AI has undergone a profound transformation over the past decade. Once a tool for niche applications, it is now the driving force behind a new paradigm in computing. The shift from perception AI, capable of recognising images and speech, to generative AI has altered the fundamentals of how machines process and generate information. This transformation is not incremental; it is structural, redefining the very nature of computing.
In his keynote at GPC in San Jose, Jensen Huang, CEO of NVIDIA, explained that AI has made extraordinary progress over the past ten years, fundamentally reshaping the computing landscape. “It first gained widespread recognition through advancements in perception AI, such as computer vision and speech recognition,” he continued. “These early innovations laid the foundation for more sophisticated AI applications, culminating in the rise of generative AI over the past five years. Rather than retrieving pre-existing data, AI now dynamically interprets queries, constructs responses, and refines its knowledge over time. This transition from static retrieval to dynamic generation represents one of the most significant shifts in computing history.”
Generative AI has overturned the traditional retrieval-based model. It no longer simply fetches stored information but interprets context, generates responses dynamically, and refines its knowledge over time. This fundamental change in computing logic affects every stack layer, from software development to data centre architecture.
The rise of agentic AI
The next frontier in AI is agentic intelligence, which can perceive, reason, and act. Unlike traditional models that follow pre-programmed responses, agentic AI understands its environment, processes context dynamically, and plans its actions accordingly. It does not simply execute a single-step command; it breaks down problems, evaluates multiple solutions, and verifies its results.
“Agentic AI represents a major advancement,” Huang said. “It is AI with agency; the ability to perceive its environment, understand context, reason through problems, and plan and execute actions,” said Huang. “Unlike previous models, it does not merely follow pre-programmed responses. It can use tools, gather information from various sources, and apply its understanding dynamically. This ability to reason is what makes agentic AI distinct.”
This development moves AI beyond mere automation into a realm where it can solve problems in ways that mimic human cognitive processes. The ability to iterate solutions at machine speed presents a step-change in AI’s effectiveness, enabling it to handle tasks far beyond previous computational capabilities.
AI’s integration into industries becomes more profound as it moves towards greater autonomy. From manufacturing to healthcare, the ability to dynamically interpret and execute complex tasks opens new possibilities. However, it also presents challenges in governance, security, and accountability. Organisations must develop new frameworks for monitoring AI decisions, integrating transparency mechanisms, and addressing potential biases in decision-making processes.
AI steps into the physical world
Historically, AI has operated within digital constraints, processing data and making predictions in software-driven environments. The next evolutionary step is physical AI, a system capable of interacting with and manipulating the physical world. This shift requires a deep understanding of real-world physics, including friction, inertia, and cause-and-effect concepts.
“AI has largely been confined to the digital world, but physical AI enables robotics by allowing AI systems to understand and interact with the physical environment,” Huang said. “This means grasping concepts such as friction, inertia, cause and effect, and object permanence, principles that govern real-world interactions. These capabilities will be crucial for the next era of AI, where intelligent systems must navigate and manipulate the physical world efficiently.”
Physical AI enables robotics to move beyond programmed automation and into dynamic decision-making. Machines must interpret and adapt to their surroundings, whether navigating a factory floor or handling delicate materials in manufacturing. This transition from digital to embodied intelligence is essential for advancing robotics, autonomous vehicles, and industrial automation.
The challenge of scale
Three critical challenges define the next phase of AI development: data, training, and scaling. AI is fundamentally a data-driven discipline, and its effectiveness depends on the quality and availability of training datasets. Ensuring access to diverse, high-quality data remains a significant challenge, particularly in industries where proprietary information is tightly controlled.
“There are three fundamental challenges in AI development: data, training, and scaling,” Huang explained. “The first is solving the data problem. AI is a data-driven discipline requiring vast amounts of high-quality data to learn effectively. Accessing and curating this data is a significant challenge.”
Another hurdle is training AI at scale without human intervention. “Human-in-the-loop training is inefficient because it is constrained by time and scale,” Huang continued. AI needs to learn at superhuman speeds and on a scale far beyond human capacity. The third challenge is scaling AI effectively. This involves designing algorithms and architectures that improve AI performance as more computational resources are applied. The industry underestimated how much computing power would be required.”
The rise of agentic AI, with its reasoning capabilities, has only intensified the need for greater processing power. Unlike previous models that generated single-token responses, today’s AI executes complex multi-step reasoning, producing and verifying sequences of responses in real-time. “Instead of generating a single token, AI now generates entire sequences of tokens, verifying and refining responses in real-time.,” Huang said. “As a result, AI is generating at least 100 times more tokens than before, and to maintain responsiveness, processing speed must increase accordingly.”
The infrastructure behind AI
The computing industry responds to these demands by significantly expanding AI infrastructure. The scale of investment in AI data centres is unprecedented, with projections suggesting global spending will exceed a trillion dollars by 2030. Cloud providers, including Amazon, Google Cloud, Microsoft Azure, and Oracle, rapidly expand AI compute capacity to meet growing demand.
“The scale of AI computing has reached an inflexion point,” Huang explained. “Cloud service providers expanding AI compute capacity rapidly. The shift from general-purpose computing to AI-accelerated infrastructure is underway, and AI-driven data centres are being built at an extraordinary rate.”
The shift towards AI-accelerated computing involves replacing traditional general-purpose CPUs with AI-specific accelerators, and GPUs optimised for machine learning workloads. “By 2030, global AI data centre investment is expected to exceed a trillion dollars,” Huang noted. “Two key factors are driving this growth. The first is the shift from traditional computing, where software was hand-coded for general-purpose CPUs, to a new paradigm where machine learning software runs on AI accelerators and GPUs. The second is the growing recognition that AI development requires substantial capital investment in computing infrastructure.”
This shift has led to the widespread adoption of accelerated computing platforms. NVIDIA’s CUDA ecosystem, for example, has become a foundational layer for AI workloads, supporting everything from structured data acceleration to physics simulations. As AI development progresses, the computing landscape will continue evolving, optimising every stack layer for AI-driven processes.
The industry has reached an inflexion point. AI is no longer just a tool but the foundation upon which modern computing is being built. The transition from retrieval-based computing to generative and agentic AI is reshaping industries at an unprecedented pace. “The computing industry has reached a tipping point,” Huang concluded. “Accelerated computing is now the dominant paradigm, and AI is transforming every aspect of computing. The shift from retrieval-based computing to generative AI represents one of history’s most significant technological shifts, and it is only just beginning.”
