Tuesday, October 17, 2006

Another look at Intel's balance sheet

We do this exercise on Intel's balance sheet again.
Curret assets -current liabilities - inventory = $17.6 - $9.05 - $4.48 = $4.07

$4.07 billion is all liquid assets Intel has in the pocket.

What else? Intel has some long term investments, but it's shrinking. It was $4.1 billion in 2Q06, now it's $3.4 billion. Long term debt stayed at $2.6 billion.

The total amount of money(net) Intel has is thus $4.07 + $3.4 - $2.6 = $4.87 billion. This includes everything, cash, short term, long term and strategic investments.

How soon can Intel BK? It can happen really fast -- two quarters after AMD FAB36 fully ramps up, Intel will BK.

Some readers asked why inventory is excluded in computing Intel's liquidity. The point is, assuming Intel continues its operation, it will always carry 4 billion inventory at any given moment of time. At the time of its BK, these inventory worth 0, because they are unwanted. You can't use that inventory, even if it's finished CPUs, to pay debt. So you have to deduct inventory from the current assets in BK analysis.

59 Comments:

Blogger Mikolaj said...

I agree with you that AMD is able to sell as much as they can produce. This is true but the we still see that AMD is not able to do that. The greatest FABs are not able to increase production capacity. So AMD is late with 65nm production and
they should announce production date for 65nm parts today. I hope.
5000wspm for Q4 is optimistic or realistic?

11:58 PM, October 17, 2006  
Blogger Mikolaj said...

If you look at Intel and AMD income statements look closely to total revenues and company values.
1. Market value
AMD is worth 12b+ATI 5.47b= 17.47
Intel is worth 120b
This is 7:1 !!!

2. Revenues
AMD+ATI = 1,3b+650m= ~2b per quater
Intel = ~8b per quater
This is 4:1

3. People
ATI+AMD=9800+3300 is 13100
Intel = 102,000

So based on this metrics AMD is undervalued, has small number of people and has control of 25% of the market.

Sharikou is right - it this situation Intel is loosing money or marketshare. AMD is not able to produce enough chips to compete with Pentium D/Pentium 4.
Every production imrovment means more marketshare.
22kwspm 200mm+ 5kwspm from FAB36+1kwspm from Chartered is not enough.
But provides stable growth in term of revenues and marketshare.

12:24 AM, October 18, 2006  
Anonymous Anonymous said...

ok lets make a deal.

if intel doesn't go bankrupt like you say they will, you stop this blog and never post another horribly one sided comment again.

if they do, you can bask in the glory of being right and tell everyone "I told you so".

please make the deal. please..

12:59 AM, October 18, 2006  
Anonymous Anonymous said...

Sharikou, you closet Intel fan pretending to be an AMD fanboy just to make AMD fanboys look bad...

Look at AMD. At $90, they make 57% profit. ($51.30)
They sell to Dell for $67/42.2% profit. ($28.30)

To make THE SAME profit as before Dell, they have to sell $81% more chips!

On the other hand, every Intel chip not sold to Dell means Intel can sell it for a smaller discount to somebody else (higher profit). This is possible because Dell's discount was at least 25% if you take AMD's numbers of $90 vs $67... but probably the discount is closer to 30% or higher.

Dell could even reduce price of their chips and sell them for more money because they don't need to worry about the Dell discount.

Conspiracy theories aside, doesn't anybody agree that competition breeds goodness for consumers and that the better AMD is the better Intel will become and vice versa???

Isn't it good that both companies are strong?

If Intel went bankrupt...
AMD's innovation would crash and vice versa.

Why not be happy when both companies succeed because it is the only way for consumers to be better off in the future???

1:02 AM, October 18, 2006  
Anonymous Anonymous said...

Sharikou, you closet Intel fan pretending to be an AMD fanboy just to make AMD fanboys look bad...

Look at AMD. At $90, they make 57% profit. ($51.30)
They sell to Dell for $67/42.2% profit. ($28.30)

To make THE SAME profit as before Dell, they have to sell $81% more chips!

