ARM presumably makes more money on the big ones, especially if they also license the GPU; but the 'Cortex-M' low power/small size microcontrollers ship in heroic quantities compared to the 'Cortex-A' application processors that actually get mentioned on the spec sheets of various cellphones and tablets. There are also the 'Cortex-R' designs aimed at tasks with hard real time constraints; those are rarely mentioned but ubiquitious in cell modems and the like.
Then you have the 3rd party designs that are ISA compatble rather than directly licenses of ARM designs, Apple and Qualcomm certainly ship a fair few of those.
15 billion may be a bit high; but 15 million probably doesn't even cover a year of new cellphones.
That said, it's exactly this ubiquity and versatility that makes me wonder what SoftBank is thinking in actually buying ARM. ARM does licensing, so if you want a CPU for your application with more customization than just buying a reel of somebody's ready-to-go silicon, they'll sell you a license on pretty favorable terms. Definitely cheaper than paying ~45% over market price to buy out the whole company. If you want the right to do your own thing with their ISA, or want them to design something to fit your particular niche, that'll cost more than a cookie-cutter license; but still substantially less than buying them out.
Plus, since their business is licensing, buying them out is more or less assured to make all their existing customers nervous: what is the new management going to do? Are they going to plunder ARM's design talent for their own pet projects? Start monkeying with license fees, release schedules, etc. to gain competitive advantage for their own products in other markets?
I freely admit that I'm no genius of the silicon supply chain; but my impression was always that ARM's success was a kind of 'for profit development consortium': They aren't running a charity; but they offer solid engineering and reasonable prices to basically everyone who comes knocking, which has made ARM a de-facto standard for a wide range of situations where a company needs a CPU to embed in their product and doesn't want to DIY a proprietary freakshow; or invest the resources necessary to deliver a competitive, updated, SPARC core or the like. Since ARM sells to everyone, they amortize engineering costs across a zillion units; and their licensees can mostly rest easy knowing that ARM, who doesn't have any real direct involvement in selling SoCs or phones, or products in other markets, isn't going to start turning the screws on their licencees in an attempt to boost their own product lines.
This laid-back attitude probably contributed to having a stock price low enough to be an acquisition target; but it also helps them make all the sales they do. If people wanted to deal with an arrogant, dangerous; but very skillful single-source, they could buy Intel silicon. If they wanted to go it alone, SPARC is free for the implementing and MIPS is practically begging people to use their ISA. So far, ARM's combination of greater engineering support than the do-it-yourself options, and greater friendliness and better prices than the Intel options have proven very popular. Now that SoftBank needs to recoup a substantial investment and do whatever they had in mind when the purchased ARM, though, is that state of affairs going to persist?