This is part of our series on the 11 Structural Shifts Reshaping AI in 2026, analyzing the trends that will define artificial intelligence this year.
Jensen Huang’s opening framework—”AI Scales Beyond LLMs”—identified three distinct scaling laws, each requiring enormous compute.
The Three Scaling Laws
1. Pre-Training Scaling (2015-2022)
The original scaling law. BERT, Transformers, GPT series—teach models to understand through massive data exposure. More data, more parameters, more compute yields more capability. This drove the initial AI revolution.
2. Post-Training Scaling (2022-2024)
RLHF and skill acquisition. Transform raw language capability into useful, aligned behavior through human feedback. This enabled ChatGPT and commercial AI products—models that could be helpful, harmless, and honest.
3. Test-Time Scaling (2024+)
The breakthrough Jensen highlighted as “revolutionary.” OpenAI’s o1 model introduced reasoning at inference—models that think before responding. They use more compute at runtime to solve harder problems. DeepSeek R1 proved this capability can be open-sourced.
This is System 2 thinking—deliberate, effortful reasoning rather than quick pattern matching.
No Plateau in Sight
Jensen’s point: each scaling law requires enormous compute, and the curves continue climbing. There is no plateau in sight.
Throughout 2025, test-time scaling dominated the research agenda. By 2026, extended reasoning chains are standard in production systems.
The Infrastructure Implications
Intelligence improves along three dimensions simultaneously:
- Pre-training (knowledge)
- Post-training (alignment)
- Test-time (reasoning)
Each dimension scales with compute. The frontier keeps moving.
Strategic Implications
Test-time compute is the current frontier. Models that “think longer” perform better—but consume more inference compute.
This creates new infrastructure demand beyond training:
- Inference clusters supporting extended reasoning chains
- Memory systems for multi-step deliberation
- Cost optimization for variable compute per query
The economics of AI now include inference scaling, not just training scaling.
The Bottom Line
Intelligence scales along three dimensions simultaneously. Each dimension scales with compute. The frontier keeps moving—and test-time scaling is where the action is in 2026.
Read the full analysis: 11 Structural Shifts Reshaping AI in 2026