On the other hand, every Intel chip not sold to Dell means Intel can sell it for a smaller discount to somebody else (higher profit). This is possible because Dell's discount was at least 25% if you take AMD's numbers of $90 vs $67... but probably the discount is closer to 30% or higher.

Dell could even reduce price of their chips and sell them for more money because they don't need to worry about the Dell discount.

Conspiracy theories aside, doesn't anybody agree that competition breeds goodness for consumers and that the better AMD is the better Intel will become and vice versa???

Isn't it good that both companies are strong?

If Intel went bankrupt...
AMD's innovation would crash and vice versa.

Why not be happy when both companies succeed because it is the only way for consumers to be better off in the future???

1:25 AM, October 18, 2006  
Anonymous Anonymous said...

Intel sales were told to be 8,7 B$.

Includet are Conroe/Woodcrest sales of 6 M pieces, but may have been 5 M pieces till 30th of september.

These are to be expected to be sold (not in inventory). So for the known prices 200 $ to 600 $ each, these CW#s could have made up for 1 B$ (minimum)- 3 B$ (maximum - every was a 6800).

If Conroes was at average price at 250 $ it would have made up for 1.250 B$.

This substracted from sales left 7.4 B$ for other business. A difference of 700 M$ to 900 M$ in relation to Q2 numbers.

If this difference is a result on less 8xx, 9xx, celeron - cpu's you can figure out the number of LOST sales in lowercost cpu's :

Divide 700 M $ by average lower cost cpu !

If ASP of lower cost cpu was 50 $, then the sale of a number of 14 M cpu's did not take place.

This offsetted by 5 M Conroes would result in a CPU ( by number ) dissale of 7 M units (a higher asp would decreas this number).

IF AMD could have managed this number (??? we know of capacity constrainments), this number would have gon to AMD (marketshare increase by numbers).

If only ONE THIRD of this TRUTH is true, the amd projected sales for Q3 (http://www.hkepc.com/bbs/hwdb.php?tid=648852&tp=amd-x2-3600-2&rid=651569) could have been realised.

1:32 AM, October 18, 2006  
Anonymous Anonymous said...

I bet if i make the same calculation tonight with AMD showing a much lower number you will tell us that all is fine and according to plan. But besides that I just made the calculation on last quarter and AMD looks like it is in good shape if you measure is one to go by. The only problem with comparing the 2 companies this way is that you do not factor in shareholder value and do remember that companies like these are run to provide profit for the shareholders. If they have to much cash lying around the shareholders are not getting the best value out of their money. Excess cash must either be invested in new profit making areas or paid to the shareholders ad a dividend. Therefore big companies seldom have more money in cash than they need to operate to company.

Also you seems to forget that Intel’s total assets are almost 47 billion and when calculation a company’s value you take all assets into consideration not just the cash. Should the unlikely thing happen that Intel start loosing money and they spend all their cash they should still be able to borrow about 26 billion (18 billion in plants + 4 billion inventory + 3 billion long term investments + 4 billion long term assets - 3 billion long therm debt).

2:18 AM, October 18, 2006  
Blogger Darius said...

Heh. Please do the same analysis on AMD balance sheet. With your logic you will prolly arrive to conclusion that AMD could BK any day now.

2:40 AM, October 18, 2006  
Anonymous Anonymous said...

yay! its more stupid math! inventory is an asset, not a liability, moron!

unless you think they are going to write off $4B in inventory, which would be yet another baseless sharikou prediction. (see Q306 "massive operating loss" claim)

4:23 AM, October 18, 2006  
Blogger Sharikou, Ph. D said...

inventory is an asset,

At the time Intel BKs, it will carry the same amount of inventory, all junk.

6:30 AM, October 18, 2006  
Anonymous xvhkdqu said...

Just for fun, I called up a couple local stores to see what was happening with their old Intel stock. I figured they would say, "Oh, it's okay for sales" or "It's getting quiet".

Believe it or not, the Pentium D's are selling like hotcakes, not from just one or two stores, but ALL of the four very diverse stores that I called.

