The 3D Semiconductor Revolution

IBM has announced a major shift in semiconductor manufacturing with its new sub-1nm chip technology. For over 60 years, the industry has relied on 2D scaling (XY-plane lithography). IBM’s new "nanostack" architecture introduces vertical (Z-direction) stacking of transistors. By staggering these nano-sheets, IBM can optimize top and bottom devices independently, allowing for better material selection and direct contact with signal and power lines. This design provides 50% better performance or 70% power savings compared to current 2nm chips, while also offering 40% area scaling, which is critical for embedding more memory (SRAM) alongside AI computing units.

The Rise of Multi-Model Orchestration

The panel discussed the emergence of new models like Sakana’s Fugu and Z.ai’s GLM-5.2, which challenge the dominance of frontier labs. A key insight is that the industry is moving away from the "single model as the product" paradigm toward "orchestration as the product." Sakana’s Fugu, for instance, acts as a router that dynamically selects the best model for a specific task.

Panelists noted that while these models post impressive benchmark scores, they are essentially "assemblages" of existing frontier capabilities. This approach offers enterprise resilience against model fluctuations but introduces new forms of non-determinism, as the quality of the output depends on the effectiveness of the routing logic rather than a single model's weights. The panel suggests that the real innovation lies in these virtual model endpoints, which could eventually be scaled down to run on commodity hardware like smartphones or laptops.

Tokenminning vs. Tokenmaxxing

The conversation highlighted a shift in AI engineering philosophy: moving from "tokenmaxxing" (throwing as many tokens as possible at a problem) to "tokenminning" (optimizing for efficiency). The panelists argued that the most important metric for AI adoption is not just raw performance, but the ability to achieve high-quality results with a smaller footprint. This efficiency is necessary for enterprise adoption, where security and data privacy constraints often prevent developers from using massive, cloud-hosted frontier models.

AI in Creative Industries

The panel touched upon the collaboration between Google DeepMind and A24, noting that the integration of AI into entertainment is becoming a standard operational reality. The consensus is that the industry is moving past the "shock" phase of AI capabilities and into a phase of integration, where AI tools are used to augment creative workflows rather than just generate content in isolation.