Learning Hub

Moore's Law

Transistors Per Chip Kept Compounding For Decades

In 1965, Gordon Moore observed that the number of components on an integrated circuit was growing exponentially. By 1975 he refined the rule of thumb to a doubling roughly every two years. The exact cadence varied, but the strategic point held: more computation kept arriving at roughly the same or lower cost.

That mattered because every layer above hardware could be more ambitious. Better chips enabled personal computing, the web, smartphones, cloud software, and finally the deep learning era. Moore's Law was not just a semiconductor story. It was the engine underneath modern digital abundance.

Representative public chip milestones on a log scale. The line is the point: compounding hardware made later AI breakthroughs economically possible.

Kurzweil

Ray Kurzweil And The Idea Of Accelerating Returns

Kurzweil's core argument is that important technologies do not advance in a single straight line. They move through overlapping S-curves. When one paradigm nears its limits, another emerges and continues the broader exponential trend. In that framing, evolution, chips, networks, and AI are not separate stories. They are stacked accelerations.

Whether you agree with every Kurzweil forecast is not the main point. The useful takeaway is that capability jumps often look slow for years, then obvious in hindsight. When multiple compounding curves align at the same time, progress feels less linear and more like a wall arriving fast.

The useful intuition: accelerating change often comes from overlapping paradigms, not one technology marching alone forever.

Compute Over Time

Training Compute Has Been Climbing Faster Than Classic Chip Progress

Moore's Law explains why hardware improved. Modern frontier AI adds another layer: organizations are also throwing vastly more total compute at training runs. Public estimates suggest that landmark systems moved from roughly 10¹⁷ FLOP-scale training runs in the early 2010s to around 10²⁵ FLOP-scale territory by the early 2020s.

This is why the public experience of AI can feel sudden. Capability did not only improve because models got smarter. It improved because architecture, data, infrastructure, and capital all compounded together.

Public-estimate trendline on a log scale. The exact numbers vary by source, but the macro pattern is clear: frontier AI compute rose extraordinarily fast.

The Hyperbolic Feeling

Why The Future Feels Like It Is Arriving In Bursts

People describe these years as hyperbolic because multiple curves are stacking at once. Chips kept improving. Cloud made compute rentable. The web created gigantic data exhaust. Better algorithms arrived. Capital concentrated around winners. Distribution moved instantly. When those curves reinforce one another, change stops feeling steady.

This is the emotional truth of the present moment: compounding systems look manageable until they do not.