Conference Presentation, Fireside Chat, Lecture, Interview
Stanford MS&E435 Economics of the AI Supercycle | Spring 2026 | The GPU Economy
- Global GDP is projected to double every 25 years, with technology's share rising significantly above the current 13% within 10 years and tech company earnings per share compounding at 15% annually compared to 6% for non-tech sectors.
- Artificial Intelligence is forecast to accelerate societal and economic progress at 10x the impact of the Industrial Revolution over a decade, with inference time reasoning demand predicted to increase by 1 billion times.
- Token consumption for reasoning models is expected to rise parabolically, driven by agents that will explode usage by an order of magnitude as they transition to taking action, potentially reaching tens of trillions of tokens consumed weekly globally.
- Grok Cloud is anticipated to grow to 4 million users from a few hundred thousand in weeks, outpacing NVIDIA's 17-year trajectory to 7 million users, while a Grok and NVIDIA partnership aims to deliver 2.5x more tokens per power footprint.
- NVIDIA is expected to reach $10 trillion in market value with $1 trillion in sales booked over the next eight quarters, though it faces competition from Cerebras, Grok, Tranium, and TPUs, with demand exceeding current supply capabilities.
- Anthropic is projected to scale revenue exponentially, having added $10 billion in annualized revenue in a single month, with new models like Mythos (TPU v7) and SPUD (Blackwell-trained) released soon.
- Unit costs for intelligence are expected to continue plummeting, having already dropped 90% in the last year and 99% over 2.5 years, despite training costs increasing 3x annually for models up to 100 billion parameters.
- AI costs are projected to fall further despite Moore's Law limits, while larger models ranging from 1 trillion to 10 trillion parameters may drive price increases for hardware like the H100 due to supply-demand imbalances.
- Gross margins for AI firms are expected to shift from negative to very positive as intelligence capabilities improve and willingness to pay increases, with the market potentially absorbing competition from custom ASICs without diminishing NVIDIA's dominance.
- Hardware constraints, such as battery life and compute intensity, are expected to limit the deployment of large models on edge devices, necessitating a shift of heavy computation to the cloud, particularly for Apple.
- The labor market is expected to shift from manufacturing and farming to the service economy, creating roles in coaching, care, and specialized services, while the value of Emotional Intelligence (EQ) rises relative to IQ.
- The "age of abundance" driven by AI is expected to generate massive wealth accumulation alongside distribution challenges, with the "Invest America" initiative playing a key role in addressing these economic imbalances.
- AI capabilities are projected to surpass collective human intelligence and eventually near an end to their exponential growth, requiring active societal engagement, policy adjustments, and social contract changes similar to past revolutions.
- Security risks are expected to harden rapidly, leading to rigorous sandboxing and testing of models like Mythos before public release, while harnesses are designed to keep models active in loops for continuous value extraction.
- Future AI development involves balancing optimistic possibilities for energy and abundance with risks of destructive applications, where industry leaders like Amodei, Altman, and Musk signal the potential end of exponential capability growth.