If you look at the marketing around the D's, people are being fooled into buying them - "Pentium D Dual-Core Do More". Or 'newest dual-core free upgrade'. And people are so used to understanding Mhz to be faster, they look at 3.6 ghz and believe that it's better.

I'm thinking that Intel will have to write off some of the P4 stock but the Pentium D's are flying off the shelves. I'd suspect the P4's are being schlocked off to businesses at a discounted rate, but aren't a big problem either.

Who is suffering from losses is the big question? I don't believe that Intel is taking the hit as much as the distributors.

Sharikou, I don't think your projections will remotely be close, especially in the 'inventory' deduction in your equation.

One last comment - people, it's spelled 'lose' not 'loose'. 'Loose' is how you would describe the lady two houses down that has 6 kids and 4 ex-husbands. Jeeze, how did that word ever become the problem it is today?

7:03 AM, October 18, 2006  
Anonymous Anonymous said...

At the time Intel BKs, it will carry the same amount of inventory, all junk.

bullshit. Even in the unlikley event of an intel Bankrupcy the inventory will still have a value. You automaticly asume that since it is not Core 2 it is crap and unsellable. This is simply not true as you just seen as intel is
still profitable. they can easily sell the older processors just because of the name and the low price.

7:10 AM, October 18, 2006  
Blogger Greg said...

Ya, where is the AMD balance sheet. I haven't seen anyone who has published it, and that's depressing, because with Intel's miniscule improvement, and the strong sales we saw this quarter, I'm betting AMD did very well. But that's still to be seen.

Too many of you are assuming that Sharikou is saying that Intel will go out of business. Even if they file chapter eleven, they could easily restructure very quickly and become competitive again. They could also easily just shrink, not restructure, and lead to a reversal. You never know. Intel isn't going away, but the fact that they're spending more than they earn (thus smaller amounts of money in assets) means their business practices are unhealthy for this time of year, or they're planning to change the way they do business.

7:27 AM, October 18, 2006  
Anonymous Anonymous said...

"If you look at Intel and AMD income statements look closely to total revenues and company values."

What about similar comparison of profits? Take the revenue for both companies and divide by profit to see how mych profit every spent dollar earns.

7:33 AM, October 18, 2006  
Anonymous Anonymous said...

I don't know how you can expect intel to go bankrupt when clearly it's competition is taking a beating and selling most of its products at bargain deals to Dell. Intel announced that they've regain market share on all segments, interest on its new products are strong without competition, while AMD products continue to have a price war with netbursts. For intel to go bankrupt, you need AMD to do the killing. All i see is the reverse until end 2009.

8:17 AM, October 18, 2006  
Anonymous Anonymous said...

At the time Intel BKs, it will carry the same amount of inventory, all junk.

Sorry, 'tis an asset. If it were truly worthless, that means they couldn't sell it at all. I guarantee you it would be more in their interest to dump that inventory for $10-20 ASP than to write it off as worthless. Why? Because they would destroy AMD in the process, and recognize revenue in the process.

Of course, if you really did an analysis (which you did not), you would see that raw materials is 0.5 billion of the inventory figure- hardly worthless. That's almost 5X more in raw materials inventory than AMD will profit this quarter. You might also note that Andy Bryant commented in the CC that inventory build is increasing for 65nm node products (ie Conroe) and declining for 90nm CPU products (but not chipsets). Implication? Intel is actually building inventory of the stuff they'll sell in the Q4 buying uptick. As opposed to leaving the channel for want... which they learned the hard way is not good business.

8:59 AM, October 18, 2006  
Blogger Isaac said...

OK, I was unsure up to now whether this blog was for real, but after reading your latest comment I am convinced. This blog just has to be a brilliant piece of performance art - it is just impossible for someone to deliberately be that obtuse. The giveaway is your calculation:

Curret assets -current liabilities - inventory = $17.6 - $9.05 - $4.48 = $4.07

Now, as someone else pointed out, inventory is an asset. OK, so it is conceivable that you could seriously argue that this inventory is worthless because all of Intel's chips are junk. The giveaway that you can't possibly be serious, however, is that you aren't content with assuming this and actually SUBTRACT the value of inventory from current assets. In the worst conceivable case, Intel could put all it's inventory into a bonfire tomorrow. The result would be that the asset would disappear from the balance sheet at no cost to them. This is the worst case - there is no conceivable justification for subtracting the inventory value from current assets.

