Hey Meta, Real Talk:
It’s Time to Move Beyond Lustre

See why the world’s most advanced AI companies are making the switch.

Schedule a Technical Review

The Hidden Bottleneck Slowing Down AI at Meta?
It’s Lustre.

Meta is pushing AI further than ever, but Lustre’s latency and scaling limits slow progress.

Faster data keeps GPUs active—maximizing output, efficiency, and impact.

Eliminate metadata drag to accelerate iteration speed and model quality.

Adaptive systems scale with demand—reducing rebuilds and protecting spend.

Performance Proven Head-to-Head

At Meta’s scale, these deltas compound into millions saved and months accelerated.

	WEKA on i3en.6xl	FSx Lustre 250 MB/sec
Cost	$194K	$215K
Read Bandwidth	227 GB/s	227 GB/s
Read IOPs	19M	1M
Write IOPs	4.7M	545K
List 1M files	~6s	~20s
List 10M files	<6 min	>28 min

NEURALMESH

Born in the Cloud.
Purpose-Built for Agentic AI.

Meta needs storage that performs anywhere. NeuralMesh™ delivers identical, high-performance storage across on-prem, multi-cloud, and hybrid — with zero performance loss.

Multi-Cloud Flexibility

Run seamlessly on AWS, GCP, OCI, or AI-native clouds.

Effortless Tiering

Move data across hot and cold storage tiers without breaking performance.

Proven in the Cloud

Stability AI (AWS), Contextual AI (GCP), and more trust WEKA.

Snap Environments Together

Cloud-to-cloud, on-prem-to-cloud, or back again — no rewrites, no re-benchmarking.

See the Proof

Solution Brief

Eliminate AI Bottlenecks. Put Your GPUs on Rails.

Lustre was built for yesterday’s workloads. NeuralMesh™ is built for agentic AI — at Meta scale.

Meta is defining the future of AI. To keep GPUs fully fed, pipelines fluid, and costs in check, storage can’t be the bottleneck.

With NeuralMesh, teams can:

Cut costs by eliminating idle GPU time
Accelerate workloads with unmatched metadata + IOPS performance
Scale without fragility using a distributed, elastic architecture

Future-proof AI infrastructure for exabyte-scale demands

Read the Solution Brief

Case Study

Why Stability AI Left Lustre for NeuralMesh Axon

Challenge: Stability AI pushed the limits of AWS with Lustre. GPUs sat idle, storage costs skyrocketed, and over-provisioning led to inefficiency.

Solution: Deploying NeuralMesh™ Axon® on AWS, Stability AI scaled to ~400 Amazon EC2 P4d instances with 2.2PB NVMe flash and 5PB object storage to power Stable Diffusion training.

Results:

🚀 93% GPU utilization efficiency — no more idle compute
💰 95% lower storage cost per TB vs. Lustre
⚡ 35% faster training cycles, cutting 3 weeks from a 60-day epoch

Get the Details

At Meta Scale, Every Bottleneck Compounds

Over the course of a year, Lustre’s inefficiencies result in millions of wasted GPU hours and delayed
time-to-value. NeuralMesh eliminates this drag.

Your GPUs Deserve NeuralMesh