Business Models: EDA Is Software But It Used To Be Sold As Hardware

Business models are really important. Just ask any internet startup company that has lots of eyeballs and is trying to work out how to monetize them. It is a lot easier to get people to use something for free, much harder to get people to pay for something especially when they don’t value it much. Different companies that look somewhat similar often have very different business models.

I was once sat on a plane next to an executive from HBO. This was in the Sex in the City and The Sopranos era. I asked him how come HBO made programs that were so much higher quality than CBS, ABC, NBC and Fox. He said that although it looked like they were in the same business, making programs for TV, in fact their business models were totally different. The big networks sold eyeballs to advertisers, the more eyeballs, the higher the price. So they wanted to make lowest common denominator programs that appealed to as wide a range of people as possible. HBO got paid by subscriber. If you subscribed to HBO for a month, then they got $3 or whatever the number was. And here is what he pointed out that I’d never thought of. They didn’t care if you liked Sex in the City as long as you liked The Sopranos enough that you were not going to drop HBO. A mostly different demographic loved Sex in the City and never watched The Sopranos. But they were not going to drop HBO because they needed their Sex in the City fix every week. So it was important to make a few really high quality programs that together covered all interests, but it was irrelevant if you liked them all. It is almost as if HBO is wasting effort if you like too many of their programs, they get paid the same if you only like one, provided you like it enough to keep signed up. I am sure today there are many people who would never drop HBO because Game of Thrones will be back for season 6. Or who love John Oliver’s show.

Once business models are set, it is very hard to change them. HBO is facing this today, wondering how to best get money from millennials who typically don’t bother with cable. If they switch to pure internet, they risk losing all those $3s from people who don’t watch stuff on the internet (or better still, who pay for HBO as part of a bundle but don’t even watch it). If they don’t, how do they reach the non-cable audience. As more an more people cut the cord (I did) then their market is moving from people who watch on Comcast to people who watch on Roku or AppleTV or their phone.

When EDA started it was in the era when there were no separate hardware and software industries. The first wave of EDA was Calma, Applicon and Computervision. They would sell you the hardware with the software already installed. For example, the Calma Graphic Design System (or GDS) was a re-badged Data General minicomputer. You paid a single price to license the hardware and the software in the same way that if you buy a digital camera you don’t pay for the software separately. Hardware was sold as an up-front purchase price and then an annual maintenance of around 15-20% of the original price. That covered the hardware and the software.

As an aside, if that GDS name seems familiar, it is indeed the same GDS as in the layout interchange format GDS II. That is actually the tape backup format from those systems dating back to the late seventies.

See also Old Standards Never Die

In the early 1980s things started to change. It is funny to look back on now, but we were genuinely worried back then whether people would pay more for software than they did for the hardware it ran on. After all, they never had before. That fear turned out to be unfounded. Initially expensive software was sold with hardware on which to run it. The DMV sold proprietary hardware in the case of Daisy and Valid, or rebadged Apollo workstations in the case of Mentor. Sun workstations came on the scene, and other offerings from HP and IBM. Everyone already had a Vax. So over time the software got unbundled. But for years the business model continued to be what it had always been: pay and up front license fee and then a maintenance fee of 15-20% per year.

During the fast growth early days of EDA this was very convenient for the EDA companies because it front-loaded the revenue needed to grow and fund R&D, but still left a sort of royalty stream going in the future years. I believe it was Gerry Hsu who first started to switch users to 3-year deals, having noticed that people seemed to like to lease cars so they got a new one every three years. That business model transition was not entirely smooth, they never seem to be.

In EDA, as a rule, you make money with software that runs for a long time (like static timing or P&R) or that people sit in front of all day (like layout). This naturally creates a reasonable license demand. Other software suffers from what I call the “Intel only needs one copy” problem. If it runs fast then a single copy can be shared around a large organization. Since they are probably not going to pay a seven figure sum for the license, this creates a problem as to how to grow revenue. If one copy serves a lot of users, it is hard to turn the beach-head of a single copy into true proliferation. When this has happened in EDA various things have been tried: per tapeout fees, per named user fees, bundling it with something where lots of copies are needed. But these all have the problem that it is hard to change a business model. Users expect to pay a normal floating license. Anything else will at worst mess up the sale and at best delay it.

The whole industry is now in a mode where deal are done for a 2-3 year period and the revenue is recognized monthly over the period, as I wrote about last week. This gives a lot of predictability since over 90% of a quarters revenue is coming out of previous backlog.

See also Synopsys Did 90% of Business From Backlog with A Deal Length of 2.5 Years. Err…What Does That Mean?

The one part of the business that still has a hardware business model is emulation. But that is because it is hardware. Even if the EDA companies wanted to recognize the revenue over 3 years, GAAP will not let them. The hardware leaves the dock, the price drops from booking to revenue. Inevitably this makes emulation revenue lumpy and less predictable than software. Just like EDA used to be in the early days. In fact, given how much is coming out of backlog, when guidance for the following quarter is given in an earnings call it is largely about how much emulation business they expect to ship that quarter.

You may also have seen the news that Sesame Street is moving to HBO. They want every child to subscribe too! There is even a business model wrinkle here too, since Sesame Street is always used as an example of just the sort of program that only PBS could make since no private company ever would. Except now they will be, and since they are richer so they will be making twice as many new programs per year as PBS could afford (the new programs will be on PBS too, but delayed a few months). Netflix makes House of Cards. Amazon is making the new Top Gear. When business models are in transition interesting things can happen.