Warning: long post.
I'd like to proffer my thoughts on the possible early warnings signs. In general they mirror much of what has been said already, but perhaps voicing them may give a slightly different perspective and hence, focus.
First I think it is reasonable to assume that downturns in this industry are (at least) bimodal.
For lack of good descriptor, I'll call the first type of downturn to be external. By this I mean anything that may occur outside the decision scope of the industry players. Examples would be the obvious like Asian financial crisis, natural disasters, wars, govt. machinations. I would also assign to this group other unexpected shifts like discontinuous technologies (say molecular or bio computers)that make it across the chasm. For most if not all of the semiconductor history, the discontinuous innovations have positively impacted it rather than negatively, but there is always the possibility.
For this group I think the only possible empirical metric available would be deviations of World Wide Semiconductor sales growth rate from the historical and projected rates. I include the projected rates purposefully, since I think investors (private and professional) will build in some level of valuation to the stocks believing the industry growth will approximate those projections.
I've tried to rationalize how to evaluate external source's influence upon unit trend effect, but am unable to do so. Although I intuitively think there is useful information to be extracted from the unit sales, I think it is lost in the noise. I see this noise stemming from functional consolidation (more functions on one chip)and application expansion. My guess is that we might be better off tracking power discreet devices on a unit base rather than all SC's. The logic being that every device has some kind of power supply. I have not yet examined this data seriously.
The second mode I'll call internal. This is essentially a supply/demand mismatch on fab capacity to MSI silicon unit demand.
I know that DQ thinks that the key indicator is DRAM pricing. I do not agree. It is very hard for me to reconcile that the DRAM price and demand for fab equipment has a first order effect on Intel, TSMC, and (the new nonDRAM)TI being capacity constrained. I think these folks have misjudged where the market size would be today at the time when they needed to make the cap ex commitments.
So we have the classic "behind the power curve" situation in which these companies decide to "increase the cap ex budget" to catch up. Often they increase it multiple times in a single year as I believe Intel has now done.
Since this capacity comes on line in relatively big chunks, the supply:demand curve is more of a step function. For 200mm fabs at 20K WSPM that would be new yearly capacity of 12 MSI. 300mm fabs at the same starts (although the ones I'm hearing about are more like 5-8K) would give 27 MSI jumps. Couple that with projections that about 30 new fabs are needed each year and the yearly new capacity can be quite large. [I'm a little fuzzy on the 30 fab number as it has been decreasing over time and some might be replacement instead of new capacity.]
So in my mind the best metric would be to compare new capacity coming on line to MSI of silicon used. Unfortunately I've never found a direct way to get this information in a timely manner. Usually it is after the fact and distributed amongst presentations by market research companies, wafer manufacturers and companies like M&W.
Unless somebody has access to these data on a timely basis, the next best thing to measure is likely the capex to sales ratio by fab companies. This data seems to have a large std. dev. For instance from 1982 to 1998 there is MITI data that shows for 12 Japanese companies it has ranged from 13% (recent) to almost 50%(1984) of sales with an average of 23%. If I remember correctly TSMC has a capex plan that is nearly equal to sales!
I've also seen slightly different data showing semi equip sales to chip sales ratioed. This ranged from 11% to almost 25% over the last 9 years.[I'm not sure if it was world wide data or NA since it was attributed to both Semi and SIA.] This data is probably the most easily obtained since it may be hard to back out the fab related capex for some companies. Gottfried has done a good job charting this info already. It could be that we can track this and either fab announcement or the results of a competitor to M&W (is there a public one? I think M&W is still owned by Jenoptiks which never gave much of a breakdown in there results.)
If we track fab announcements, we can make reasonable capacity guesses, but probably won't see old capacity taken offline, which might skew the results.
Having said the above, I'd like to say that I think sales are solid for the next 4 qtrs (discounting external events), but I'm not sure about the ability to grow those sales. Primary reason is the lack of people. There are lots of open positions up and down the food chain. This is not a problem that can be cured in one year. Worse for those eqpt. firms in the Valley, it is becoming increasingly more difficult to attract new engineers from external areas due to housing costs.
At some companies there seems to be production capacity problems that are facility related as well. These can be fixed, but the people problem remains.
For many firms, they have exceeded historical valuations and I think that if the growth rates weaken these valuations are at risk. Consequently I've reduce some of my exposure already (better early than late is my experience in this arena) and will likely continue to do so over time.
As a conclusion I'd like to mention two factors that I see which may ameliorate the business cycles in this segment. First I think the foundries allow better global utilization balancing by taking the smaller companies out of the capacity swings.
Secondly, AMAT top managers have been preaching the single wafer fab model for at least 18 months now. I really believe their model is correct. It should reduce WIP and allow cap expansion in smaller increments.
Unfortunately, I don't think either factor is(will be) strong enough to soften the next cycle much.
FATBOY |
