IP Cores and the Fabless Model — How Arm Rules Silicon Without Making a Single Chip

Why a company that only sells blueprints ends up at the center of the chip industry. License fees versus royalties, the shipment volume where buying beats building, the three grades of IP (soft, hard, architectural), and what RISC-V actually made free — and what it did not.

Not just fabless — they don't sell chips at all

"Fabless" describes a chip company that owns no fabrication plant. Apple, NVIDIA and Qualcomm have never etched a wafer with their name on it. They design, and hand manufacturing to a foundry such as TSMC.

Arm sits one step further back than that. Arm doesn't make chips, and it doesn't sell chips either. What it sells is a circuit blueprint and the right to build things that follow it. And yet nearly every smartphone application processor in the world runs Arm instructions, with the same architecture spreading into servers and microcontrollers. A company that ships no physical object sits at the middle of an intensely physical industry. This article is about the structure that makes that possible.

The full path from sand to finished part is mapped in The Semiconductor Supply Chain. Here we pull out only the very top of that chain — the layer where the design itself is the product — and follow the money.

An analogy: architects, contractors, and licensed drawings

Think about putting up a building.

• The architecture firm draws the plans
• The general contractor actually builds
• The owner commissions the work and uses the result

In semiconductors, the foundry is the contractor, and the fabless company is both owner and lead designer. So what is Arm? Arm is the party that owns proven structural patterns that architects reuse — stair details, shear wall layouts, fire egress arrangements — and rents out the right to use them. Borrowing a pattern with a track record and adapting it is faster and safer than inventing one. The borrower pays to use it, and pays again in proportion to how many buildings go up.

Those patterns are IP cores (intellectual property cores, or design assets). CPUs, GPUs, the physical layers of USB and PCIe, memory controllers, crypto engines — a modern SoC is mostly a composition of blocks licensed from other companies, and the portion drawn in-house is usually the minority.

The intuition: it all comes down to who carries the fixed cost

Why borrow? The answer isn't performance or philosophy. It's the shape of the cost.

Chip design costs are almost entirely fixed costs incurred before a single unit ships: engineer salaries, EDA tool licenses, compute for verification, and a mask set for the prototype. That money leaves whether you sell zero units or a hundred million, and selling a hundred million barely raises the total. In semiconductors, the cost of being able to make it dominates the cost of making it — and that share grows with every process node.

Worse, this fixed cost grows multiplicatively, not additively. Each generation adds design rules, adds combinations to verify, and shrinks the range where physical effects (interconnect delay, IR drop, heat) can be ignored. The fact that the growth is not linear is what drives nearly every strategic decision in this industry.

When fixed costs rise by orders of magnitude, only two options remain: raise volume by orders of magnitude and recover thinly, or share the fixed cost with other people. IP licensing is the second. If Arm builds a CPU core once and a thousand companies use it, each carries a thousandth of the development. From the borrower's side, a verified design far beyond what they could build alone becomes a variable cost tied to shipments. IP licensing is a machine for converting fixed cost into variable cost — that is its economic essence.

The mechanism: license fee, royalty, and the crossover volume

IP is usually priced in two tiers: an upfront license fee paid at signing, and a royalty paid on every chip shipped. Write the license fee as $L$, the royalty rate as $r$, the chip's selling price as $p$, and lifetime shipments as $N$. Then the total cost of borrowing is:

That line in words: what you pay is the lump sum up front plus a fixed fraction of the price on each unit sold — nothing more. $L$ is owed even at zero shipments; $r p N$ grows only as units move.

Building the same function yourself costs $C_{\text{own}}$, which is essentially all fixed cost $D$:

In words: in-house design means paying everything at the start and nothing per unit afterwards. The absence of $N$ on the right-hand side is the statement — build a hundred units or a hundred million, the total doesn't move. The shipment count where these two lines cross is the decision point:

That fraction in words: take the fixed cost you avoid by borrowing, and divide it by the extra amount you hand over on every unit sold. The numerator $D - L$ is the money licensing saves you up front; the denominator $r\,p$ is the royalty on one chip. The quotient is simply how many chips it takes for those royalties to eat the saving.