|
Seismic shift hits foundries |
|
|
|
Oct 31, 2007 at 03:06 PM |
|
The top four pure-play foundries (TSMC, UMC SMIC & Chartered) have now all provided their latest quarterly results with one very important underlying trend!
Capital spending taps have basically been turned off for 2008 and beyond, even though utilization rates have recovered to the point when capital spending would historically have kicked in to further increase available capacity.
TSMC guided that CapEx would be significantly reduced in 2008 and that the ratio of CapEx to sales would be set at a lower level going forward. UMC said the very same thing while highlighting that spending would focus on its existing tool installed base by migrating more of its current capacity to smaller geometries.
SMIC and Chartered have guided that a tightly managed CapEx-to-demand ratio would be maintained, indicating that spending in 2008 will be at best comparable with 2007.
We haven’t seen this level of spending cuts since the dotcom bust, yet all the foundries are seeing high utilization rates at leading-edge nodes as well as at trailing-edge!
The increasing move on the part of IDMs to adopt an asset-lite business model would have given the impression that the last thing we expected was for the foundries to cut CapEx. However, it is quite apparent that asset-lite moves take several years to pick up momentum and 2007 was definitely not a momentum year!
But, on the current face of it, 2008 doesn’t look that much better!
We also have new 300mm fabs under construction at TSMC and UMC while SMIC will manage two new 300mm fabs for provincial Chinese Government agencies. With respect to TSMC and UMC, it now looks like these fabs may see little or no activity in 2008!
I would welcome readers’ views as to why the major foundries are reacting this way, as it does not seem to make sense to me.
Here are my current speculations:
One aspect may be down to the very slow adoption of 65nm. The capital spending, especially this year by most foundries, has been allocated to 65nm technology and capacity tool buys, while the customers stayed away. Although things are improving, Chartered doesn’t expect an improved 65nm ramp environment until the second half of 2008!
Another is the underlining wafer price that has seen ASPs decline in 2007, while tool depreciation costs have risen and utilization rates were low (1H07). The ROI issue is more acute at Chartered and SMIC compared to the other two foundries, but TSMC has highlighted that it does not expect to grow above the industry average and indeed is expecting flat growth for 2007.
However, I am sure there is something missing from this reasoning and look forward to your comments.
|
Comment by GUEST on 2007-11-07 12:08:29
My opinion entails a question:
Would it be possible that the merchant fabs are seeing the opportunity for skipping further investments at the 300mm @ 65nm node and going straight for 300mm @45nm, to be followed soon after by 450mm @ 32nm?
I suspect that as Moore's Law ends, and the financial risk/reward explodes, good strategists may consider jumping to a small as possible CD, and then staying in that node for a long enough time in order to attain a decent ROI.
Assuming that this analysis holds any water then the merchant fabs might be reserving the CAPEX for additional 300mm @ 45nm in 2009-2010, while doing capital buildup for investment at 450mm @ 32nm starting 2011-12. The problem with investments in the 300mm @ 65 nm is that, at this point, they will be short lived.
Intel’s earlier than predicted push to 300mm @ 45 nm seems to be putting pressure on everyone and if continuing unabated, may end up rendering much of the existent 300mm @ 65nm; and even new infrastructure obsolete before it has enough time to attain target ROI.
The fact that all merchants have adopted a similar strategy suggests there is some asymmetry of information affecting financial projections of the whole merchant industry as well as those of the IDMs.
My theory is that merchants are becoming increasingly aware of their winning position on the logic IC industry and realigning strategies for the check-mate of the weakest IDMs in 2009-2012.
Under this analysis, consolidation will be rampant and accelerating from now on, and will propagate as well to the equipment suppliers and R&D centers.
Cost cutting through M&As seems unavoidable at this point. | Comment by julian1959 on 2007-11-02 10:33:27
I think part of the problem in the slow migration to 65nm is that the design cycle is getting increasingly long and exponenetially longer over 90nm and 130nm. Each node adds excessive time in verification and RET enhancement checks. If time to market is an issue then it might drive decsions on node design to be done at 90nm where the cycle is more predictable. While it appears there are designs at 65nm in the pipeline, I still think the overall cost is not giving IDMs and the OEMs the necessary ROI. There's no question that each technology node allows for additional functionality of the chip itself, but long gone is the shrink for more die per wafer, except in DRAMs (maybe) | Comment by GUEST on 2007-11-01 16:59:25
Now that I've thought about it for five minutes, while I still think a cartel is unlikely I want to modify what I said:
there is a huge incentive to cheat on the cartel but it also takes a couple of years to increase capacity and your competitors will know if you drop the $2B to build a new fab. A cartel could actually work out very well in this industry and now would be the perfect time to be doing it as companies go asset-lite.
That being said, I think that's highly unlikely that this has to do with short and medium term demand and market composition trends.
What does the market segment breakdown of foundry revenues look like? | Comment by GUEST on 2007-11-01 16:57:08
I'm not an industry insider, but what you've said makes sense. Another possible piece of the puzzle: what about across the board improvements in efficiency and throughput that allow for reduced costs and increased output with the existing tool sets? They could be expecting that this will be sufficient to supply the industry for the next couple of years. Didn't Intel reduce capex this year by $600 million for just this reason? Wait a couple of years for asset lite to start kicking in and then begin capacity buildouts. This obviously also has the advantage of avoiding even more idle fabs if the global economy goes to pot in the next couple of years.
As an off the wall observation: If the high utilization rates and reduced capex are unanimous across the foundries, I suppose it's possible that we could be looking at the beginnings of an industry cartel -- but that is highly unlikely. Call it $93/barrel envy.
It would seem to me that the economic incentives to cheat on a cartel agreement in this case would be far too great.
| Comment by GUEST on 2007-11-01 16:55:04
Hi, I am Reiko from SMIC Corporate Relations Center. I have read the article and would like to clarify that SMIC only manage and operate ONE 300mm fab for WXIC and ONE 200mm fab for Cension. |
|
|
|
Blogs 
