Nvidia CEO says millions of AI agents will soon work alongside 75,000 staff as

Nvidia’s chief executive Jensen Huang has taken the idea of “AI‑augmented workforces” from speculation to a concrete timeline, telling investors that the company expects “millions of AI agents” to be operating side‑by‑side with its 75,000‑person staff within a few years. The projection, reported by Supercar Blondie, reflects Nvidia’s confidence that its next‑generation hardware and software stack can support a scale of autonomous agents far beyond today’s pilot projects (Supercar Blondie). Huang framed the vision as a natural extension of the firm’s core business—delivering the compute power that underpins generative AI—arguing that the same GPUs that train large language models will soon run the inference workloads of countless specialized agents in real time.

The roadmap hinges on Nvidia’s recent partnership with Salesforce, announced on March 16, 2026, which integrates the company’s Nemotron 3 Nano model into Salesforce’s Agentforce platform (Salesforce Dictionary). Nemotron 3 Nano is purpose‑built for enterprise deployment: it offers a 1 million‑token context window, allowing an agent to retain a full customer history or a multi‑document workflow without truncation, and it employs a Mixture‑of‑Experts (MoE) architecture that activates only the most relevant subnetworks for each task. According to the partnership brief, this design “dramatically increases computational efficiency, reduces reasoning tokens, and lowers the overall compute cost” (Salesforce Dictionary). By embedding these agents directly into regulated, on‑premises environments, Salesforce aims to make AI a seamless part of daily CRM operations rather than a peripheral add‑on.

Nvidia’s GTC conference in San Jose underscored how the hardware ecosystem is being aligned to this vision. Coverage from CNET highlighted that the event featured extensive demos of the H100 NVL GPU, which delivers the bandwidth and tensor‑core performance required to run MoE‑based models at scale (CNET). The same sessions showcased robotic process automation and “AI‑claw” prototypes that illustrate how physical and software agents can cooperate in manufacturing and logistics. While the GTC narrative was broad, the consistent thread was the need for “cost‑efficient, high‑throughput inference” to power the millions of agents Huang described, a point reinforced by the MoE efficiency gains cited in the Salesforce collaboration.

Analysts see the hybrid workforce concept as both an opportunity and a risk. On the upside, the ability to deploy autonomous agents that can handle routine ticket routing, data entry, or even complex decision support could free up human talent for higher‑value tasks, potentially boosting productivity across Nvidia’s own internal operations. On the downside, the sheer volume of agents raises questions about governance, data privacy, and the compute budget required to keep them running 24/7. The partnership’s emphasis on “governed AI agents” within regulated environments suggests Nvidia and Salesforce are pre‑emptively addressing those concerns, but the broader industry still lacks standardized frameworks for monitoring millions of concurrent agents.

If Huang’s timeline holds, Nvidia will not only be a supplier of the chips that train AI but also a platform provider for the next wave of AI‑driven work. The company’s ability to marry high‑performance hardware with purpose‑built models like Nemotron 3 Nano could set the template for how enterprises transition from a human‑only workforce to a hybrid ecosystem where AI agents handle the bulk of repetitive and data‑intensive tasks, leaving humans to focus on strategy and creativity.