Therefore, this entire blog must have been a brilliant and prolonged parody, one of the best I've seen. I congratulate you sir!

9:08 AM, October 18, 2006  
Anonymous Anonymous said...

Give up this obesession with intel BK. You embarass people who really like AMD and wish it well.

9:17 AM, October 18, 2006  
Anonymous Anonymous said...

http://www.businessweek.com/technology/content/oct2006/tc20061017_137788.htm?chan=technology_technology+index+page_today%27s+top+stories
Business Weeks top story, Intel decline continues as AMD puts the pressure on.
It appears Intel declining market share problems have worsened by more than another 10% loss to AMD.
Many AMD chips are in short supply but Intel has a massive inventory of chips that just wont sell and nobody wants including the new conroes.
AMD is enjoying huge sales gains with many chip shortages of top performing models.
AMD has increased more than 10% market share as Intel flounders in misfortune of huge inventories and declining chip prices.
AMD chip prices are rising but Intels continue to decline.
It appears that the supercharged pentium 3 conroe is no match for the AMD Market Leader OPTERON with HYPER TRANSPORT.

9:20 AM, October 18, 2006  
Blogger 180 Sharikou said...

Sharikou - I'm very curious how you arrived at the conclusions in this post. Either you are intentionally misinterpreting the data from the earnings report or you just don't understand how to read it.

Some flaws in your thinking:

1. If all Netburst was crap how come Intel coninues to sell these products? And if they can continue to sell them now in spite of Athlon and Core 2 being better products (the hypothesis for your claim they are junk) then on what basis do you claim they will have to trash it all in the next 2-3 qtrs before Netburst is EOLed.

How can you be sure:

2. That Intel will turn 500 million of raw materials into Netburst and hence it's junk in your calculation.

3. That the rest of the 4 billion inventory is all Netburst and hence it's junk.

Inventory is an asset until written off. So far, they have written off 100 million out of ~4.4 billion. Hardly an indicator they will write off 4 billion of inventory.

But the most important thing is you continue to miss on your predictions on the big write off on the sales of the communications businesses and the GAAP loss. Having missed in Q2...you have given yourself a life till Q4. The indicators are completely the opposite with 1.4 billion profit and an increased Q4 forecast. Perhaps you should retract this claim before it's too late.

Visit my blog over the weekend when I'll be able to sit down and do an assessment of both Intel & AMD's earnings. Perhaps that will help you improve your understanding of how to read the data and figure out what's happening to the business.

http://sharikou180.blogspot.com
(A more balanced POV)

9:39 AM, October 18, 2006  
Anonymous Anonymous said...

Intel Chief Financial Officer Andy Bryant reported to Reuters stated “world's biggest chipmaker gained market share back from its smaller rival AMD”. I wonder if that is a temporary claim if they are including P4 inventory dumping at ridiculous low prices.

If it is true AMD is selling all it can produce then someone is doing creative book keeping or lying or both. Both companies can’t be gaining market share.

9:54 AM, October 18, 2006  
Blogger Sharikou, Ph. D said...

Inventory is an asset until written off.

The point is, assuming Intel continues its operation, it will always carry 4 billion inventory at any given moment of time, at the time of its BK, these inventory worth 0, because they are unwanted. You can't use that inventory, even if it's finished CPUs to pay debt. So you have to deduct inventory from the current assets in BK analysis.

10:39 AM, October 18, 2006  
Anonymous Edward said...

"Look at AMD. At $90, they make 57% profit. ($51.30)
They sell to Dell for $67/42.2% profit. ($28.30)

To make THE SAME profit as before Dell, they have to sell $81% more chips!"


