Build This Now
Build This Now
Qu'est-ce que le code Claude ?Installer Claude CodeL'installateur natif de Claude CodeTon premier projet Claude Code
How LLMs WorkAI Image GenerationHow AI Agents WorkC'est quoi le agentic coding ? Un guide en français clairC'est quoi le vibe coding ? Un guide en français clairAI TokensVector EmbeddingsChatGPT Dreaming MemoryAI Browser InjectionAI Energy & WaterIs AI a BubbleEU AI ActAI Voice ScamsAgentic CommerceWhy AI Uses GPUsHow HTTPS Works
speedy_devvkoen_salo
Blog/Handbook/Core/Why AI Uses GPUs

Why Does AI Run on GPUs, Not CPUs? (One Genius vs. a Thousand Interns)

A CPU is a few brilliant workers doing tasks one at a time; a GPU is thousands of simple workers doing the same math all at once. AI is mostly that simple math at massive scale — here's why GPUs won.

Vous voulez le framework derrière ces projets ?

Obtenez le système Claude Code que nous utilisons pour planifier, construire, tester et livrer des logiciels en production.

Découvrez ce que nous construisons pour les entreprises →
speedy_devvkoen_salo
speedy_devvWritten by speedy_devvPublished Jun 13, 20267 min readHandbook hubCore index

AI runs on GPUs instead of CPUs because the core work of a neural network is a staggering amount of simple, repetitive math done all at once — and that's exactly what a GPU is built for. A CPU is like a few brilliant workers who each do complicated tasks one after another; a GPU is like thousands of simpler workers doing the same basic calculation in parallel. For AI, you don't need a few geniuses — you need a thousand interns all multiplying numbers at the same time. That single mismatch is why Nvidia became one of the most valuable companies on earth.

Here's the intuition, no engineering degree required.

Table of Contents

  1. What AI Actually Computes
  2. CPU vs. GPU: The Core Difference
  3. Why AI Is a Perfect Fit for GPUs
  4. Why This Made GPUs Scarce and Nvidia Huge
  5. Frequently Asked Questions

What AI Actually Computes

Underneath the magic, a neural network is mostly multiplication and addition — billions of tiny numbers (the model's "weights") multiplied against your input and summed up, over and over. There's no single hard calculation. There are enormous numbers of trivial ones, and they can mostly be done independently of each other.

That last part is the key: if a million little multiplications don't depend on each other, you don't have to do them one at a time. You can do them all at once — if your hardware can.

CPU vs. GPU: The Core Difference

Both are chips full of "cores" that do math. The difference is the trade-off each makes:

CPUGPU
CoresA few, very powerfulThousands, individually simpler
Best atComplex tasks, one after anotherThe same simple task, massively in parallel
AnalogyA few geniuses working sequentiallyA thousand interns working simultaneously
Wins whenWork is varied and step-by-stepWork is uniform and parallel

A CPU is a generalist: great at running your operating system, a browser, a game's logic — lots of different, sequential decisions. A GPU was originally built for graphics, which means coloring millions of pixels with the same kind of math at once. Turns out that "same math, millions of times, in parallel" is also the shape of AI.

Why AI Is a Perfect Fit for GPUs

Picture adding up a million pairs of numbers.

  • On a CPU (say, 8 powerful cores), you do them in big batches but still largely in sequence — fast, but fundamentally a line.
  • On a GPU (thousands of cores), you hand one addition to each core and they all finish at nearly the same moment.

Neural networks are made of exactly this kind of bulk parallel arithmetic (technically, matrix multiplication). So a GPU can be tens or hundreds of times faster than a CPU for AI — not because each GPU core is smarter, but because thousands of them work at once. Use the wrong tool and training a model that takes days on GPUs could take months on CPUs.

Why This Made GPUs Scarce and Nvidia Huge

Once everyone realized AI's appetite is essentially "as many parallel math units as you can buy," demand for GPUs exploded — and Nvidia, which makes the dominant AI GPUs and the software ecosystem around them, became the picks-and-shovels supplier of the AI gold rush. That's why GPU supply, data-center buildouts, and Nvidia's revenue are constant news, and why they sit at the center of the AI bubble debate.

It's also tied to why AI uses so much energy: thousands of cores running flat-out draw enormous power and throw off enormous heat. The very thing that makes GPUs fast for AI is what makes AI data centers power-hungry.

Frequently Asked Questions

Why does AI use GPUs instead of CPUs?

Because AI's core work is a massive amount of simple, repetitive math that can be done all at once. GPUs have thousands of cores built to run the same calculation in parallel, while CPUs have a few powerful cores built for varied, sequential tasks. AI fits the GPU's strength almost perfectly.

What's the difference between a CPU and a GPU?

