Conference Presentation, Fireside Chat, Panel
Fractile AI, Rogo, Scribble VC, San Francisco Compute: The AI Trillion-Dollar Opportunity
- San Francisco Compute aims to launch an accessible compute marketplace, engage in US and California regulatory discussions to mitigate risks through industry input, and sustain operations by modeling downside risks tenfold while pursuing deep tech moats; the company anticipates non-technical teams accelerating infinitely to delay specialist hiring, though complex orchestration skills for VMs and Kubernetes will remain in demand.
- Scribble Ventures plans to act as a first-check investor for the top 1% of founders, expecting the rapid evolution of model technology to force daily product reinvention while predicting that top talent will remain the critical factor for success despite fears of excessive headcount reductions.
- Fractile intends to develop a novel silicon platform running frontier models 50 times faster and 10 times cheaper than flagship GPUs within a couple of years, following a 12-to-18-month architecture phase and a subsequent six-month shipping window for a total close to two-year timeline; the company forecasts a shift from tokens-per-second pricing to quality-of-service monetization as the world navigates supply chain separations across the US, Europe, and China.
- Rogo expects to achieve product credibility suitable for scrutiny within a two-year horizon, with an 18-month roadmap focused on building and establishing trust, while projecting that AI research will rely on tools for thought partnership and that senior banking skills in market dynamics will grow in relevance as finance shifts from manual calculation to AI-driven processes.
- Elizabeth Weill forecasts that within a decade, ubiquitous access to personal private tutors, memory retrieval via AI, personalized medical prescriptions, and advancements in robotics and space exploration will transform society, while noting that organizational structures will likely remain familiar despite the current era possessing the "dumbest" models ever used.
- Walter Goodwin posits that compute serves as the ground truth for AI progress, anticipates a future where humanity shifts focus to energy consumption efficiency similar to the industrial revolution, and expects the first silicon deployed in data centers to have a lifecycle of three to six years.