No, you do not understand, semiconductor is a fixed-cost business. The cost of production is roughly the same whether you make more or less products.

AMD's Q2 gross margin was 57%, which means at that average selling price AMD only need 40% of its production to break even. Any more sale adds to profit - they could be sold as cheap as the material cost (something like $20-$30 per wafer) with no negative effect on profit.

This applies equally to Intel, BTW. The only difference is that AMD's sales volume is increasing, while Intel's is decreasing (year on year).

10:48 AM, October 18, 2006  
Anonymous Anonymous said...

If Intel could afford to buy nVidia, they would have by now. It is obvious to me that Intel must be facing serious financial problems.

10:59 AM, October 18, 2006  
Anonymous PENIX said...

I predict that Intel will file for BK sooner rather than later. The AMD 4x4 platform will be the nail in the coffin for Intel. After the release, the value of the conroe and P4 will plumet to near 0, because chips that do not sell are worthless.

Intel will be looking to file for BK with the largest loss numbers possible. If they declare BK immediately, they can write off their entire inventory at current market value. If they wait, the market value will be 0 and their numbers will be much smaller.

Intel chips are worth more in a landfill than they are in stores.

11:13 AM, October 18, 2006  
Anonymous Anonymous said...

The point is, assuming Intel continues its operation, it will always carry 4 billion inventory at any given moment of time, at the time of its BK, these inventory worth 0, because they are unwanted. You can't use that inventory, even if it's finished CPUs to pay debt. So you have to deduct inventory from the current assets in BK analysis.

Selective listener?
Obtuse?
Ignorant?
Or all of the above?

Let me state clearly, and you can address one by one:

0.5B in raw materials is never worthless.

At clearance pricing, finished goods are still have a finite value >0. Bankruptcy in no way, shape, or form changes demand for product.

Your assumption that WIP is all low-end product that can not sell is spurious. I believe the CFO more than you. Prove otherwise and I'll change my mind.

Intel is still generating substantial amounts of cash. Hard to go bankrupt as long as FCF is still being generated. And one-time events are clearly a very small portion of this cash flow: positives and negatives totalled ~$87M out of 1.3B, and again- the CFO said no inventory writedown expected in Q4.

Please refute on a point-by-point basis. Glittering generalities and weak logic will receive zero credit.

11:18 AM, October 18, 2006  
Anonymous Anonymous said...

Sharikou - AMD will go south today after market close. $20? How much do Hector dear to lie ...

Run the numbers again in 2 hrs and cry ...
Money talks.

11:36 AM, October 18, 2006  
Anonymous Anonymous said...

Very interesting, but, anybody here know what "Free Cash Flow Analisys" mean??? That is what you should be really doing to "predict" a BK situation, is Intel cannot respong to its cash flow needs. I understand AMD tech edge and the fan position about it, because I like it too, but it is unbeliable that you state something like Intel is going BK. There is a famous saying about "Too Big To Fail" that I believe applies to Intel. Anyway, I would like more to read an opinion of Sharikou about today AMD news (Before them, of course) to have a comparison tomorrow.

12:15 PM, October 18, 2006  
Anonymous Anonymous said...

the point he is making guys is that they are still producing 75% D's. So that 4billion in inventory will stay in inventory and also be mostly D's. It's not like they have stopped producing D's, and can just keep pooling from the inventory they have. So if they do go BK, what he is saying is true. That they will have this 4Billion worth of useless chips that they can't sell so are removed from assets and written off.

12:36 PM, October 18, 2006  
Anonymous Anonymous said...

To put Intels status into perspective Amercian airline just reported they had revenue of $5.8b and made a massive $15m profit, compared to Intel which had $8.6b and $1.3b respectively.

If Intel is going to go bust then American Airlines must be really worried :)

Can a company that makes $1.3b in 3 months actually go bust or is it wishful thinking ?

I think AMD might make a losss this report.

12:49 PM, October 18, 2006  
Anonymous Graham said...

Both companies can’t be gaining market share.

