What is a CNC Machine
Everything you need to know to walk into an IC meeting and explain what CNC machining is, why the programming bottleneck exists, and why it has resisted automation for 40 years.
A computer numerically controlled (CNC) milling machine
A CNC machine takes a solid block of metal and carves it into a precision part. It does this by spinning a cutting tool at thousands of RPM and moving it through the material along programmed paths, removing metal layer by layer until only the finished shape remains.
CNC stands for Computer Numerical Control. Every movement of the cutting tool is specified by coordinates, speeds, and cutting forces in a program. The machine follows that program to tolerances of thousandths of a millimetre. A human could not hold these tolerances by hand. The machine can, repeatedly, for hours.
A single machine costs between $50,000 and several million dollars. A typical shop has 15 to 20 of them. The US has roughly 300,000 CNC machines across an estimated 35,500 machine shops. Globally, the installed base exceeds 3 million machines in over 338,000 facilities.
The infrastructure layer of manufacturing
If cloud computing has servers, manufacturing has CNC machines. They are the compute layer. Every physical industry runs on them. Nothing precise and metal gets made without one.
The list does not end. Agricultural equipment, marine propulsion, rail, robotics, scientific instruments, firearms, jewellery, motorsport, prosthetics. If it is metal and it needs to be precise, it was CNC-machined. The machines that make other machines are CNC-machined. The moulds that shape plastic parts are CNC-machined. The tooling that stamps sheet metal is CNC-machined. It is turtles all the way down.
How it works
An engineer designs a part in 3D CAD software. That model defines what the part looks like. Then a machinist has to figure out how to make it: which cutting tools to use, in what order, at what speeds, approaching from which angles, how deep to cut, how to hold the part so it does not move. That set of instructions is the CNC program. The machine executes it.
The physical cutting is remarkable. A carbide end mill spins at 10,000 to 20,000 RPM, shearing metal into chips that fly off at speed. Coolant floods the cut zone to manage heat. The machine moves on multiple axes simultaneously, tracing curves through solid material with sub-micron repeatability. A five-axis machine can approach the part from virtually any angle, cutting features that would be impossible to reach from a fixed orientation.
The machine itself is not the bottleneck. Modern CNC machines are extraordinarily capable. The constraint is the program that tells them what to do.
Why programming is hard
A 3D printer builds parts by depositing material layer by layer. The slicing software that generates the instructions is straightforward: take the 3D model, slice it into horizontal layers, fill each layer with material. The geometry of each layer is determined by the shape of the model at that height. One path through the problem.
CNC machining is the opposite. You start with a solid block and remove material. The cutting tool can approach from any direction. It can cut at different depths, different speeds, with different tools. Every feature on the part can be machined in dozens of different ways, and the order in which you machine features affects everything: tool access, rigidity, heat distribution, surface finish, whether the part vibrates or deflects under cutting forces.
3D Printing
Slice model into layers. Fill each layer. One viable path.
The software decides: layer height, infill pattern. That is roughly it.
A beginner can set up a print in minutes.
CNC Machining
Start with solid block. Remove material from any direction. Millions of viable paths, most of them wrong.
The programmer decides: tool selection, cutting strategy, approach angles, speeds, feeds, depth of cut, stepover, lead-in geometry, coolant strategy, fixture design, operation sequence. Hundreds of interdependent parameters per part.
A machinist needs roughly a decade of experience to program complex parts reliably.
The search space is combinatorial. For a moderately complex part, the number of valid machining strategies runs into the millions. Most of them will produce a correct part. A small number will produce it efficiently. An even smaller number will produce it without breaking a tool, damaging a surface, or crashing the machine. Finding the right strategy is a judgment call that depends on the specific material, machine, tooling, and tolerance requirements.
This is why the problem resisted automation for 40 years. It is not a slicing problem. It is a planning problem with hard physical constraints, where the cost of a wrong answer is a ruined part, a broken tool, or a crashed machine.
Training timeline
A CNC machinist typically needs 4 to 5 years to become competent and roughly 10 years to program complex multi-axis parts independently. The knowledge is accumulated through doing: learning how specific materials behave, how tools deflect, how machines differ from each other. Most of it is never written down. When a senior machinist retires, that knowledge leaves with them. Today, 25% of the US machinist workforce is over 55.
Walk through a machine shop and count the machines that are not cutting. In a typical shop, more than half of them are idle at any given moment. Spindle utilization across the industry averages 25% to 40%. A machine that cost $400,000 is cutting metal fewer than four hours out of every ten.
The machines are not broken. They are waiting. Waiting for a program. Waiting for a setup. Waiting for an operator to finish programming the last job. For every hour a machine spends cutting, the programmer spent four hours writing the instructions. In complex aerospace work, that ratio reaches 8:1.
The owner walks through the shop and sees capital burning. Half a million dollars of iron, sitting cold, because the one person who can program it is working on something else. He cannot hire another programmer. There are 354,100 machinists in the US. 25% are over 55. Projected retirements in the coming decade: 89,000. The pipeline is not replacing them.
The scale of the problem
The US alone spends $24 billion per year on machinist cognitive labour: the programming, tooling decisions, and setup planning that turns a 3D model into a finished part. Globally, wage-adjusted, that figure is approximately $100 billion. Including the full cognitive stack across all factory roles (inspection, production planning, quoting, DFM), the US figure is $61 billion and the global figure approaches $300 billion.
Every dollar of that spend is existing payroll. It is not a market that needs to be created. It is wages being paid to human brains doing work that software can do faster. The opportunity is the gap between rising demand for machined parts (defence, reshoring, electrification) and a shrinking workforce that cannot keep up.
What to read next
Anatomy of a Machine Shop follows one order through an 18-machine shop, from RFQ to shipment. The $61B Cognitive TAM sources every number in the market model. The Machinist Shortage models the retirement cliff through 2035.