As enterprises push generative and agentic AI into production, the conversation is, kind of shifting, beyond GPUs. The bigger challenge now is moving, storing, and managing massive quantities of data, without slowing down AI workloads. That is the problem WEKA is targeting with the launch of NeuralMesh 6 and WEKApod 3, both products built to work together as one unified AI infrastructure platform, not just separately.
NeuralMesh 6 covers the software layer, while WEKApod 3 brings the hardware built to support it. Together they’re aimed at enterprises, cloud providers, AI service companies, and research organizations running large-scale training plus inference workloads.
NeuralMesh 6 introduces a set of upgrades meant to make enterprise AI operations less painful. One of its most important additions is native multi-tenancy. This lets multiple teams or customers use the same infrastructure, yet still keep workloads isolated via separate networking, encryption, authentication, and quality of service controls. Instead of deploying a dedicated cluster for each project, organizations can securely run thousands of isolated AI environments on shared infrastructure, which sounds simple but it isn’t.
The platform also unifies file and object storage, so AI teams can reach the same data through standard POSIX file systems or Amazon S3 object storage, without creating duplicate copies. That approach makes it easier to manage training datasets, inference pipelines, and enterprise applications, from a single place, rather than juggling multiple systems.
Another key capability is intelligent data mobility. NeuralMesh 6 supports metadata-first replication, asynchronous replication, and remote caching, allowing distributed AI workloads to begin accessing datasets before complete transfers are finished. For organizations operating across multiple cloud regions or hybrid environments, this reduces delays while improving collaboration between teams.
WEKA has also focused on storage efficiency. NeuralMesh 6 includes always-on deduplication, compression, fingerprinting, and similarity detection that the company says can reduce AI dataset storage requirements by up to six times while maintaining low performance overhead. The release also adds Kubernetes-native deployment through a dedicated operator and introduces NeuralMesh Observe, a monitoring platform that provides centralized visibility, diagnostics, and alerts across AI infrastructure.
Complementing the software is WEKApod 3, the latest generation of WEKA’s AI storage and memory platform. Built with custom hardware, PCIe Gen 6 architecture, and NVIDIA ConnectX networking, it is designed to maximize performance without increasing datacenter footprint.
Also Read: Breaking the Test-Tube Bias: How OpenAI’s ‘Deployment Simulation’ Redefines Enterprise Software and Security
Its flagship Prime Max configuration supports up to 1.1 exabytes of effective storage capacity in a single rack, backed by 441.5 petabytes of raw capacity. According to WEKA, the platform can deliver 10.2 terabytes per second of throughput and 210 million IOPS per rack, making it suitable for large language models, AI factories, and high-concurrency inference environments.
The WEKApod 3 portfolio includes three configurations to address different workloads. Nitro focuses on maximum performance with all-TLC flash storage. Prime balances speed and capacity through WEKA’s AlloyFlash technology, which intelligently places frequently accessed data on faster TLC storage while moving less active workloads to QLC flash. Prime Max targets organizations managing extremely large AI datasets where storage density is a priority.
The combined launch kind of reflects a broader shift happening across enterprise AI. Businesses are realizing that just adding more GPUs, doesn’t really solve the infrastructure bottlenecks. Data movement, storage efficiency, memory capacity and resource utilization are starting to matter as much, or more, as AI models keep getting larger, and more complex.
For cloud providers and AI infrastructure operators, platforms that blend software-defined data management with purpose-built hardware can make deployment easier while also pushing better GPU utilization. Enterprises may also see less storage strain, quicker access to distributed datasets, plus infrastructure that scales, without having to keep expanding physical datacenter space again and again.
The announcement also underlines that AI infrastructure is turning into a full technology stack, not simply a bunch of separate pieces. Software that handles data more intelligently and hardware built for AI workloads are increasingly being developed together, so performance bottlenecks can be removed before they ever show up in production systems.
With NeuralMesh 6 and WEKApod 3, WEKA is positioning itself around that changing reality. Instead of focusing solely on faster storage or larger capacity, the company is targeting the broader challenge of helping enterprises build AI environments that are easier to manage, more efficient to operate, and capable of supporting the next generation of production-scale AI applications.
