# WEKA® NeuralMesh™ Object Store

**Object Storage That Keeps GPUs Moving**

High-performance S3 for AI, with GPU-scale concurrency, linear scaling, and unified file and object access.

## What It Takes to Run S3 at AI Speed

### Performance Bucket

**Keep S3 Fast Under AI Load**

Fast ListObject enumeration, Lightweight ETAG hashing, and Data Integrity Freedom target the exact operations that saturate first under AI load — catalog scans, high-concurrency writes, and checksum overhead. S3 over RDMA moves data directly into GPU memory, cutting latency from the inference path.

- [Read the Architecture](/resources/white-paper/wekaio-architectural-whitepaper)

### Distributed Metadata

**Add Nodes, Add Performance**

Fully distributed metadata eliminates centralized controllers, so every node added contributes throughput instead of contention. Performance scales linearly from six nodes to hundreds.

- [Meet NeuralMesh](https://weka.io/product/neuralmesh)

### Unified Object and File

**One Data Layer for S3 and POSIX**

S3 and POSIX operate on the same physical data blocks, with cross-protocol coherency enforced at the architecture level. A file written through NFS reads immediately through S3, no transformation, no copy.

- [Read Our Protocol Docs](https://docs.weka.io/?_gl=1*1vtogq0*_ga*MTQ5MzA3OTQ5OC4xNzg0NDA2NTUw)

### Data Reduction

**Always-On, No Performance Cost**

Data Reduction runs inline with under 5% write overhead, using similarity-based compression to capture redundancy across blocks. Background-first processing keeps writes at native NVMe speed, so capacity gains never cost GPU utilization.

- [Read About Data Reduction](/article/data-reduction-capacity-and-performance)

### AlloyFlash

**The Right Flash, From the First Write**

AlloyFlash places each write on the flash best suited for it as it happens — small writes to high-performance TLC, larger writes to high-density QLC, so AI infrastructure gets both performance and flash economics without manual tuning. 

### Multitenant Operations

**Scale Tenants Without Scaling Operations**

The NeuralMesh Kubernetes Operator provisions isolated tenants, each with dedicated S3 buckets, IAM policies, and lifecycle rules, through the same workflow already used for compute. NeuralMesh Observe brings visibility into one view.

- [Read the Kubernetes Brief](/resources/solution-brief/weka-kubernetes-operator)

## Proven at AI Scale

- **3x** *Greater object throughput than nearest alternatives*
- **4x** *Faster enumeration across a 280-million-object dataset*
- **6x** *Data reduction with under 5% write overhead*

## Customer Story

> We needed something that’s much more scalable than existing NAS solutions — an infrastructure that could grow to hundreds of petabytes. Our existing solution couldn’t provide that scale and wasn’t performing as well in these magnitudes — that’s what drove us to WEKA.

— David Ardley, Genomics England

## Frequently Asked Questions

### What is NeuralMesh Object Store?

NeuralMesh Object Store is a high-performance S3-compatible object storage system for AI workloads, built on WEKA's NeuralMesh architecture. It provides S3 and POSIX access to the same physical data, with no gateway or separate object tier.

### Can the same data be accessed through both S3 and POSIX?

Yes. S3 and POSIX access the same physical data in NeuralMesh. Data written through file protocols is immediately available through S3, and vice versa, without copying or moving the data between storage systems.

### Does NeuralMesh Object Store work with existing S3 applications and tools?

Yes. NeuralMesh Object Store provides S3-compatible APIs for existing S3 applications, tools, and AI pipelines. It also supports capabilities including IAM policies, temporary STS credentials, lifecycle rules, versioning, event notifications, and delete protection.

### Where can NeuralMesh Object Store run?

NeuralMesh Object Store can run on qualified commodity servers or on WEKApod, WEKA's fully integrated appliance for customers seeking maximum rack-scale performance and density.

### How does NeuralMesh Object Store manage security and identity?

NeuralMesh Object Store supports IAM policies, temporary STS credentials, and LDAP-mapped S3 identities. S3 API activity can also be streamed as structured events to SIEM platforms for centralized security monitoring and auditing.

### How does NeuralMesh Object Store support multitenant environments?

NeuralMesh Object Store provisions isolated S3 tenants through standard Kubernetes manifests. NeuralMesh Observe provides visibility into request rates, latency, and per-tenant consumption, simplifying operations as multitenant environments grow.

## Related Resources

- [Feature Focus: S3 Object Storage That Never Makes GPUs Wait](/article/neuralmesh-fast-s3-object-store) (Article)
- [The S3 Object Store That Never Makes GPUs Wait](/resource/the-s3-object-store-that-never-makes-gpus-wait) (Solution Brief)
- [NeuralMesh Fast S3 Object Storage](/resource/neuralmesh-fast-s3-object-storage) (Technical Brief)