Jensen Huang emphasizes that Moore's Law is dead as a primary performance driver. The 50x gain from Hopper to Blackwell came overwhelmingly from architecture and computer science breakthroughs, with raw transistor improvements providing only marginal benefit.

The current 3x annual growth in AI compute is not easily accelerated and may be unsustainable. It is fundamentally constrained by the slowing of Moore's Law (1.4x), the fixed production rate of ASML's EUV machines for new fabs (1.2x), and the near-total absorption of leading-edge wafer capacity from other sectors (1.8x).

While AI model providers may overstate demand, the most telling signal comes from TSMC. Their decision to significantly increase capital expenditure on new fabs, a multi-year and irreversible commitment, indicates a strong, cynical belief in the long-term reality of AI compute demand.

The memory market is projected to see unprecedented growth, with 2026 revenues expected to increase by roughly $600 billion in a single year. This incremental growth alone is larger than the entire annual market for smartphones, PCs, or servers, highlighting the massive economic shift driven by AI infrastructure.

The critical constraint on AI and future computing is not energy consumption but access to leading-edge semiconductor fabrication capacity. With data centers already consuming over 50% of advanced fab output, consumer hardware like gaming PCs will be priced out, accelerating a fundamental shift where personal devices become mere terminals for cloud-based workloads.

In a striking economic anomaly, the cost to rent older NVIDIA H100 AI chips is increasing, not decreasing. This is because the growth in AI's usefulness is outstripping the tripling annual supply of compute. It signals that the value being generated by AI models is growing faster than our ability to manufacture the hardware to run them.

A counterintuitive view of Moore's Law is that for it to hold, the economic value of computation must halve every 18 months because we historically run out of uses for it. The recent rise in H100 GPU rental costs suggests AI is the first application where demand is growing faster than supply, breaking this trend.

Unlike railroads or telecom, where infrastructure lasts for decades, the core of AI infrastructure—semiconductor chips—becomes obsolete every 3-4 years. This creates a cycle of massive, recurring capital expenditure to maintain data centers, fundamentally changing the long-term ROI calculation for the AI arms race.

The history of computing has cycled through different constraints. It began as capital-bound (acquiring expensive mainframes), shifted to engineering-bound with the rise of software, and has now returned to being capital-bound due to the massive compute costs for training AI models. This historical context reframes today's landscape.