Most Products Run on the Wrong Chip

Every successful hardware product eventually hits the same wall: the chip that got it to market becomes the chip holding it back.

Most products are not built on the right chip. They’re built on the chip that was fastest to ship with.

At the beginning, that’s not a mistake. It’s the rational decision.

If you’re building a new hardware product, you don’t start by designing custom silicon. You start with what exists: a general-purpose processor, an FPGA, or an off-the-shelf dev kit. That gets the prototype working. It gets customers in the door. It gives the team speed when speed matters most.

But as the product scales, something subtle happens.

The Hidden Tax of Off-the-Shelf Silicon

The same shortcut that helped the product get to market starts becoming its biggest bottleneck.

Suddenly the product is burning too much power, carrying too much compute, pushing too much data to the cloud, and costs too much to deploy at scale. The hardware still works, but around the wrong components:

• Power budgets blown
• BOM inflated
• Board space wasted
• Thermal limits hit early
• Too much data sent off-device
• Too little hardware differentiation

The company ends up paying, over and over, for hardware that was chosen for convenience rather than fit.

This pattern shows up everywhere once you know what to look for.

Apple didn’t build its own silicon because Intel chips couldn’t run software. It did it because generic chips were no longer the right fit for the exact product Apple wanted to build.

• Waymo didn't build custom silicon because off-the-shelf GPUs couldn't process sensor data, they did it because autonomous driving requires real-time inference at a power and thermal efficiency generic chips can't deliver.
• GoPro transitioned from merchant image processors to their custom GP-series to unlock hyper-smooth stabilization without melting the camera in ten minutes.
• DJI dominates the drone market because their custom silicon handles complex computer vision and low-latency video at a weight and power envelope standard chips can't touch.

The Frontier-Node Distraction

Tech is obsessed with frontier-node AI chips.

But for a huge number of physical products, chasing the frontier is the wrong problem. They do not need the densest transistors on Earth.

They need to stop paying for generic overhead.

For many products, the win comes from removing general-purpose bloat, collapsing repeated functions, and matching the silicon much more tightly to the actual job. They need better performance per watt, lower always-on power, fewer packages, and a faster path from product idea to real hardware.

What Companies Actually Want

The biggest misunderstanding in chip design is that the bottleneck is just tooling.

Better Electronic Design Automation (EDA), AI-generated RTL, and open-source flows all matter. But the deeper problem is the gap between knowing what a product needs and actually getting to manufacturable, economically sensible silicon.

The real barrier to custom chips is that the path from idea to production-ready silicon has historically required a company to take on enormous complexity, risk, and organizational overhead. Most companies do not want to become chip companies just to make one part of their product better.

Most companies are not shopping for “custom silicon” in the abstract. They are shopping for outcomes: lower power, lower cost, simpler boards, and faster products. The chip is just the mechanism.

If you look at the world through the lens of who wants to design a chip? , the answer is: not many people.

If you look at the world through the lens of who is paying a hidden tax for generic hardware that no longer fits their workload?, the answer is: a much larger market than the industry has historically served.

The Next Semiconductor Shift

The next big shift in semiconductors won’t be just better tools for chip designers.

It will be making the right silicon reachable for companies that were never supposed to have it.

If your engineering team is bleeding cycles optimizing around hardware constraints, you are already paying the price of custom silicon. You just don't have the chip to show for it.

Jordon Kashanchi - CTO @ Visibl Semiconductors (YC W26)