ASIC

in English: application-specific integrated circuit

advancedwidespread

An ASIC is a chip designed for a single task or narrow range of tasks, so it performs that task much faster and with less energy than a general-purpose processor.

In plain terms

Your computer's processor is like a Swiss Army knife: good for almost everything, but the best at nothing. An ASIC, by contrast, is a scalpel: made for a single task, but unbeatable at it.

ASICs are designed for a specific task, such as video compression, network traffic routing, cryptocurrency mining, or running AI models. Since every unnecessary part is omitted, they perform the same work with less energy and faster.

In the AI world, this became important because training and running large language models requires vast amounts of electricity and money. Major tech companies are therefore designing their own AI-specific chips to avoid total dependence on expensive, general-purpose graphics processing units (GPUs).

The price is inflexibility: you cannot change the manufactured circuit afterward. An AI chip can run multiple models, but if the nature of the task changes fundamentally, a new chip must be designed.

How it works

Engineers precisely define what the chip must be able to do, such as performing a specific type of AI calculation very quickly.
The circuit is designed so that only the necessary components are included, arranged in the shortest and most efficient layout possible.
The plan is checked for months with simulations, because a circuit defect in the manufactured chip can only be fixed with a new, expensive production run.
Based on the final design, a chip factory (such as Taiwan's TSMC) manufactures the chips on silicon wafers, which is expensive and takes months.
Ready-made chips go into servers or devices, where they perform the same task for years with high efficiency.

An everyday example

When your phone recognizes your face in seconds or plays a high-resolution video without lag, this is typically done by a circuit designed for this purpose. Similarly, when you ask an AI assistant a question, your answer is increasingly generated on a custom AI chip in a data center.

Who is affected

everyonePurpose-built chips like these also work in your phone, router and car, and they partly determine how much AI services cost and how fast they are.

entrepreneur, executiveThe price and availability of AI services are strongly influenced by the chips on which providers run their models.

bank customer, investorA key question in how investors view chipmakers and tech companies developing their own chips is who wins the race for custom AI chips. → Money

industrial worker, engineerA significant share of the electronics built into machines, vehicles and measuring instruments is custom-designed, purpose-built chips, whose limitations need to be known. → Industry

A common misconception

MythAn ASIC is better than a GPU in every way, so it will soon completely displace it.

In realityASICs are superior only in the tasks they were designed for. Since AI models evolve rapidly, more flexible GPUs remain essential, especially in research and where tasks change frequently.

Limits and risks

It cannot be modified after production: if the task or the structure of AI models changes, the chip may become partially or fully obsolete.
Design is very expensive and usually takes years, so it is only worthwhile for large volumes or massive usage.
Only a few factories can produce the most advanced chips, making the supply fragile and exposed to geopolitical risks.
Speed and consumption figures provided by manufacturers often apply only to specific favorable tasks, so treat them cautiously.

Similar, but not the same

GPU (grafikus processzor) – A GPU is programmable for various parallel computations, while an ASIC is fixed to a single task type, though more efficient in that area.
FPGA (field-programmable gate array) – An FPGA circuit can be reprogrammed after manufacturing, while an ASIC's is final, but the latter is cheaper, faster, and more energy-efficient in large volumes.
CPU (central processing unit) – The CPU is general-purpose and runs any program, while the ASIC performs only the task it was designed for.

Related concepts

Broader

integrated circuit, semiconductor, AI-hardver, computing infrastructure

Narrower

AI accelerator, TPU, NPU, mining ASIC

On the edge

GPU, FPGA, data center, inference, rendszerchip (SoC)

A brief history

Custom-designed chips became widespread in the early 1980s, when manufacturers began offering customizable circuits. The general public first heard about them in the 2010s in connection with Bitcoin mining machines. In AI, the breakthrough came with Google's TPU chip: the company started using it in its own data centers from 2015 and publicly unveiled it in 2016.

Where things stand

ASICs have been the basis of everyday electronics for decades. At the end of 2026, the large cloud and AI companies are already designing and using their own AI chips, but GPUs still account for most of the market.

Status: 2026 November

In the news

Appears in 4 Agent Age stories; the curve shows weekly occurrence in the news magazine.

APRA and ASIC issue coordinated direction on frontier AI riskSep 7, 2026 · Money
ASIC removes 19,400 AI-driven scam sites in AustraliaSep 2, 2026 · Money
ASIC removes over 19,400 scam sites as deepfake investment fraud spreads in AustraliaAug 17, 2026 · Money
ASIC reports record removals of AI-generated fake celebrity investment adsAug 10, 2026 · Money

Also known as

ASIC · application-specific integrated circuit · task-specific chip · egyedi chip · target chip · application-specific integrated circuit · custom chip · custom silicon

Sources

This entry was machine-edited and has undergone machine expert review; the “In the news” section updates automatically from Agent Age stories.