Road to AI 02: Transistors and ICs, the Origin of AI Cost Curves

Why episode 2 matters

Episode 1 covered the birth of computing.
Episode 2 answers a practical question: why did capability rise so fast while cost kept dropping across decades?

The core answer is the transistor and the integrated circuit (IC).

The real shift: size, power, reliability

Vacuum-tube machines were large, hot, and fragile.
Transistors changed the economics of computing by making systems smaller, cooler, and more stable.

Three structural changes

1. Miniaturization: more compute components in the same physical space
2. Power efficiency: lower operating cost and thermal burden
3. Reliability: fewer failures and better service continuity

This is when computing began to move from lab-grade infrastructure to product-grade infrastructure.

Timeline: from transistor to chip era

Year	Event	AI-relevant meaning
1947	Transistor invented	Practical electronic compute accelerates
1958-59	Integrated circuit emerges	Complex circuits compressed into chips
1965	Moore's law articulated	Performance growth becomes an industry roadmap
1971	Commercial microprocessor	Foundation for mass, general-purpose computing

Why this still controls modern AI costs

Today's AI operations still follow the same logic:
deliver similar or better quality with less compute and more stable execution.

Link 1: compute unit cost

Higher integration lowered cost per operation over time.
That long curve made modern large-scale training and inference economically possible.

Link 2: power and cooling

If efficiency is weak, service unit economics break immediately.
That is why GPU choice, batching strategy, and quantization matter in LLM production.

Link 3: reliability and operability

AI products run continuously under variable load.
Failure rate, recovery time, and burst handling are product quality factors, not just infra metrics.

Operator checklist

1. Split KPI tracking into model quality and infra efficiency (latency/cost).
2. Measure before/after cost for every model change (token cost + average latency).
3. Audit hardware concentration risk (single accelerator, single region, single vendor dependence).

One-line summary

The transistor and IC era made "more compute in less space" normal,
and that same principle now appears as the core AI question: higher quality at lower cost.

Next episode

Episode 3 will cover how operating systems and software engineering practices determine AI product stability and shipping speed. ai-evolution-chronicle-02-transistor-and-ic 2026-02-10 ai_road_104720dc evolution_to_1147226f chronicle_ai_12472402 02_02_13472595 transistor_transistors_c471a90 and_and_d471c23 ic_ics_e471db6 ai_the_f471f49 evolution_origin_18472d74 chronicle_of_19472f07