Disclosure: My company, Tirias Research, has consulted for IBM, Nvidia, and other companies mentioned in this article.

Growth drivers

The surge in generative AI adoption was more dramatic than even the most optimistic projections could predict. Then, a fresh spike in demand emerged in September 2024 with the launch of ChatGPT-o1, the first broadly implemented "reasoning" model. Unlike earlier versions, it didn’t just provide answers; it worked through questions, delivering more considered, logical, and nuanced responses. This kind of reasoning required significantly more behind-the-scenes "reasoning tokens" per session, leading to a sharp increase in token generation. Furthermore, user engagement levels soared. By the close of 2024, time spent generating content using generative and reasoning AI models had grown by over 22 times compared to the prior year.

Innovation also accelerated at an unprecedented pace. From the introduction of transformers in 2017 to ChatGPT-1 in 2022 and reasoning models in 2024, the timeline of breakthroughs continues to shrink. Sophisticated model architectures like mixture-of-experts (MoE) enable more efficient reasoning while keeping active parameter usage low. Open-source models such as Meta’s Llama series challenge proprietary solutions by offering lighter, faster alternatives that can run locally on laptops and smartphones. Meanwhile, optimizations like sparse attention and conditional computing are producing more efficient models like DeepSeek R1 (launched in 2025), which originally used only 37 billion active parameters per token, compared to Llama’s 405 billion or over 1 trillion in some closed models.

Token demand by the numbers

Tirias Research anticipates continued growth in user numbers, visit frequency, time spent, and AI-generated content. Additionally, with agentic APIs becoming available in 2025, AI agents will begin autonomously linking AI models together, forming thoughts, executing tasks, and interacting with other services. Human prompting will no longer be the sole catalyst for AI activity once autonomous agents start generating usage independently. Consequently, the annual rate of token generation is projected to jump from 677 trillion in 2024 to 2,092 trillion by the end of 2025 and reach 77,000 trillion (77 quadrillion) by the end of 2030.

Simon Solotko, Senior Analyst at Tirias Research, states: "The AI ecosystem is under extraordinary pressure. Multimodal capabilities, user demand, and agentic and multimedia workflows are progressing so rapidly that even improvements in compute hardware and software won’t be enough to counteract the rise in demand."

A forecast snapshot from 2028 shows that the use of AI assistants and agents will likely be dominated by a few providers. On the infrastructure side, however, API-accessed AI models are expected to power a broad array of business and consumer applications by embedding AI capabilities into customer-facing service providers.

The industry may shift toward a natural monopoly similar to Google's dominance in internet searches. With its early entry via ChatGPT and strong brand recognition, OpenAI currently leads the market in AI models and token generation. Whether OpenAI maintains its edge remains unclear.

Future Trends

Larger models will keep increasing in size and complexity, surpassing hardware advancements. The biggest models already exceed the memory capacity of any single accelerator, requiring clusters of GPUs and entire racks to process workloads. However, progress in distillation and efficiency will help scale down to smaller, specialized models. The arrival of DeepSeek marked a major leap in model efficiency, redefining performance benchmarks.

AI Agents will become widespread. Industry leaders, including Nvidia’s Jensen Huang and IBM’s Arvind Krishna, envision every employee working alongside multiple AI agents. Some agents will reside in machines, others in virtual environments, and still others in physical robots. AI agents will also start collaborating with one another.

Competition in AI will intensify. As models mature, differentiation isn't just about size or speed anymore—it now involves a broader set of factors. Services are embedding AI models into workflows, APIs, and interactive applications, driving toward full automation of tasks and entertainment. At the same time, cost constraints are pushing every player to adopt cutting-edge methods for faster training, better inference, and reduced computational costs. This competition extends beyond the enterprise—AI is now influencing global politics as nations race to innovate.

Moreover, AI will continue evolving. By the end of the decade, AI-generated images and video may surpass text as the main form of AI-generated content and the primary driver of future compute needs. A significant portion of this content could be created on edge devices. Media generation combined with autonomous AI agents and machines will usher in the next phase of AI development.

Final Thoughts

Unlike previous technology adoption cycles, generative AI shows no signs of slowing. Rapid advances in both capability and efficiency are fueling increased demand. As agentic AI expands beyond human interaction, the number of "users" of generative AI will grow exponentially.

I will explore the rising AI demand for images, video, autonomous agents, and autonomous machines, along with the global infrastructure requirements and total cost of operation (TCO) of generative AI in upcoming articles.

The above is the detailed content of The Unstoppable Growth Of Generative AI (AI Outlook Part 1). For more information, please follow other related articles on the PHP Chinese website!