Earlier this week, we published a piece on Qualcomm’s growth prospects (or lack thereof). We also listened to Nvidia’s GTC developer event. Through this we stumbled on an important difference between the two companies that highlights a growing issue with semis companies.

In our note on Qualcomm, we looked at the external challenges Qualcomm faces (spelled Apple), and noted that one potential solution for Qualcomm is to develop products that are so compelling they could make a shot at holding onto Apple. A big problem with that approach is an internal factor. Put simply, Qualcomm does not do sales well.

Qualcomm has a seriously siloed corporate structure. Each product line operates largely independently from others, not just a separate RF team and a separate modem team, but further sub-divisions within each of those segments. This structure helped Qualcomm stay nimble in its early years when it was fighting for survival, but has long since outgrown its efficacy. 

Adding to this, Qualcomm does not have a strong sales culture. Again, in the early days of the company traditional sales models did not really apply. Qualcomm won business by its mastery of the standards process and having the first product to market. Even as recently as 2013, Qualcomm had almost no competition for modems for the first few years of the 4G standard. They did not need a complex sales org to generate revenue.

This has been changing in recent years. They hired a Chief Commercial Officer in in 2022. Which is definitely encouraging, but it only happened in 2022!. It is not just a question having sales people, it is a question of having a sales culture. For Qualcomm, this muscle is just not well developed, and it will take time to develop. In practice, that means that Qualcomm customers often have to interface with multiple Qualcomm teams. This works fine for low level engineering decisions – the handset engineer can pick and choose the modem, apps processor, power amplifier, filters, tuners, power management and other parts individually. This is not what we would call a system sale. And while at a high level (say CEO to CEO) there may be some broader perspective, most of the day-to-day work is fragmented. This presents all sorts of challenges, not least of which bringing on new customers or selling a major customer on some form of system-level advantage.

This contrast very sharply with Nvidia’s approach. For us, the overriding them of GTC was the sheer number of “partners” included in the event. During his keynote, CEO Jensen Huang would periodically stop to rattle off a half dozen companies Nvidia was working with on a particular product. The company issued 37 (!) press releases for the show, 60% of those mentioned some customer or partner in the headline, including one press release where they just say “Nvidia and 100+ Partners help Move Enterprises to AI”. The keynote walked through about a dozen new products, Almost all of these were purpose built for a specific industry (healthcare, lithography, drug discovery, warehouses) or specific development approaches (LLMs, generative AI, graphics, etc.). And crucially, none of these new products were just a chip, they were all some module or system combining multiple parts, including those from competitors.

Nvidia has sold more than discrete chips for al long time, beginning with complete graphics cards for PCs. But as the company has transitioned into the dominant force in “AI” and increasingly in the data center, they have broadened that approach tremendously. This is not customizing a chip for a particular customer, it is a broader approach, providing complete solutions for specific industry needs. This looks a lot like the future of compute that we have been talking about so much lately.

Qualcomm tried this approach over a decade ago when they saw Mediatek as a major threat. Qualcomm launched a line of “reference designs” (aka Dragon Boards) that provided a complete blueprint for customers lacking sophisticated engineering skills. This did not work out well, in large part because of the fragmentation at the heart of the the company’s organizational structure, each of the constituent products on the design went their own direction.

Now, we are not saying Nvidia has perfected sales. Their organization has its own set of challenges (forecasting channel inventory, to name one). Nor are we saying that Qualcomm is doomed, they have made real progress moving forward on many of their structural issues in recent years. That being said, it is hard to compare the two models and not conclude that Nvidia’s approach is the future.