricing page, Retail API, Learn overview): four GiB/h tiers, Durable SSD, LRS three copies, Blob integration, AKS CSI, default encryption. Editorial calculations: 1 TiB=1,024 GiB, 730h; TiB monthly; capacity×tier; total throughput=capacity×tier; per MB/s=TiB monthly÷tier; 10/50/100 TiB and 2,500/12,500 MB/s examples.

Seven limitations: (1) eastus USD list prices only, excluding tax/contract discounts/regional differences; (2) capacity/throughput/clients are editorial assumptions; (3) excludes minimum capacity/step (not on the pricing page), Blob requests/storage/egress, AKS nodes/network, clients/operations/backup; (4) no real parallel IO/latency benchmark; (5) tier throughput is official naming; actual performance depends on clients/network/file count; (6) only Durable SSD is listed; scratch/other media follow the official page; (7) prices/tiers change—recheck official pages and bills before purchase.

Sources and basis

Data cut-off 11 November 2026; prices/features follow Microsoft official pages and bills; this is not procurement, architecture or legal advice.

Turn capacity, tier and throughput into a checkable budget

Bring data size, peak throughput, clients/protocol, Blob/AKS needs and region; we turn Managed Lustre cost into a checkable budget.

Book a Lustre cost review