Eric Schmidt & Andrew Feldman: The Race to SuperIntelligence

On stage at RAISE Summit (Paris), Andrew Feldman (CEO & Co-founder, Cerebras Systems) joins Dr. Eric Schmidt to unpack how wafer-scale compute and system-level co-design are accelerating the path to frontier AI. Feldman reveals why Cerebras built the world’s largest chip (the WSE) to keep data on-chip and crush memory bottlenecks, while Schmidt frames the bigger picture: transformer-era scaling, software lock-ins, and the economics (and electricity) behind the next wave of intelligence.
‍

Key discussion points include:

• Hardware–software co-evolution: From sparse linear algebra to prompt caching—why end-to-end engineering matters.
• Wafer-scale advantage: Keeping activations and weights on a single die for massive throughput and ultra-low latency.
• Speed wins: In inference, latency = retention; tokens/sec becomes a north-star metric for real products.
• Beyond GPUs: Where specialized architectures excel for inference while GPUs dominate training.
• Scaling laws & S-curves: Deep learning, RL, and test-time compute—why demand may flip to inference-heavy.
• Open source & geopolitics: Adoption of open-weights models globally (incl. China) and what that means for competition.
• Sovereign AI: Lessons from large national deployments (e.g., UAE) and Europe’s opportunity—if compute and power are prioritized.
• Energy reality check: Cheap, clean electricity (incl. nuclear) as a gating factor for AI data centers and national strategy.
• Alignment & safety: The hard problem of testing fast-improving systems before release.

If you care about AI infrastructure, sovereign compute, and real-world inference at scale, watch this session. Like, comment, and subscribe for more expert conversations from the RAISE Summit.