NVIDIA CEO Jensen Huang is reframing how enterprises think about artificial intelligence resources, arguing that the future of work will require companies to allocate "token budgets" to employees just as they do computing hardware. Speaking at NVIDIA's GTC 2026 conference, Huang outlined a vision where AI token consumption becomes as fundamental to engineering work as having a laptop, fundamentally reshaping how organizations budget for and deploy AI capabilities .

What Does a "Token Budget" Actually Mean for Your Company?

Huang's concept centers on a simple but powerful idea: tokens, the basic units of text that AI models process, will become a measurable resource that companies must allocate and track. Just as IT departments manage laptop budgets and computing resources, they will soon need to manage token consumption across their workforce. This shift reflects the reality that agentic AI, which refers to AI systems that can autonomously perform tasks and make decisions, has reached an inflection point where it's becoming essential to daily work .

"What used to be a thing for engineers is when you come to work, they give you a laptop. Now when you come to work, they give you laptop and tokens. And token budget is now a real thing. Every engineer is going to have a token budget," said Jensen Huang, founder and CEO of NVIDIA.

Jensen Huang, Founder and CEO, NVIDIA

The implications are significant. If a company hires a $300,000 engineer who consumes no tokens in their work, Huang suggests that raises fundamental questions about how that employee is contributing to the organization. This framing positions token consumption as a key performance indicator and resource allocation metric for the enterprise AI era .

How to Prepare Your Organization for Token-Based Resource Management

• Establish Token Budgets: Begin allocating token consumption limits to teams and individual engineers, similar to how cloud computing budgets are managed today, to ensure efficient use of AI resources and prevent runaway costs.
• Monitor AI Usage Patterns: Implement tracking systems to measure how different departments and roles consume tokens, identifying which workflows generate the most value per token spent.
• Plan Infrastructure Investments: Recognize that tokens must be "manufactured" through computing infrastructure, meaning organizations need to invest in GPU capacity and AI factories to support their token budgets.
• Develop OpenClaw Strategies: Evaluate how your company will integrate NVIDIA's OpenClaw, an open-source AI operating system, into your infrastructure, as Huang indicated every company will need a clear strategy for this technology.

Huang emphasized that computers are transitioning from being mere tools to becoming "manufacturing equipment" for tokens. This distinction matters because it changes how organizations should think about capital expenditure and resource allocation. Rather than viewing AI infrastructure as a one-time purchase, companies should see it as ongoing manufacturing capacity that produces the tokens their workforce consumes .

Huang

Why Is NVIDIA Projecting $1 Trillion in Orders Through 2027?

The token budget concept directly connects to NVIDIA's ambitious growth projections. At GTC 2026, Huang announced that NVIDIA has "line of sight" to $1 trillion in orders, extending through 2027, up from the $500 billion figure outlined at the previous year's conference . This projection specifically covers Blackwell and Rubin GPU architectures, which are designed to handle the massive token generation demands that enterprises are beginning to experience.

Huang clarified that this $1 trillion figure represents strong visibility and demand forecasts backed by actual purchase orders, not speculative projections. The timeline extension through 2027 reflects the reality that building AI infrastructure at scale takes time, but the underlying demand is concrete and measurable .

The inference inflection point, where AI systems are increasingly being used to generate responses rather than just train models, is driving this demand surge. Companies are racing to build what Huang calls "AI factories," massive computing facilities dedicated to producing tokens at scale. This represents a fundamental shift from the training-focused AI infrastructure of previous years to inference-focused systems that generate outputs for end users .

What Role Will Open-Source AI Models Play in This Ecosystem?

NVIDIA is positioning itself not just as a hardware provider but as a comprehensive AI ecosystem player. The company is developing open-source models like Nemotron, which Huang indicated will be "near the world's best" in performance. This strategy provides enterprises with strong, fine-tunable language models that can be customized for specific use cases without requiring companies to build models from scratch .

Huang also highlighted OpenClaw, NVIDIA's open-source AI operating system, as a revolutionary technology that will become essential infrastructure for enterprises. While it may seem like a toy in its early stages, Huang compared its importance to historical technology shifts like the adoption of Linux, the internet, and mobile cloud strategies. Every software company and enterprise will need to develop a clear OpenClaw strategy, according to Huang .

How Is NVIDIA Expanding Its AI Infrastructure Ecosystem?

Beyond GPU manufacturing, NVIDIA is building a comprehensive ecosystem through strategic partnerships. In March 2026, NVIDIA announced a major partnership with Marvell Technology, investing $2 billion in the company and connecting Marvell's custom silicon and networking capabilities to NVIDIA's AI infrastructure through NVLink Fusion, a rack-scale platform for semi-custom AI systems .

This partnership enables customers to build heterogeneous AI infrastructure that combines NVIDIA's GPUs, networking, and storage technologies with Marvell's custom processors and optical interconnect solutions. The collaboration addresses a key market need: enterprises and hyperscalers want flexibility to customize their AI infrastructure while maintaining compatibility with NVIDIA's ecosystem .

"The inference inflection has arrived. Token generation demand is surging, and the world is racing to build AI factories. Together with Marvell, we are enabling customers to leverage NVIDIA's AI infrastructure ecosystem and scale to build specialized AI compute," stated Jensen Huang, founder and CEO of NVIDIA.

Jensen Huang, Founder and CEO, NVIDIA

The companies are also collaborating on silicon photonics technology and AI-RAN (AI Radio Access Network) solutions for 5G and 6G telecommunications infrastructure. This expansion signals that NVIDIA's ambitions extend beyond data center AI to encompassing the broader infrastructure that will support AI applications across industries .

What Does This Mean for Enterprise AI Adoption?

The shift toward token budgets and AI factories represents a maturation of enterprise AI adoption. Rather than treating AI as an experimental tool or departmental initiative, organizations are beginning to view it as core infrastructure requiring dedicated capital investment and resource management. The $1 trillion in projected orders reflects genuine demand from hyperscalers, regional cloud providers, industrial companies, and enterprises building on-premises AI systems .

Huang noted that NVIDIA's relationship with hyperscalers extends beyond simply selling hardware. By maintaining CUDA compatibility and supporting AI developers, NVIDIA effectively acts as a sales force for cloud service providers, bringing customers and workloads to their platforms. This ecosystem approach creates stickiness and ensures that NVIDIA remains central to how enterprises deploy AI, regardless of which cloud provider or infrastructure path they choose .

The physical AI inflection point, which Huang expects to arrive in a few years, will further expand this opportunity. Industrial applications, manufacturing, robotics, and other physical-world AI use cases will require on-premises and edge deployments, potentially growing the addressable market from the current 40 percent of on-premises deployments to a much larger share. Huang estimates that $50 trillion to $70 trillion of global industries will eventually require physical AI capabilities, dwarfing the digital AI market .