The Hidden Cost of Chip Design Licensing

Fexingo (0:01)

Look at the latest numbers from the Semiconductor Index.

The SOX is up nearly 2% over the last 5 days, driven largely by Nvidia and Micron posting strong momentum. But if you look closer at the balance sheets, there is a subtle shift happening in how these companies value their own designs.

Luna (0:21)

You mean the licensing deals? Because it feels like every major player is suddenly trying to build their own core instead of just buying them off the shelf.

Fexingo (0:30)

Exactly.

Let's talk about Arm, the British chip designer that essentially holds the keys to the mobile world. For years, they've operated as a toll collector, charging fees based on the number of chips shipped or the performance tier of the processor. It is an incredibly lucrative model because it scales with volume without requiring them to build factories or manage complex manufacturing logistics.

Luna (0:55)

But recently, I have seen reports that some of their biggest clients are getting nervous about the long-term cost of those royalties, especially as chip complexity explodes.

Fexingo (1:06)

Right. And that brings us to the specific case of Apple. When Apple decided to move away from Intel processors for its Mac lineup and switch to its own custom silicon based on ARM architecture, it wasn't just a hardware change. It was a strategic renegotiation of power.

By designing its own cores, even if they still license the instruction set, Apple captures more margin and reduces its exposure to per unit fees.

Luna (1:33)

Does ARM actually lose money when a company like Apple designs its own cores, or do they still get paid for the base architecture?

Fexingo (1:41)

They still get paid, but the dynamic changes.

Apple pays for the right to use the ARM instruction set and perhaps some foundational IP blocks, but they keep all the profits from the advanced customization. This has triggered a ripple effect across the industry. Competitors like Qualcomm and MediaTek are watching closely, realizing that sticking strictly to standard ARM designs might leave them vulnerable on margins.

Luna (2:07)

So it is less about who makes the best chip now, and more about who owns the deepest pockets to fund custom R&D.

Fexingo (2:14)

Precisely, and this isn't limited to mobile.

We are seeing this play out in the data center, too. Nvidia, which trades as NVDA, has been heavily invested in ARM-based servers for AI workloads. But even they are facing pressure to optimize their internal designs to keep costs down as training models grow exponentially larger.

Luna (2:36)

It seems like the era of easy access to cutting edge chip technology is ending, replaced by a high barrier club where only the giants can afford to compete.

Fexingo (2:47)

That is a fair characterization, but there is another angle here regarding the actual pricing structure of these licenses.

ARM has been experimenting with different fee models, moving away from simple per chip counts toward performance-based tiers. This means if your chip runs faster or handles more threads, you pay significantly more regardless of how many units you sell.

Luna (3:10)

That sounds risky for any startup trying to enter the market with a niche, high-performance product. They could get crushed by fees before they even scale.

Fexingo (3:19)

It is exactly that risk, which is why we are seeing a resurgence of interest in Risc-V, an open-source instruction set architecture. It offers the flexibility of ARM without the licensing fees.

However, the ecosystem around Risc-V is still fragmented, and major players are hesitant to bet their entire future on it just yet.

Luna (3:42)

So ARM remains the king of the hill, but the throne is getting shaky as clients try to build their own castles.

Fexingo (3:49)

I would say the throne is secure for now, but the moat is being dug by the clients themselves. Let me pause here for a quick moment to talk about how this show operates.

Luna (4:00)

We really appreciate the listeners who help keep these deep dives going, especially on topics like this that require serious research.

Fexingo (4:08)

A small group of supporters already covers our production costs through buymeacoffee.com/fexingo, allowing us to stay independent and ad-free. If you find value in these analyses, consider joining them.

Luna (4:22)

It makes a huge difference in being able to dig into the specifics without chasing clicks. Thanks for supporting the network.

Fexingo (4:30)

Back to the chip wars.

One area where this licensing tension is becoming acute is in the automotive sector. As cars become computers on wheels, the demand for specialized processors is skyrocketing. Companies like Tesla are designing their own AI chips, bypassing traditional suppliers entirely.

Luna (4:51)

Tesla has always been vertical, but are other automakers following suit? Or is that too expensive for most of the industry?