Can Qualcomm compete in Custom Silicon?

On this week’s episode of the Circuit Podcast we dive deep into the custom ASIC business, where chip companies help non-chip companies bring their silicon dreams to life. And last week, we spoke with a Chinese Tech site about this topic, and the interviewer asked us why Qualcomm is not more active in this business. The interviewer had a very well-reasoned argument in favor of this plan, and we think the answer tells us a lot both about Qualcomm and the broader business of custom silicon.

On its surface this plan makes a lot of sense. Qualcomm’s customers – the big handset companies – are all developing their own slate of custom chips ranging from small-ish projects like image sensors and AI accelerators to full-blown applications processor SoCs which include all of the above and many other features. These companies are spending a lot of money on the effort but are each encountering a range of difficulties. Xiaomi’s silicon expenses are huge, Oppo actually gave up at one point, only to be coming back in now several months later. Having a seasoned chip designer help them would be hugely helpful to their efforts. Last week, Mediatek announced a big expansion to its custom ASIC work, but so far Qualcomm has been largely silent on the subject.

The easy explanation is that Qualcomm does not want to help grow its competition. Those customer designs would be used to displace some of Qualcomm’s core products such as its Snapdragon line. This is a reasonable concern. Mediatek pointed out that it was only working on chip designs where it would not sell a competing product, such as for the data center. On the other hand, this instinct is not stopping Qualcomm’s peers like Marvell and AMD, who both reasoned that it was better to join than fight. Their reasoning seems to be that if they help those customers design chips, it will drive attach rates for their other products. In fairness, those companies have broader product lines than Qualcomm, but AMD still appears to be willing to help customers design CPUs and AI accelerators, businesses which are central to their operations.

Another important factor is that Qualcomm lacks a sales culture and is in desperate need of retooling its efforts to approach customers more strategically. Qualcomm does not really have an organization capable of providing this service, both in terms of making sales calls and in supporting designs going into production. They can handle their own production needs, but that operations team does not have any external support capacity.

All of those could be solved, but we think there is one more critical reason sitting beneath all of them that is likely the ultimate blocking factor – intellectual property (IP). Qualcomm is a company founded on its R&D and IP, it generates ~30% of profits from licensing that IP, and all those patents play a foundational role for the company’s mobile products. If they start helping other companies design chips, they could theoretically put that IP at risk. Today, even basic sales agreements (i.e. no custom design involved) require dense contracts to protect both vendor’s and buyer’s IP. When we start talking about jointly designing a chip, the contracts get even longer. Imagine Qualcomm helps Vivo design an AI accelerator to help image processing, just to name an example. Vivo would not want its design work to end up in Xiaomi’s phones, and would seek to limit the sharing of IP within Qualcomm to prevent that. Even if Qualcomm could build a completely siloed design team dedicated to Xiaomi (by no means easy), they would not want to sign a contract that prevents them from selling image sensing chips to other customers. Given that the contract has to be signed before design work really begins, there is no telling what would go into that joint work until is done. By which point it may look identical to what the Qualcomm internal product looks like. Vivo could claim a violation, and Xiaomi wold not want to buy a product which would open it up to a lawsuit from a competitor. And if Qualcomm patented any of the work it did for the internal chip, that would risk contaminating the entire patent pool. Can of worms – opened. There may a way to overcome these hurdles, put enough lawyers in a room to figure out, but then again maybe there is no solution. Companies like AMD face similar IP concerns, but they do not have massive businesses built on licensing that IP, and so their problem is solvable.

In theory, Qualcomm could take the Mediatek approach of only working on products which they do not sell, but what would that leave for the company to really sell. They would be competing with companies like Mediatek which have been doing this for much longer, and then factor in all the other obstacles above. Qualcomm would be left as a third or fourth pick. Ultimately, we think Qualcomm will take some version of this path, but only for a small number of customers. Maybe if they had a stronger focus on strategic partners this could become important to the company, but they have a long way to go before they can get to the point where being a custom ASIC vendor would merit much of their attention.

Leave a Reply