---
id: "claim-compute-scaling-rate"
type: "claim"
source_timestamps: ["§ Transforming Moats"]
tags: ["compute", "hardware", "scaling-laws"]
related: ["concept-recursive-algorithmic-development", "entity-nvidia-blackwell", "entity-meta-llama-4"]
confidence: "high"
testable: true
speakers: ["Toby E. Stuart"]
sources: ["futures"]
sourceVaultSlug: "hbr-seg-futures"
originDay: 2
articleStem: "hbr-nm-99-genai-end-incumbent-advantage"
sourceUrl: "https://hbr.org/2024/11/could-gen-ai-end-incumbent-firms-competitive-advantage"
sourceTitle: "Could Gen AI End Incumbent Firms’ Competitive Advantage?"
---
# AI Compute Is Scaling at 4× Moore's Law

**Claim:** The computational power devoted to training AI models has doubled approximately every **6 months** over the last decade — explicitly **four times the rate** of advancement in underlying semiconductor technology described by **Moore's Law**. The blistering pace is illustrated by **AlexNet (2012)**, which took *a week* to train on *two GPUs* and could now be trained in **about 5 minutes on a single state-of-the-art [[entity-nvidia-blackwell|NVIDIA Blackwell GPU]]**. The scale is further shown by [[entity-meta-llama-4|Meta Llama 4]] training on a cluster of **more than 100,000** state-of-the-art GPUs. This pace underwrites [[concept-recursive-algorithmic-development|recursive algorithmic development]].

**Confidence: high · Testable: yes.**

**Enrichment / Validation.** *Directionally supported*: training compute for frontier models has grown far faster than Moore's Law (roughly doubling every 3–6 months across 2012–2018 per widely cited analyses), and clusters of tens of thousands of GPUs are already in use. The *exact* figures — "doubling every ~6 months for a decade" and "5 minutes on a Blackwell GPU" — should be treated as illustrative/extrapolated rather than rigorously documented in the enrichment search set. The Llama-4 100,000-GPU cluster is forward-looking; large clusters are credible, but the specific numbers and model naming are unvalidated. Counter-point: energy, capital, and supply-chain constraints may slow future scaling.