Podcast, Interview
SpaceX’s $2T Case, Nvidia’s Shock Selloff, America Turns on AI, Trump Pulls AI Order, Bond Crisis?
- Recursive self-improvement and continual learning are predicted to drive an order-of-magnitude annual improvement in model quality, potentially establishing a new form of Moore's Law where AI outpaces human iteration.
- Annual Recurring Revenue for private language model companies (Anthropic, OpenAI, Gemini, xAI) is forecasted to reach $200–400 billion by the end of the current year, supported by approximately 80% gross margins on inference.
- Current competitive advantages are estimated with Anthropic holding a three- to six-month lead over peers and a three- to twelve-month lead over open-source models, while data center deployment by SpaceX is expected to grow dramatically faster than anticipated.
- Cost per token is anticipated to halve due to architectural shifts toward smaller, networked systems, and a "minor breakthrough" is expected regarding orbital compute becoming a reality in the second half of 2028 to the first half of 2030.
- SpaceX is projected to achieve rapid Starship reusability within one to two years, enabling the sale of compute capacity to major clients like Google and OpenAI, with terrestrial data centers alone generating $100–200 billion in revenue by 2030–2032.
- Hardware lifecycles are expected to extend to 10–15 years for GPUs when combined with domain-specific accelerators, while AI technology diffusion is described as extremely asymmetric, creating power imbalances comparable to the nuclear era.
- Geopolitical risks include a potential credit crisis driven by inflation reaching 4–6% in Q2 and 10-year Treasury yields hitting 4.6% or higher, alongside scenarios where a Strait of Hormuz closure is relatively beneficial to the US due to energy self-sufficiency.
- Global power dynamics are predicted to shift based on US AI advancement, with a possibility of a 20- to 30-year arrangement regarding Taiwan to avoid conflict, while the US maintains a relative global advantage despite rising rates and inflation due to AI and energy strengths.
- Model improvement cycles may become seasonal due to user behavior, though this trend is expected to alter with the introduction of GenIC, and a future point is predicted where AI improves models more than human-in-the-loop processes.
- A global balance of power is considered necessary, as the speaker posits that if the United States does not advance its AI technology, other entities will, necessitating continued strategic competition.