We write a lot about non-chip companies rolling their own semis designs. This is an important topic for the semis industry today, lots of big customers are moving to their internally designed solutions, sapping demand and reshaping the market. Our core thesis is that these projects only work when the chips being designed convey some form of strategic advantage, merely saving money on chips is not sufficient. But the price of chips is still an important factor in this process. Most semis companies are currently enjoying historical record-best margins, the result of two decades of steady industry consolidation. Twenty years ago, every chip segment had a dozen or two competitors vying for each design win, today most segments are down to three or fewer suppliers.

Many look at this and assume that the high price of chips is the reason companies have shifted to their own designs. But as we linked to above, the math for this does not work out. Instead, the dearth of competitors makes itself felt along a different, crucial, axis – control of the roadmap.

Building a semi, especially processors, ASICs and SoCs, take a long time to design – months if not years. As a result, a key part of every chip company’s process is locking down their roadmaps. We have been involved in many of these, and they are serious events – multiple days of meetings with dozens of stakeholders arguing it out. Big customers typically do not attend these meetings in person, but in well-run organizations, they are there in spirit, strongly represented. Semi companies need to make sure they are designing in the features customers want. The lack of these features can doom the product, but too many features can blow up the cost of the chip. So there are always trade-offs, played out across design, pricing, SKUs – product definition. The result is some form of lowest common denominator – optimizing between all the customers’ and the designers’ interests.

For the biggest customers – and in the data center there are less than ten – the result of their vendors’ roadmap trade-offs are almost guaranteed to be unsatisfying. In the early 2000’s when there were ample competitors, a big company could co-opt one vendor’s roadmap and get something close to exactly the part they want. Today, with far fewer vendors, even the biggest customer has to settle for something less. (The exception being Apple who has taken this to an art form.)

And so it is no surprise that they are instead turning to designing their own chips. This lets them control the roadmap completely – getting exactly the features they want. It is hard to overstate how important this is. Chips run software, the closer that software can get to the silicon the better it will run, and the more performance it will deliver. The few companies that can pull this off can use these chips to massive advantage.