Future 100T Param Model to Cost >$250B to Train, Says Joseph Jacks
JosephJacks_ · x · 2026-07-30
Tech observer Joseph Jacks shared an insightful observation regarding the future of AI compute: a future frontier AI model with 100 trillion parameters could require over $250 billion to train. However, due to hardware advancements, this massive model could eventually be run privately on 5,000x less hardware (equivalent to 1-200 TB of HBM memory, costing around a $50M inference cluster). He also noted that frontier model weights are the fastest depreciating digital assets in history.
More from AGI Musings
- Facebook AI Head Warned Deep Learning Would Hit a Wall in 2019—Still Waiting — haider1 · 2026-07-30
- Parallelization Bottlenecks Could Delay the Technological Singularity — Jsevillamol · 2026-07-30
- Academics More Willing Than AI Pros to Discuss Post-Human Future — danfaggella · 2026-07-30
- CFXS Paper: Using LLMs to Uncover Hidden Job Transition Paths — soumitrashukla9 · 2026-07-30
- Valar Atomics Founder: Cheap Energy Will Always Create Its Own AI Demand — No Priors · 2026-07-30
- Researchers Find Anomalous Narrative Fulfillment Tendencies in Claude Opus 5 Base Mode — repligate · 2026-07-30