A CPU has a few very capable cores optimized for complex, step-by-step work — like running your operating system. A GPU has thousands of simpler cores optimized for doing the same operation on lots of data simultaneously, like coloring millions of pixels or multiplying millions of numbers for AI.

Is a GPU always faster than a CPU?

No — only for work that's highly parallel, like AI math or graphics. For varied, sequential tasks (most everyday computing), a CPU is better. GPUs win specifically when you have huge numbers of similar calculations that don't depend on each other.

Why is Nvidia so important to AI?

Nvidia makes the dominant GPUs used to train and run AI, plus the software ecosystem developers rely on. As AI demand exploded, so did demand for its chips, making Nvidia the key supplier of the AI boom — which is also why its sales feature heavily in bubble debates.

Why do AI GPUs use so much electricity?

Because thousands of cores running at full speed draw a lot of power and generate a lot of heat, which then needs cooling. The same parallel design that makes GPUs fast for AI is what makes large AI data centers so energy- and water-intensive.

Continue in Core

  • La Fenêtre de Contexte 1M dans Claude Code
    Anthropic a activé la fenêtre de contexte 1M tokens pour Opus 4.6 et Sonnet 4.6 dans Claude Code. Sans header beta, sans surcharge, tarification fixe, et moins de compactions.
  • AGENTS.md vs CLAUDE.md : expliqué
    Deux fichiers de contexte, une seule base de code. Comment AGENTS.md et CLAUDE.md diffèrent, ce que chacun fait, et comment utiliser les deux sans rien dupliquer.
  • Why a Hidden Line of Text Can Hijack Your AI Browser
    AI browsers read the whole web page — including text hidden from you. That's the door behind prompt injection, OWASP's #1 AI security risk in 2026. Here's how the attack works, in plain English.
  • AI Research for Builders: The Latest Breakthroughs, Explained Monthly
    A monthly digest of the latest AI research — agents, reasoning, efficiency, and models — with every claim traced to its source and translated into what it means if you build with AI.
  • 15 AI Research Breakthroughs (June 2026)
    The latest AI research, explained: DeepSeek shipped DSpark and a million-token V4, open coding models closed the gap, AI disproved an 80-year-old math conjecture, and inference costs kept dropping. What each finding means if you build with AI.
  • Did Anthropic Call for an AI Pause? What It Actually Said
    Anthropic did not call to halt the AI boom. Here is what its June 2026 'recursive self-improvement' post actually said, why the 80%-of-its-own-code stat spooked it, and what it means if you build with Claude Code.

More from Handbook

  • Techniques de réflexion approfondie
    Des phrases déclencheurs comme think harder, ultrathink et think step by step poussent Claude Code en raisonnement étendu et en plus de calcul au moment du test, même modèle.
  • Modèles d'efficacité
    Les frameworks de permutation transforment 8 à 12 builds manuels en un template CLAUDE.md que Claude Code utilise pour générer les variations 11, 12 et 13 à la demande. Capturé une seule fois.
  • Le mode rapide de Claude Code
    Le mode rapide route tes requêtes Opus 4.6 sur un chemin de service prioritaire dans Claude Code. Mêmes poids, même plafond, réponses 2,5x plus vite à un tarif token plus élevé.
  • Optimisation de la vitesse
    Le choix du modèle, la taille du contexte et la spécificité de l'invite sont les trois leviers qui décident de la rapidité des réponses de Claude Code. /model haiku, /compact, et /clear covered.

Vous voulez le framework derrière ces projets ?

Obtenez le système Claude Code que nous utilisons pour planifier, construire, tester et livrer des logiciels en production.

Découvrez ce que nous construisons pour les entreprises →
speedy_devvkoen_salo

Agentic Commerce

Agentic commerce is when an AI agent handles the whole purchase — finds the product, pays, and checks out — from a goal like 'order trail shoes under $150 that arrive Friday.' Here's how it works and who's building it.

How HTTPS Works

The padlock in your browser means your connection is encrypted — scrambled so only you and the website can read it. Here's how HTTPS works: the handshake, the two-key trick, and what it does and doesn't protect.

On this page

Table of Contents
What AI Actually Computes
CPU vs. GPU: The Core Difference
Why AI Is a Perfect Fit for GPUs
Why This Made GPUs Scarce and Nvidia Huge
Frequently Asked Questions
Why does AI use GPUs instead of CPUs?
What's the difference between a CPU and a GPU?
Is a GPU always faster than a CPU?
Why is Nvidia so important to AI?
Why do AI GPUs use so much electricity?

Vous voulez le framework derrière ces projets ?

Obtenez le système Claude Code que nous utilisons pour planifier, construire, tester et livrer des logiciels en production.

Découvrez ce que nous construisons pour les entreprises →