Powerful on-device AI won't be a single large model. The effective paradigm is a smaller "orchestrator" model that acts as a router. It handles simple tasks, calls specialized local models (e.g., for PII filtering), and intelligently decides when to escalate complex queries to more powerful cloud-based models.

A Stanford study found that the vast majority of queries sent to powerful frontier models don't require their advanced capabilities. These tasks could be handled by smaller, faster, and more private local models at virtually no cost, revealing a massive inefficiency in the current API-centric approach.

The shift from simple chatbots (one user request, one API call) to agentic AI systems will decouple inference requests from direct user actions. A single user request could trigger hundreds or thousands of automated model calls, leading to an exponential increase in compute demand and cost.

The next wave of AI compute demand won't be from generating more outputs, but from agents performing exponentially more data collection for a single task. For example, a financial model could trigger an agent to analyze vast datasets, like satellite imagery, multiplying token usage for one result.

The future of AI isn't just in the cloud. Personal devices, like Apple's future Macs, will run sophisticated LLMs locally. This enables hyper-personalized, private AI that can index and interact with your local files, photos, and emails without sending sensitive data to third-party servers, fundamentally changing the user experience.

The next wave of AI adoption involves 'agentic' workflows, where AI performs complex tasks autonomously. This shift from simple queries to agentic use is expected to increase token consumption by approximately 10x per task. This will drive a massive explosion in compute demand across all knowledge-work industries, not just coding.

The evolution of AI towards complex, autonomous "agents" makes relying solely on the cloud slow and expensive, as users burn through token budgets. Nvidia's bet is that running these agents locally on powerful new PC chips will be faster and cheaper for consumers, driving a major hardware shift away from pure cloud computing.

A cost-effective AI architecture involves using a small, local model on the user's device to pre-process requests. This local AI can condense large inputs into an efficient, smaller prompt before sending it to the expensive, powerful cloud model, optimizing resource usage.

Local models shouldn't be seen as direct competitors to frontier cloud models on raw power. Instead, their strategic value is as a 'generator in the garage'—a resilient, offline backup ensuring core AI workflows continue even if the main 'grid' (cloud AI) goes down.