Conference Presentation, Keynote
Yuri Frayman, Co-Founder & CEO, Cast AI: Robots Will Feed the World… If We Build the Rails
- Vision & Core Thesis: SoftBank Vision Fund portfolio companies aim to build the infrastructure ("rails") required for robots to feed the world, enabling AI to transition knowledge work into the physical world.
- Economic Projection: AI is already redefining every industry, with a long-term trajectory toward global prosperity similar to the adoption of steam, electricity, and the internet.
- Employment Outlook: While the fear of mass job displacement is a recurring myth, historical technological shifts suggest a need for workforce re-education rather than job destruction.
- Critical Resource Constraints:
- Energy: Unrestricted energy generation is a prerequisite for AI abundance; all potential sources (e.g., windmills) must be utilized to produce the necessary "microns" of power.
- Compute: The proliferation of billions of intelligent devices generating petabytes of data requires massive GPU expansion for cloud-based fine-tuning and training.
- Regulatory & Competitive Landscape:
- Emerging regulations (e.g., "big beautiful bill" in the US) may inadvertently favor well-funded commercial labs over smaller players if mandates vary by state.
- Forward-looking strategy: Anticipated emergence of open-source or decoupled models to bypass state-specific regulatory fragmentation and ensure competition against billion-dollar entities.
- Infrastructure Gaps:
- Workforce: Despite rhetoric about self-coding, a significant shortage of AI/ML engineers remains; the supply of human engineering talent must be increased to sustain future innovation.
- Scale: Planetary-scale infrastructure projects, such as Project Stargate, are necessary to support purpose-built, GPU-intensive platforms.
- CAS Inference Architecture:
- Supply/Demand Mechanism: A cloud-agnostic model that invisibly sources affordable GPUs for applications in real-time, ensuring costs remain efficient and friction is minimized.
- Security Requirement: Data and models must remain within a private control environment to maintain trust and security during distributed inference.
- Future Roadmap: The development of these open, fast, and fair infrastructure rails over the next decade will determine which entities benefit from AI abundance and which are left behind.