Evening Panel

Joint Technology/Circuits Panel – Tapa 1-3

Tuesday, June 15, 8:00 PM – 10:00 PM

Organizers

AI: Grand Vision or Grand Delusion?

Moderators

Gary Bronner, Rambus
Vita Pi-Ho Hu, National Taiwan University

The AI world stands at a critical inflection point.  Ambitious scaling goals are colliding with fundamental  physical and economic constraints. OpenAI’s call for $1T  level  investments in  AI infrastructure will require massive investment in new datacenters,  fab capacity and  energy generation.  There is tension between the desired growth trajectories and manufacturing feasibility, highlighting a broader disconnect between AI aspirations and infrastructure realities. While inference represents the primary monetization opportunity for AI systems, the massive capital expenditure required for data center expansion raises questions about optimal deployment strategies.  Will 100x scaling in AI datacenters deliver 100x benefit? Power is a particular concern with targets of hundreds of GW of power needed by the 2030s and each GW of power needing the equivalent of a nuclear reactor.   Is today’s AI ready for such investment?  Is building out AI datacenters at this scale the right move or is more research needed to improve AI efficiency.  Will scaling compute in support of Large Language Model (LLMs) take AI to the next level?  Or are we setting ourselves up for an AI winter where AI won’t be able meet its promise.  Are we living through a “dot.AI” bubble, similar to the “dot-com” bubble of the internet era?  Or are those concerns overblown? As the industry grapples with these competing forces, the path forward requires careful consideration of whether current scaling strategies can overcome fundamental physical limitations or whether paradigm shifts in computational approaches will be necessary to realize AI’s transformative potential.

Panelists