Moore's Law vs Wright's Law

Time vs. Experience: The Fundamental Divergence

Moore's Law is a clock. It predicts that improvements happen on a schedule, driven by the semiconductor industry's massive R&D investments and lithography roadmap. This made it an extraordinarily useful planning tool: product managers could design hardware two years out knowing roughly what transistor budgets they'd have. Wright's Law is a flywheel. It predicts that improvements are earned through cumulative production — every unit built teaches the system something, from yield optimization to design simplification.

The practical difference becomes clear when production rates vary. During the PC boom of the 1990s, semiconductor production grew so steadily that Moore's Law and Wright's Law produced nearly identical predictions. But when a technology's adoption curve is nonlinear — as AI inference is today — Wright's Law captures the acceleration that Moore's Law misses. AI inference costs fell roughly 280× in two years not because of a ticking clock, but because hyperscalers deployed at unprecedented scale, driving the exact learning-curve dynamics Wright described in 1936.

Where Moore's Law Still Reigns

Moore's Law remains the best framework for understanding semiconductor roadmaps. Intel's push into the Angstrom Era — with 20A (2nm) and 18A (1.8nm) nodes launching in 2025–2026 — and the industry's target of one trillion transistors per package by 2030 are fundamentally Moore's Law milestones. Gate-all-around (GAA) transistor architectures like Intel's RibbonFET and TSMC's nanosheet designs are engineering responses to keep Moore's trajectory alive. Advanced 3D packaging, where TSMC plans to stack electrical and photonic dies in 2026, extends the law's spirit even when planar scaling stalls.

For hardware architects designing the next generation of GPUs or AI accelerators, Moore's Law — in its evolved, packaging-inclusive form — still sets the envelope of what's physically possible. Huang's Law, which tracks GPU performance gains, is itself built on top of Moore's transistor budgets combined with architectural innovation.

Where Wright's Law Dominates

Wright's Law is the superior model for any technology where production volume is the primary driver of cost reduction. Solar panels have followed an ~24% learning rate for decades. Lithium-ion batteries have tracked a ~18% rate. And AI inference has exhibited perhaps the steepest Wright's Law curve in history: GPT-4-class performance cost $60 per million output tokens in March 2023; by early 2025, equivalent capability was available for under $0.40 — a curve that maps cleanly to the explosive growth in cumulative inference volume.

For platform strategists in the virtual economy, Wright's Law provides the actionable insight: the way to make AI-powered features affordable isn't to wait for better chips (Moore thinking) but to deploy at scale now and ride the learning curve down (Wright thinking). This is why companies like OpenAI and Anthropic price aggressively — they're making a Wright's Law bet that volume will pull costs below current pricing.

The AI Inference Inflection

The most dramatic current illustration of the Moore-vs-Wright divergence is AI inference pricing. Between 2023 and 2026, inference costs collapsed by orders of magnitude — not primarily because GPUs got more transistors (Moore), but because of cumulative deployment experience: quantization techniques, speculative decoding, mixture-of-experts architectures, model distillation, and infrastructure optimization. These are classic Wright's Law mechanisms — process innovations driven by production experience.

The data is striking: the median rate of AI inference cost decline recently accelerated from roughly 50× per year to 200× per year. This pace far exceeds what semiconductor scaling alone could deliver. It represents Wright's Law operating at AI speed, where "production experience" accumulates in months rather than decades because AI systems can simulate and optimize millions of inference runs digitally.

Strategic Implications for Platform Builders

The choice between Moore's Law thinking and Wright's Law thinking has concrete strategic consequences. A Moore's Law strategist waits: they delay adoption until the next hardware generation delivers better price-performance. A Wright's Law strategist scales: they deploy now, accept current costs, and bet that their cumulative production experience will create a cost advantage competitors can't easily replicate.

In the agentic economy, Wright's Law thinking dominates. The platforms accumulating the most AI agent deployment experience — processing the most transactions, handling the most edge cases, optimizing the most inference pipelines — are building structural cost advantages that compound over time. This is precisely the dynamic behind deflationary technology: costs fall not on a schedule but in proportion to cumulative adoption.

Limitations and the Emerging Synthesis

Neither law is complete. Moore's Law fails to account for demand dynamics — it assumes the industry will keep investing regardless of market conditions. Wright's Law assumes a stable learning rate, but 2025 research from construction economics suggests learning rates shift over time, making long-range forecasts unreliable. For hardware specifically — transistors, DRAM, laser diodes — Wright's Law's traditional formulation still outperforms alternatives.

The emerging synthesis treats Moore's Law as setting the physical ceiling (what's possible given semiconductor technology) and Wright's Law as governing the economic trajectory (how fast costs actually fall given real-world adoption). Both are necessary for understanding the full picture of exponential growth in technology. The smartest strategists use Moore to plan chip designs and Wright to plan business models.