NeuReality Unveils OS for AI Factories to Tame Chaotic Inference Market

TEL AVIV, Israel – March 12, 2026 – As the artificial intelligence boom fuels an insatiable demand for computing power, Israeli startup NeuReality today unveiled a new software platform designed to bring order to the chaos. The company introduced NR-NEXUS, which it bills as an “inference operating system” aimed at transforming today’s fragmented and often inefficient AI infrastructure into streamlined, production-ready “AI token factories.”

For enterprises and cloud providers grappling with the astronomical costs and complexities of running large-scale AI models, the announcement signals a potential shift in focus from raw hardware acquisition to the sophisticated software needed to manage it. Already deployed with beta customers, NR-NEXUS promises to orchestrate the entire AI inference stack, a critical process where trained AI models generate answers, predictions, or content—often measured in “tokens.”

The Operating System for the AI Factory

NeuReality is positioning its new platform as the foundational software for what many in the industry, including NVIDIA CEO Jensen Huang, are calling the “AI factory.” This concept reframes the data center not as a collection of general-purpose computers, but as a specialized, industrial-scale plant designed for one purpose: transforming raw data and electricity into valuable intelligence. The output of this factory is measured in the speed and efficiency of token generation.

NR-NEXUS aims to be the operating system for this new type of computer. Just as Windows and macOS provided a standardized platform for the personal computer era, NeuReality hopes its software will do the same for the intelligence era, unifying disparate components into a cohesive whole.

“AI inference is rapidly becoming one of the largest computing markets in the world, yet the infrastructure stack around it remains fragmented,” said Moshe Tanach, CEO of NeuReality, in the company's official announcement. “With NR-NEXUS, we are defining the operating system for AI token factories – enabling organizations to run and scale inference workloads efficiently across GPUs, emerging XPUs, hyperscalers, and dedicated AI clusters.”

Tackling Fragmentation and Vendor Lock-In

A core challenge for organizations deploying AI is the inefficiency baked into current systems. Expensive, powerful GPUs often sit underutilized, waiting for data or held back by processing bottlenecks. This fragmentation across different runtimes and systems drives up costs and limits the return on massive hardware investments. The market for AI inference management is a complex patchwork of solutions from hyperscale cloud providers, specialized platforms, and hardware giants, each with its own ecosystem.

NR-NEXUS directly confronts this issue with a hardware-agnostic approach. The company claims the operating system works across any CPU, GPU, or network interface card (NIC), and can be deployed in hyperscale cloud environments, on dedicated private GPU clusters, or with emerging specialized processors, known as XPUs. This flexibility is a significant selling point for enterprises wary of being locked into a single vendor's hardware and software stack, a common concern in a market dominated by a few major players.

By creating a universal software layer, NeuReality argues that organizations can optimize their existing infrastructure and future-proof their investments without being forced into costly re-architecture. “As open-source models and AI-native applications proliferate, operators need infrastructure that gives them flexibility rather than lock-in,” Tanach noted. “NR-NEXUS provides that foundation.”

Under the Hood: How NR-NEXUS Boosts Efficiency

The platform's promise of higher utilization and lower costs is rooted in its sophisticated orchestration capabilities. NR-NEXUS incorporates a feature called “Runtime Intelligence,” which dynamically analyzes each AI inference request and routes it along the most optimal path. This involves making real-time decisions about which inference engine to use, how to disaggregate workloads, and how to manage the flow of tokens and data caches for maximum efficiency.

Its modular architecture includes an “Orchestrator” to scale operations across a cluster and a “Governor” to manage incoming requests with security and observability. This system transforms what is often a chaotic collection of inference stacks into a governed, multi-tenant, and production-grade system. This approach builds on NeuReality's established expertise in AI hardware architecture, including its previous work on AI-native chips designed to eliminate CPU bottlenecks and push GPU utilization toward 100%.

Early use cases demonstrate the platform's ability to handle complex, real-world workloads. NeuReality has successfully used the system to deploy Meta's Llama 3.1 70B model on clusters integrating Qualcomm's Cloud AI 100 Ultra inference cards, showcasing its capacity to manage cutting-edge large language models and specialized hardware from different vendors.

A Market Hungry for Solutions

NeuReality is entering a market experiencing explosive growth. Projections show the global AI inference market soaring from around $100 billion in 2025 to over $250 billion by 2030. This expansion is driven by the widespread adoption of generative AI, which places unprecedented strain on computing infrastructure.

Enterprises are discovering that simply buying more GPUs is not a sustainable strategy. The primary pain points are not just the initial hardware cost, but the unpredictable operational expenses, deployment complexity, and the fundamental mismatch between traditional data center designs and the demands of AI. For many, the challenge of running AI inference at scale is becoming as significant as the initial migration to the cloud.

By focusing on optimizing the entire system to lower the cost per token, NeuReality is addressing the most critical economic question in the AI era. As organizations pour billions into the foundational hardware of their AI factories, the strategic focus is inevitably shifting toward the intelligent software layer that can guarantee a tangible return on that monumental investment.