Nvidia seized the CES spotlight in Las Vegas to formally introduce Rubin, its next-generation artificial intelligence platform. Scheduled to reach customers in the second half of 2026, the system has already attracted substantial interest from cloud providers and AI companies, including Meta, Microsoft, and xAI.
While this might seem like another processor announcement, Rubin signifies far more than incremental improvement. The platform reflects a fundamental reconception of computing architecture – one designed for an AI landscape where systems operate as continuous agents rather than isolated responders, maintaining persistent context, coordinating across processes, and engaging with physical environments through robotics and autonomous systems.
Nvidia founder and CEO Jensen Huang characterized the shift directly. "AI has evolved beyond single-interaction chatbots into intelligent collaborators that comprehend the physical world, reason across extended contexts, ground themselves in factual accuracy, deploy tools for substantive work, and sustain both immediate and long-term memory," he stated.
From component supplier to system architect
Nvidia built its reputation on graphics processing units. Even as the company became synonymous with AI computing, the GPU remained the gravitational center with supporting components orbiting around it. Rubin breaks that pattern decisively. The graphics processor retains prominence but no longer commands exclusive primacy.
The platform presents an integrated architecture conceived as a unified infrastructure from inception. Graphics processor, central processor, memory, storage, connectivity, and networking were jointly engineered (codesigned) toward a singular objective: supporting AI operations over time with sustained memory and multi-process coordination. The distinction resembles contrasting engine upgrades for existing vehicles against a ground-up automotive design. Rubin reconstructs the entire computing framework rather than optimizing individual elements.
Among Rubin's notable departures from convention, the inclusion of a proprietary central processor stands out particularly. Alongside the graphics unit, Nvidia introduced Vera – a CPU component positioned at the platform's core, claiming territory historically dominated by Intel and AMD. The emergence of Nvidia-designed central processing within the architecture signals a transformation in how the company approaches AI-optimized data center infrastructure.
Contemporary data centers, including those purpose-built for artificial intelligence workloads, rely on central processors as system orchestrators. These CPUs operate system software and management frameworks, handle request routing, govern memory and input-output functions, and schedule graphics accelerator operations. While GPUs handle computational intensity, CPUs determine operational sequencing and component integration – functioning as conductors that coordinate computing resources.
Vera's integration within Rubin fundamentally alters this operational model. Rather than general-purpose CPUs managing systems externally, Nvidia embedded specialized processing within the platform architecture itself. Vera coordinates agent interactions, governs complex workflows extending over time, and executes logic unsuitable for GPU implementation. This assumes responsibilities traditionally handled by Intel or AMD processors, enabling Nvidia to architect unified systems where computation, memory, networking, and scheduling function cohesively.
The approach reflects broader evolution in data center design, particularly facilities engineered specifically for AI operations (AI-first infrastructure). Instead of general-purpose CPUs hosting graphics accelerators, Rubin offers an architecture in which central processors themselves are optimized for artificial intelligence workloads from design inception. For Intel, decades-long data center dominance is challenged as the inference-and-agent computing era transforms not just GPU roles but also the CPU's foundational position.
Memory, storage, and what lies between
Rubin's most consequential innovation may involve memory infrastructure. Contemporary AI systems, particularly agent-based implementations, generate substantial contextual data during execution. Historically, performance requirements mandated retaining this information in GPU memory, which was constrained and costly.
Rubin challenges this constraint directly. The platform redirects portions of contextual memory to dedicated storage layers that exhibit memory-like characteristics while remaining outside GPU boundaries. This enables context preservation across time, cross-process information sharing, and rapid access without consuming expensive computational resources.
Israeli engineering influence
Israeli technical contributions permeate the architecture. Rubin incorporates six distinct chips – Nvidia's Israeli development centers created four networking and connectivity components, while the Vera central processor also received partial development in Israel.
Nvidia has steadily expanded its Israeli operations in recent years, amplifying both workforce scale and the scope of technical responsibility. The shift toward architectures in which GPUs share prominence with integrated components emphasizes domains where Israeli engineering demonstrates particular strength – networking, communications, security, and complex system architecture.
Viewed comprehensively, Rubin transcends the status of a technological platform, demonstrating the strengthening position of Nvidia's Israeli operations within one of the world's most significant technology companies.
