Nvidia GTC DC 2025: Weka Unveils New NeuralMesh Architecture Built for Nvidia BlueField-4

At GTC Washington, D.C. 2025, Weka.IO Ltd. announced that it is developing the next-gen of NeuralMesh by Weka, its intelligent storage system, for the newly announced Nvidia BlueField-4 data processing unit (DPU), marking a transformational shift in how AI infrastructure will be architected and deployed.

This approach eliminates the need for standalone CPU servers, instead leveraging the unprecedented 800Gb/s networking bandwidth and 6x compute improvement of Nvidia BlueField-4 to deliver what Liran Zvibel, CEO, Weka, hails as “the perfect foundation for the future of AI data infrastructure.“

Reimagining AI Infrastructure Architecture
Rather than deploying NeuralMesh on traditional servers with separate CPUs, the new-gen of Weka’s NeuralMesh storage software will run directly on Nvidia BlueField-4, fundamentally simplifying AI infrastructure deployment while improving economics, performance, and power efficiency. This architectural shift represents the firm‘s most substantial alignment yet with Nvidia’s infrastructure roadmap and vision for the AI data center.

Perfect Complement to Modern AI Infrastructure
Nvidia BlueField-4 brings programmable infrastructure, acceleration, zero-trust security, and massive networking capabilities that align with NeuralMesh’s microservices-based mesh architecture. When paired with Nvidia’s latest accelerated compute platforms, BlueField-4 and NeuralMesh create a unified, efficient foundation for gigascale AI factories.

“The rapid evolution of AI demands intelligent data infrastructure that can adapt and scale with unprecedented efficiency,” said Liran Zvibel, co-founder and CEO, Weka. “Weka’s NeuralMesh architecture, built for Nvidia BlueField-4, is a tremendous leap forward, providing the foundational storage for next-generation AI factories.”

“The demands of AI reasoning and other inference workloads require a new foundation for data infrastructure – one that unifies compute, network, and storage for unprecedented efficiency and scale,” said Justin Boitano, VP, enterprise products, Nvidia Corp. “With Nvidia BlueField-4 integration, Weka’s NeuralMesh storage system supports this architectural transformation to power efficient data processing in gigascale AI factories.“

Benefits of the BlueField-4-Powered Architecture
Weka’s next-gen NeuralMesh storage system, built on BlueField-4, will deliver:

• Power Efficiency: Delivering over 100x tokens/watt improvement for agentic AI workflows
• Simplified Infrastructure: Eliminates the need for dedicated storage servers with standalone CPUs, reducing complexity and footprint.
• Superior Economics: Lowers infrastructure costs by consolidating compute, networking, and storage functions.
• Enhanced Performance: Leverages BlueField-4’s 800Gb/s networking and 6x compute improvement for unmatched data movement speeds.
• Seamless Integration: Aligns with Nvidia’s infrastructure roadmap and AI factory architectures.
• Zero-Trust Security: Built-in security acceleration and tenant isolation at line rate through the Nvidia DOCA software framework.

Collaboration for Age of Reasoning
Weka and Nvidia have collaborated closely for close to a decade, spanning Nvidia AI Enterprise software, Weka’s achievement of multiple Nvidia certifications, and collaboration on the introduction of new Nvidia technologies. NeuralMesh storage software built on BlueField-4 represents Weka’s deepest integration of Nvidia technologies yet.

“Power efficiency is key to bringing gigascale AI factories online. Nvidia BlueField-4 and Weka’s NeuralMesh storage software will directly address this challenge, delivering breakthrough tokens-per-watt efficiency without compromising performance, scale, or operational simplicity. We view NeuralMesh’s integration with BlueField-4 as a strategic growth engine that can simplify and accelerate Enterprise AI factory deployments for our customers,” said Nilesh Patel, chief strategy officer, Weka.

Availability and Next Steps
The company is actively developing its next-gen NeuralMesh software release for Nvidia BlueField-4, with detailed roadmap information to be shared with customers and partners in early 2026.