[Update: We have added SMIC to this analysis.]
In 2009, we wrote a piece about Apple, looking not at the iPhone’s market share, which was tiny then, but at its share of industry profits, which were massive. Even in those early days of single digit iPhone share, Apple was already consuming 80%+ of handset industry profits. With all the foundry news lately, we though we would try something similar with semiconductor manufacturing.
Of course, there is no way to recreate this analysis for the handset industry today, as all the vendors left have stopped breaking out most of the relevant data. For foundries, the task is in many ways simpler, these are public companies and we are looking at their entire operation, so all the data is readily available. On the other hand, two big players do not break out foundry revenue – Intel and Samsung. For Intel, we hacked backed our ways into the numbers. We treated Intel’s Cost of Goods Sold (COGS) as Intel Foundry Services (IFS) revenue – this is far from perfect as there are a bunch of other items in those numbers, not least the fact that many Intel products are fabbed at TSMC. Even more problematic was ascribing operating expenses. There is no easy way to do this, part of the problem of splitting Intel into two companies is how tightly the design and manufacturing sides of the company are. So we guessed estimated – we ascribed 75% of Intel’s corporate R&D and 40% of other opex (MG&A) to IFS, and then for good measure we just knocked all these numbers down by 10%. There are more elegant ways to do this, but we believe this is close enough. Which is admittedly better than how we approached Samsung. Samsung’s foundry business is both buried inside a multi-division business and generates a significant share of its “revenues” from production for Samsung’s own chips. Samsung just does not provide enough information for us to even hazard a guess, so we left them out. Finally, there is the question of all the other companies with internal fabs (IDMs), notably the large analog companies. We will leave them for a future post, especially since this is the market that matters most to Global Foundries, who we did include on this list.
That leaves us with the following foundries:
- Global Foundries
- “Intel Foundry Services”
- Tower Jazz
- SMIC
- TSMC
- UMC
One final note on methodology. For Global Foundries, we only have data going back to 2018. Tower Jazz is in the process of being acquired by Intel and so has not released 2022 figures yet. Also, for TSMC and UMC we used a constant exchange rate NT$0.033 to US dollar. NT$ fluctuated a bit during this period, but since it is essentially a manipulated currency those variations are fairly small in the context of this analysis.
By market share, the data comes out as we would have expected, TSMC is the largest, but IFS is also fairly large, with everyone else much smaller.
The Profit chart is somewhat a repeat of market share, but a few items stand out. First, there are some negative numbers here – notably Global Foundries which turned its first operating profit just last year. Note also the steady shrinking of IFS. In the early days of this period they commanded a massive share of profits, which steadily dwindles from 62% down to 30%. Our best guess is that if our methods greatly overstates IFS’s profitability. If it were a standalone entity today, it would likely be losing money. There are some significant dis-synergies in splitting the pieces of Intel, and our math here masks much of that. Ultimately, IFS’s profitability comes down to some fine-grained cost accounting in any split, and we do not envy the people who end up tasked with determining those breakdowns.
The other thing that really stands out from this data is TSMC’s acceleration. We have data for Intel, TSMC and UMC that goes back over a decade, and while once all these companies were fairly close, TSMC has outgrown the others in both revenue and profits, a trend that hit overdrive during the pandemic.
In many ways, this data is unexciting in that it is unsurprising. It maps to our intuition, in a way that the Apple data did not. But it does highlight the fairly dramatic shift in market share, something we highlighted in our look at Intel’s prospects a few months back. It should also go without saying that Tower Jazz will have minimal impact on Intel’s market or profits share. Finally, we find ourselves wondering how Samsung’s figures shift this analysis. We suspect that the size of their memory business, if treated as a foundry customer, would position Samsung in the top 2, but we really have to wonder about the profitability of their foundry operations. Our best guess is that this business would show profitability much lower down the list.
Below we present our data for this series as well as a separate set with IFS removed