Machine Code

The processor is the computing part of a computer that carries out all instructions. It understands no words and no sentences, only numbers. Machine code is exactly this sequence of numbers: a very long chain of zeros and ones, in which each section means a single instruction. Such an instruction is tiny, for example: add two numbers, or fetch a value from memory. For humans this is practically unreadable, but for the processor it is the only comprehensible form. Every program on your phone or laptop ultimately exists as machine code.

The lowest level that every piece of software reaches

Humans write programs in languages like Python, Java, or C. These languages use English words and characters that can be read and discussed. The processor can do nothing with them. That’s why, at the end of every programming task, there is a translation into machine code. Without this step, even the most elegant program text remains just text in a file.

Machine code also explains why software doesn’t run everywhere. Different processor types understand different numerical instructions. A program for the chips in older Windows computers therefore doesn’t simply run on the Apple chips in newer MacBooks. The two families are called x86 and ARM, and their machine code is simply not the same. This is exactly why many programs have multiple download versions.

This level is also crucial for a computer’s speed. How fast a program runs depends on how cleverly the instruction sequence is structured. Two programs can accomplish the same thing and still take different amounts of time to compute.

From source code to instruction sequence

The translation is handled by a program called a compiler. It reads the text written by a human, checks it for errors, and generates the corresponding sequence of numbers from it. A single instruction in the source code often becomes dozens of machine instructions. The compiler also arranges them so that the processor has to wait as little as possible.

A single machine instruction typically consists of two parts. The first part says what is to be done, such as adding or comparing. The second part says with which data. So that humans can talk about this, there is assembly language: a notation in which each machine instruction is given a short mnemonic, for example ADD for addition. Assembly language is thus almost the same as machine code, just in readable notation.

Not every language is fully translated in advance. Python programs are processed piece by piece at runtime, which is more flexible but slower. A common misconception is that modern languages make machine code superfluous. That’s not true: it merely arises later and more invisibly.

Why the topic keeps coming up with AI chips

In everyday life you barely notice machine code, but you constantly run into its consequences. When an app in the store is offered separately for iPhone and Android, this difference is, among other things, behind it. Virus scanners, too, partly operate at this level, because malware often exists only as a finished instruction sequence.

In business news you mostly encounter this topic in connection with AI hardware. Nvidia's graphics chips are so successful because there is mature software for them that translates AI computations into efficient machine code. Competitors don’t just have inferior chips, but above all weaker translation tools. Analysts call this a moat, meaning a hard-to-attack advantage.

The reverse direction is also interesting: AI models now write program text themselves. In doing so, they almost always generate readable source code, not machine code directly. The translation remains the compiler’s job, because this step must be reliable and verifiable.