NVIDIA CEO Predicts Software Development Will Shift to a Token-Driven Model

Speaking at the recent Morgan Stanley technology conference, NVIDIA CEO Jensen Huang made a bold prediction that could permanently change how we understand the IT industry. In the view of the head of the world's leading semiconductor company, software development is on the verge of a radical transformation and will soon become fully token-driven. The concept, which Huang calls token-driven, means abandoning the manual writing of lines of code in favor of orchestrating complex AI systems. The NVIDIA chief's statement does more than record the current automation trend. It predicts a fundamental shift in which traditional software companies will cease to exist in their present form and give way to autonomous agent networks.

To grasp the scale of this forecast, it helps to look at how quickly large language models have developed over the past few years. For a long time, the software industry relied on human capital: programmers translated business logic into languages machines could understand. With the arrival of advanced generative networks, however, the basic unit for measuring computational work became the token, a fragment of a word or symbol that a neural network processes to understand and generate text.

Huang, whose company is a key beneficiary of the AI boom thanks to its production of the most sought-after GPUs, sees the focus of the technology process shifting. Software development is ceasing to be a matter of typing commands on a keyboard and is becoming a process of continuous token generation and consumption by compute clusters.

The central element of the new paradigm will be the mass adoption of so-called "digital employees," which will fundamentally change the internal structure of IT companies. In Huang's vision, organizations will evolve into agent networks in which specialized AI agents continuously interact with one another to solve complex tasks. One agent might handle architecture, another write a first draft of the code, a third test it and hunt for vulnerabilities, and a fourth deploy the finished product.

In this ecosystem, the human is no longer a craftsman-coder but the head of a digital department who sets high-level business goals. Every iteration and every data exchange between these intelligent entities requires processing information, which translates directly into a continuous stream of burned and generated tokens.

The shift to a token-driven model means the economics of software development will be measured by entirely different metrics. Where the main resource used to be the time of a skilled developer, it is now the throughput of neural networks. The token becomes a universal internal currency and the main fuel for a new generation of autonomous systems. Companies will start planning budgets not only around the headcount of human employees but also around quotas for AI compute consumption. This creates a unique environment in which the efficiency of building digital products depends directly on how well the interaction architecture of autonomous agents is designed and how well their communication is optimized in token terms.

The obvious consequence of such a transformation will be an unprecedented, explosive growth in demand for compute. Digital workers do not tire, do not need breaks, and can run around the clock, so the volume of code they generate and data they analyze will exceed human capacity by many orders of magnitude. Keeping millions of such AI agents running worldwide will require colossal server resources. This aspect of Huang's forecast underscores NVIDIA's own strategic position in the market. The industry will need entire token factories made up of hundreds of thousands of GPUs to sustain new corporate infrastructures built entirely around artificial intelligence.

Such changes will also inevitably transform the labor market in the technology sector. The role of the entry-level programmer, who writes boilerplate code and handles routine tasks, will lose value rapidly, because tireless digital employees will take over those functions entirely. Systems engineers, AI solution architects, and specialists in managing agent networks will move to the forefront.

People will have to adapt to a reality in which their main job is not to program by hand but to direct the computing power of machines where it is needed, monitor the quality of the algorithms' work, and minimize the risk of logical errors by the AI. Educational institutions and corporate training centers will have to urgently revise their workforce training programs for this new era.

Ultimately, Huang's vision paints a picture of a future in which the line between creating software and directly consuming compute is erased for good. The transformation of traditional IT companies into networks of autonomous AI agents is not just another turn of technological progress but a fundamental change in the paradigm of creation in the digital space. The industry is moving fast toward a moment when ideas will become finished, working services solely through the uninterrupted conversion of electricity into silicon computation, and then into billions of precious tokens.

What sounds like a bold futurological prediction today will become everyday economic reality tomorrow, and the leaders will be those who learn faster than others to manage armies of digital creators.