By now, most of us are accustomed to the truth that TSMC operates a major quantity of trailing edge manufacturing capability. Lengthy after they’ve moved on to ever extra superior nodes, they proceed to run the older fabs.
Editor’s Word:
Visitor writer Jonathan Goldberg is the founding father of D2D Advisory, a multi-functional consulting agency. Jonathan has developed progress methods and alliances for corporations within the cell, networking, gaming, and software program industries.
In 2024, TSMC generated nearly 50% of income from nodes which might be 5 years or older – 7nm and up. This stands in distinction to Intel, which famously shut down outdated nodes when transferring on to a brand new course of. When Intel made the choice to do that, it made sense for his or her enterprise mannequin, however now that they’re attempting to enter the foundry enterprise, that lacking capability is another impediment.
Nonetheless, we have been curious as to only how worthwhile TSMC’s older nodes are. The corporate breaks out income by node, however little or no past that. Armed with that and some guesswork discovered estimates, we did some math.
Beneath is a graph exhibiting their 2024 income breakdown by node…
Income by Node
The superior nodes – 3nm and 5nm mixed – contribute 52%. In contrast, the chart beneath exhibits working revenue by node.
Working Revenue by Node
And for comparability listed here are the numbers…
| Node | Income | Op Revenue |
|---|---|---|
| 3 nm | 18% | 13% |
| 5 nm | 34% | 14% |
| 7 nm | 17% | 27% |
| 18 nm | 8% | 12% |
| 20 nm | 0% | 0% |
| 28 nm | 7% | 10% |
| 40/45 nm | 4% | 6% |
| 65 nm | 4% | 6% |
| 90 nm | 1% | 2% |
| 110/130 nm | 2% | 3% |
| 150/180 nm | 4% | 6% |
| > 250 nm | 1% | 2% |
To place this in context, TSMC’s superior nodes generate 52% of income however solely 27% of revenue. That being stated, it is very important do not forget that the nodes – particularly 3nm – are nonetheless in very early levels and are advancing in profitability fairly quickly.
The figures for 2023 really present each 3nm and 5nm to be loss-making. They’ve made a whole lot of progress lately, and our estimate is that by subsequent 12 months, income share and revenue share will align extra intently. Simply in time for a brand new node to launch.
A be aware on methodology:
We acknowledge that not everybody enjoys the intricacies of constructing fashions as a lot as we do. So we saved this portion for the tip, for individuals who get it.
TSMC solely breaks out income by node, not profitability or bills, however that is sufficient to get began.
The massive driver on this mannequin is depreciation. The important thing to TSMC’s funds is that the outdated nodes are absolutely depreciated. The corporate depreciates tools over 5 years. So 7nm, which launched in 2017, is the cutoff. Every thing newer (3nm & 5nm) nonetheless carries depreciation load.
The corporate accounts for over 90% of its depreciation expense in Value of Items Bought (COGS). So, for our functions, we allotted depreciation to the 2 superior nodes by their share of income. Then we allotted the remaining COGS (minus depreciation) throughout all nodes by share of income. That is most likely a bit of bit off – older wafers most likely value much less and require fewer steps – however our sense is the distinction is minimal.
We then checked out R&D. We assumed that almost all of R&D goes to the newest nodes, and once more allotted the expense throughout 3nm and 5nm. That is really a bit off, as our assumption is that almost all of R&D goes to nodes not but in manufacturing.
Nonetheless, we imagine each the superior nodes nonetheless incur a good quantity of R&D prices related to working up the training curve and enhancing yields (particularly 3nm). So this overestimates prices to a level, however to a level we’re comfy with.
Lastly, SG&A we allotted evenly throughout all nodes by their respective share of income. This most likely overstates the prices of the older nodes, which presumably don’t want as a lot administration time or gross sales assets.
