A Vision Talk - Deep Space Preset

The next decade of AI will be built on constraints, not compute

Why the companies that win the 2030s will optimize for efficiency, trust, and specificity rather than raw model size.

Press N for speaker notes - Arrow keys to navigate
The scaling wall
100x

more compute for only 2x better performance

The diminishing returns curve of raw model scaling, 2024-2026
Three tectonic shifts

Where the value migrates next

  • From parameters to data quality: curated, domain-specific data beats raw internet scale
  • From general to specific: vertical models that know one thing deeply beat horizontal models that know everything shallowly
  • From inference to evaluation: the bottleneck moves from generating answers to knowing which answers are correct
The future of AI is not a bigger brain. It is a more disciplined one.
The Constraint Thesis
The opportunity

Build for the constraint, not the capability.

The next wave of AI companies will not be the ones with the most GPUs. They will be the ones with the best data, the tightest feedback loops, and the deepest domain expertise.

Start specific