Azure's NVMe Comeback: How Laosv4 Changed Everything

Azure just dropped a bomb on the cloud storage market. The new Laosv4 instances are so ridiculously over-provisioned with NVMe capacity that even a tiny 4-core VM gets 2.88 TB of blazing-fast local storage. This isn't an incremental improvement—it's a complete reset of what "storage-optimized" means.

From Lsv2 Disappointment to Laosv4 Domination

Let's be brutally honest about where Azure was. The original Lsv2 series (launched 2018) was... underwhelming. Old architecture, inconsistent performance, limited configurations. If you needed serious NVMe performance, you looked at GCP or AWS. Azure's storage-optimized instances were barely on the radar.

The Lsv2 felt like a checkbox feature—"yes, we have NVMe instances too"—rather than a genuine attempt to compete. It worked, technically, but nobody was excited about it.

The Ridiculous NVMe Capacity Story

Azure didn't just improve with Laosv4—they went absolutely overboard in the best possible way. Look at these numbers:

Notice the pattern? 0.72 TB of NVMe per vCPU across the entire lineup. This is unprecedented. Azure isn't skimping on the small instances—they're giving you the same NVMe density whether you rent 4 cores or 32.

Heating Up the Competition

Azure's Laosv4 move forces GCP and AWS to respond. Here's how the competitive landscape looks now:

GCP: Still the Balanced Champion

GCP GCP's N2 instances remain excellent—good NVMe options with strong networking (32 Gbps on n2-standard-32). But their local NVMe comes in smaller increments per core. A comparable n2-standard-4 with local SSDs doesn't match Laosv4's capacity.

GCP advantage: Better network bandwidth, more mature ecosystem
Azure advantage: Ridiculous NVMe capacity, even on small instances

AWS: Steady and Proven

AWS The AWS i3en/i4i series are rock-solid. Great for production workloads that value reliability over bleeding-edge capacity. But instance pricing for equivalent NVMe capacity is higher than Laosv4.

AWS advantage: Battle-tested, excellent docs, wide availability
Azure advantage: More NVMe capacity per dollar

The Quota Wars: Scarcity = Success

Want to know if a cloud product is actually good? Check the quota limits. Laosv4 quotas are ridiculously hard to get.

Try spinning up an L32aos_v4 in a popular region like East US—instant quota hit. Submit a request—wait 2-3 weeks for approval. This isn't Azure being stingy; it's a signal that demand is crushing supply.

The old Lsv2? Never had quota problems because nobody wanted them. Laosv4? Can't build capacity fast enough to meet demand.

The Network Reality Check

Before we get carried away, let's address the weakness: networking still lags behind GCP.

• L4aos_v4 (4 cores): 10 Gbps network
• L32aos_v4 (32 cores): 16 Gbps network

Compare to GCP n2-standard-32 at 32 Gbps, or AWS i4i.8xlarge at 18.75 Gbps. Azure's networking is functional but not leading-edge.

This matters for distributed storage. When we tested MayaScale on L4aos_v4 with active-active replication, write IOPS peaked at 99K before the 10 Gbps network became the bottleneck. The NVMe drives could push more—but network replication consumed the bandwidth budget.

For single-node workloads (local databases, build caches, staging areas), the network is fine. For multi-node distributed storage, you'll hit the ceiling faster than you'd like.

Azure's High-Bandwidth Secret Weapon

But wait—Azure has a solution for bandwidth-hungry workloads: the DNSv6 and other N-series instances.

These aren't storage-optimized, but they support NVMe temp disks and deliver up to 80 Gbps networking on larger SKUs. If you need both massive NVMe and high network bandwidth, N-series is your answer.

Azure Strategy Pick your bottleneck:

• Storage bottleneck? → Laosv4 (max NVMe capacity)
• Network bottleneck? → N-series (high bandwidth + NVMe)
• Both? → N-series (pay premium for both)

Real-World Testing: MayaScale on Laosv4

We've been running MayaScale across all three clouds, and Laosv4 delivers some impressive numbers:

L4aos_v4 Performance (4 cores, 2.88 TB NVMe)

• Read latency: 139 μs (0.139 ms) at QD1
• Write latency: 222 μs (0.222 ms) at QD1
• Peak read IOPS: 352K (sub-1ms latency)
• Peak write IOPS: 77K (sub-1ms latency)

These are phenomenal numbers for a 4-core instance. The write IOPS ceiling at 77K is network-limited, not storage-limited—the NVMe drives can do more.

Where Laosv4 Excels

Perfect for:

• Single-node databases (PostgreSQL, MySQL, MongoDB primaries)
• Build systems and CI/CD pipelines needing fast local cache
• Data staging and ETL processing with massive temp storage needs
• In-memory databases with large persistent backing stores
• Development and testing environments with real production-like storage

Where N-Series Wins

Choose N-series when:

• Running distributed storage (MayaScale, Ceph, GlusterFS) across multiple nodes
• Heavy network replication or synchronization requirements
• Applications need both NVMe and 40+ Gbps networking
• Cost is secondary to raw performance

The Economics: NVMe Per Dollar

Here's where Azure's Laosv4 strategy becomes clear: they're competing on storage capacity per dollar, not overall compute price.

For workloads that need massive local storage but moderate compute, Laosv4 absolutely crushes the competition. You're getting 2.88 TB of NVMe on a 4-core instance for a fraction of what an equivalent capacity configuration would cost on GCP or AWS.

The Multi-Cloud Verdict

Azure's Laosv4 fundamentally changes the conversation. For years, the answer to "where should we run storage-intensive workloads?" was "GCP or AWS." Now Azure has a legitimate—and in some cases superior—answer.

Bottom Line: Competition Works

Azure's Laosv4 series proves that cloud competition benefits customers. After years of playing catch-up in storage-optimized instances, Azure came back with a vengeance—delivering NVMe capacity that's frankly ridiculous in the best way.

Yes, the network bandwidth still lags GCP. Yes, quotas are tight. But Azure fundamentally changed what customers can expect from a "storage-optimized" instance. 2.88 TB on a 4-core VM is a game-changer.

The cloud NVMe wars just got interesting. Azure is back in the fight, and they're swinging hard.