In context: For years, hardware makers have observed the attention and valuation multiples enjoyed by software companies with envy. Employees at hardware companies have also longed for the fancy perks their peers receive at software companies, while their hardware teams are fortunate to even have coffee at work. Software may be eating the world, but does that mean only software companies get foosball tables at work?
In recent years, many of these “have-not” companies have increasingly discussed their goals of attaching software revenue streams to their hardware. This ranges from the reviled, like BMW’s heated seats-as-a-service, to Nvidia’s cloud ambitions with its Omniverse platform.
We were reminded of this recently by a TechCrunch article about carmaker Hyundai’s plans for software profits. Full credit to the Hyundai team, as they have some great ideas for generating software revenue from their customers. Their vision is to sell ways to customize car interiors with unique sounds and sports team-themed displays.
This approach is reminiscent of how game makers monetize free games by selling character skins and other decorative items. Hyundai hopes to achieve $15 per month per customer in subscription revenue. With nearly 4 million cars sold last year, this could amount to $700 million in extra high-margin revenue. While it’s an ambitious goal, compared to the $7 billion in profits the company earned, it’s not that dramatic.
Hardware companies essentially have three ways to sell software:
- Sell subscriptions for nonessential but appealing features, like custom skins for in-car displays or heated seats.
- Build a software adjacency to chip functionality.
- Sell standalone software.
In the context of semiconductor companies, all of these options are likely to prove challenging.
To date, we can’t think of a single semiconductor company that has created a software business from scratch. They all do extensive software work, but that’s not the same as selling software.
A few chip companies have acquired software companies, but these acquisitions don’t necessarily provide shining examples for others. Broadcom has acquired several software companies, but they function more like a private equity fund with a portfolio of semiconductor companies.
Intel has also acquired multiple software companies over the years, such as McAfee and Wind River, but they mostly ended up divesting them for less than they paid.
The other two scenarios are equally challenging. Adding extra circuitry to a chip for infrequently used features is already a common practice among chip companies. Loosely speaking, this is usually referred to as binning. For instance, a CPU with 50 cores can be sold at a lower price with 4 cores turned off, or vice versa, a chip with 46 working cores can be upgraded to 50 cores with a firmware update. Either way, it’s still one chip. The problem is that adding features requires space on the chip, which comes at an additional cost, so it’s not always easy to incorporate. Regardless, this is not really a software service, except insofar as it is marketed as one.
Alternatively, chip companies can offer an add-on service that is somehow tied to circuits on a chip. This is essentially what Nvidia is trying to achieve with Omniverse, a cloud GPU service that has all of Nvidia’s features activated 100%. However, Nvidia is an outlier, as they have a dominant position in AI training right now, which grants them sufficient market power to pull this off.
That being said, this model is not entirely far-fetched. We believe it’s something startup chip companies should explore. As we’ve noted previously, the biggest hurdle to this model is selling it to customers. Most chip companies lack software sales expertise. Startups can build that capability from day one, which is much easier than integrating a salesforce into a company with thousands of employees. Even Nvidia, with all its market control, relies on Microsoft and Google to really sell this service to end customers.