NVIDIA made a bold bet on the future of computing at GTC 2026 this week, unveiling a two-pronged strategy that could reshape how AI reaches users and devices: agent computers designed for personal AI assistants, and AI grids that turn the world's telecommunications infrastructure into a distributed inference platform.
The Rise of Agent Computers
The company introduced "agent computers" as a new category of device, purpose-built for running personal AI agents locally. The flagship DGX Spark desktop AI supercomputer leads the charge, featuring 128GB of unified memory capable of supporting models with over 120 billion parameters.
"The paradigm of consumer computing has revolved around the concept of a personal device — from PCs to smartphones and tablets," NVIDIA noted. "Now, generative AI — particularly OpenClaw — has introduced a new category: agent computers."
Alongside DGX Spark, NVIDIA announced new open models optimized for local agents:
- Nemotron 3 Super 120B: A 120-billion-parameter open model with 12 billion active parameters, designed for complex agentic AI systems. On the new PinchBench benchmark for OpenClaw performance, it scored 85.6% — the top open model in its class. - Nemotron 3 Nano 4B: A compact model for resource-constrained hardware, ideal for building action-taking conversational personas in games and apps. - Mistral Small 4: 119 billion parameters with 6 billion active parameters, unifying capabilities across chat, coding and agentic tasks.
The company also released NemoClaw, an open-source stack for OpenClaw that optimizes agent experiences on NVIDIA devices with enhanced security and local model support. DGX Spark and RTX PRO GPUs can run these models locally, enabling users to have proactive AI assistants reachable from their preferred messaging apps.
AI Grids: Telecom Goes Distributed
On the infrastructure front, NVIDIA announced that major telecommunications operators are building AI grids — geographically distributed AI infrastructure leveraging existing network footprints to power inference closer to users and devices.
The vision is striking: telcos and distributed cloud providers operate approximately 100,000 distributed network data centers worldwide, spanning regional hubs, mobile switching offices and central offices. These sites have enough spare power to offer over 100 gigawatts of new AI capacity over time.
AT&T is partnering with Cisco and NVIDIA to build an AI grid for IoT, running AI inference closer to where data is created. This enables mission-critical applications like public-safety use cases with Linker Vision, supporting faster detection and response while keeping sensitive information under customer control at the network edge.
"Scaling AI services that are both highly secure and accessible for enterprises and developers is a core pillar of our IoT connectivity strategy," said Shawn Hakl, senior vice president of product at AT&T Business.
Comcast is transforming one of the nation's largest low-latency broadband footprints into an AI grid for real-time, hyper-personalized experiences, working with NVIDIA, Decart, Personal AI and HPE.
This represents "a structural change in how AI is delivered, putting telecom networks at the center of scaling AI rather than just carrying its traffic," NVIDIA said.
What Comes Next
Jensen Huang also revealed NVIDIA's seven-chip strategy aimed at reaching $1 trillion in revenue by 2027, though details were sparse from this report. The combination of powerful local hardware and distributed edge infrastructure suggests NVIDIA sees the future as hybrid — part personal AI device, part globally distributed compute grid.
The GTC announcements make clear that NVIDIA is no longer just a chip company. It's positioning itself as the architect of an AI ecosystem that spans from the desktop to the edge of every cellular network.