The Agentic Post
Breaking
Gemini’s Multimodal Features, Explained  ·  ChatGPT Custom GPTs, Explained  ·  What Is Constitutional AI? Explained  ·  AI Capex Explained for Investors  ·  AI Startup Valuations: How They Are Set  ·  How to Reskill for an AI Job Market  ·  
Home/Hardware & Robotics/Chips & GPUs
Why Are AI Chips So Expensive?

Why Are AI Chips So Expensive?

Chips & GPUs

An explainer on why AI chips cost so much, covering manufacturing precision, specialized memory, and persistent supply-demand imbalance.

AP
Staff
September 8, 2026 · 1 min read

The price tag on leading-edge AI chips reflects genuine, compounding costs at every stage, not just margin.

Our explainer on how AI chips are made covers why: modern chips are etched at nanometer-scale precision using photolithography equipment that costs enormous sums per machine, and yield is never 100%. Every failed chip on a wafer raises the effective cost of the ones that work.

High-bandwidth memory, needed to feed data to AI chips fast enough, is itself expensive and complex to stack onto the package. Supply has repeatedly struggled to keep pace with demand.

Basic economics plays a role too. Every major lab wants the newest generation as fast as possible, and manufacturing capacity for leading-edge chips takes years to build out, creating persistent scarcity that keeps prices elevated well beyond raw production cost.

Chip cost flows directly into every API call and cloud instance built on it, part of why aggressively cheap models stand out so much when they appear. See TSMC’s own site for more on the manufacturing side.

Related

TPU vs GPU: What Actually Differs
Chips & GPUs

TPU vs GPU: What Actually Differs
Figure AI Bets $6 Billion That Robots Need More GPUs, Not More Data
Chips & GPUs

Figure AI Bets $6 Billion That Robots Need More GPUs, Not More Data
AI Chips and GPUs in 2026, Explained
Chips & GPUs

AI Chips and GPUs in 2026, Explained
Up Next
What Is a Token in AI? Explained Simply

What Is a Token in AI? Explained Simply

AI Glossary

A plain-English explainer on what a token actually is in AI, why pricing and context windows are measured in tokens, and how tokenizers affect cost.

AP
Staff
September 8, 2026 · 1 min read

“Token” is one of the most foundational terms in AI, and one of the least explained in plain language, despite showing up on every pricing page.

A token is the basic unit a model reads and generates, roughly a word or a word-fragment. “Understanding” might be one token or split into “under” and “standing” depending on the tokenizer. Rough rule of thumb: a token averages about four characters of English text.

Tokens, not words or characters, are what a model actually processes, so pricing, context limits, and generation speed are all measured in tokens. The same text can cost slightly different amounts across models, each has its own tokenizer splitting text differently.

Our coverage of Claude Sonnet 5’s pricing change flags a real consequence: a new tokenizer can count meaningfully more tokens for the same text than an older one did. An unchanged per-token rate can still mean a real cost increase in practice.

A model’s context window is measured in tokens too, the total text it can consider at once. Try OpenAI’s own tokenizer tool to see this in action.

Powered by