Yes both companies could theoretically gain market share. AMD and INtel are not the only PC processor suppliers. THey just own about 96% of the market. If both companies took 1% of the that 4% they don't both own, then they could both own market share. Via is a pretty small player but they do have a share. Does Transmeta still have any marketshare?

1:49 PM, October 18, 2006  
Anonymous Anonymous said...

"Some readers asked why inventory is excluded in computing Intel's liquidity. The point is, assuming Intel continues its operation, it will always carry 4 billion inventory at any given moment of time. At the time of its BK, these inventory worth 0, because they are unwanted."

So if you assume inventory to be excluded why is your total liquid asset calculation SUBTRACTING inventory? Shouldn't this just be Curret assets -current liabilities = 17.6-9.05 = $8.55?

Why are you subtracting inventory as a liability/debt?

"The point is, assuming Intel continues its operation, it will always carry 4 billion inventory at any given moment of time."

This is a terrible assumption! Has Intel ALWAYS carried $4Bil in inventory? This # will obviously fluctuate over time- you have taken the last 2 quarters as data points and have drawn a straight line....

2:20 PM, October 18, 2006  
Anonymous Anonymous said...

"No, you do not understand, semiconductor is a fixed-cost business. The cost of production is roughly the same whether you make more or less products."

So wafer cost is fixed to....one would think that would be a variable cost as spending on Si would depend on the # of wafers. Chems/gases, spares, test wafers, labor? This is ~1/3 (maybe slightly more) of actual wafer cost.

Equipment is ~50% - which I think is what you are arguing about this being a fixed cost business. Then start talking about cost of distribution and sales which may or may not be considered "fixed cost"

So outside of equipment/fab what else exactly is fixed cost?

2:34 PM, October 18, 2006  
Anonymous Anonymous said...

Even assuming that Intel ever ended up in this position (which, let's face it, is as likely as Jessica Simpson being in the position to win an Oscar), odds are that Dubya and his corrupt buddies would somehow bail them out. So, the odds of your least favourite semiconductor company disappearing in the near future are the same as Jessica Simpson ever... ah, you get the idea.

4:28 PM, October 18, 2006  
Anonymous Edward said...

"Equipment is ~50% - which I think is what you are arguing about this being a fixed cost business. ...
So outside of equipment/fab what else exactly is fixed cost?"


Equipment and fab operation, including power/maintenance and human resource, are the major expense of semiconductor production, which is mostly fixed. A line running at 50% capacity takes abou the same power & maintenance & human workforce as a line at 100% capacity.

The raw material cost is relatively cheap - AFAIK it's around $30 per wafer (6 in.) at small quantity. For 8 in. it could be a bit more expensive, but not much when buying in large quantity like AMD. Even conservatively at $60 per wafer and 150 good dies/wafer, it's only $0.4 per die.

Packaging and distribution cost are also mostly fixed. Packaging materials are even cheaper, where the line operation, testing, and binning are the main cost. These costs are fixed whether a line is running at 50% or 100%.

So yes, the semiconductor chip production can be seen as a fixed-cost business, with ~ 1% variation on the final selling price.

4:41 PM, October 18, 2006  
Anonymous Anonymous said...

"Sharikou - AMD will go south today after market close. $20? "

You were right.. AMD did go south on the markets today. Investors are worried and with reason. AMd now need to sell more CPU's in order to make the same amount of money they did for 3 years now. With production at Capacity they simply will not be able to sustain another price drop. Therefore Intel's 45nm technology may put a serious strain on AMD's financial portfolio. We will see.

6:05 PM, October 18, 2006  
Anonymous Anonymous said...

Edward - do yourself a favor and stop commenting on Si processing.

You're saying a 300mm SSOI wafer is $60? ("conservatively"?)

Just to show how insane you are -

1)what is the area increase going from 6" to 12" (>4X)
2) Additional SOI cost or is that free when you do the comparison to the 6" cost you are quoting?
3) SSOI cost or is the strained Si layer that is transferred the same cost as Si?
4) think the yield rate for 12" wafer boules is as good as 6"?


A 300mm bare Si p+ substrate is >$300 wafer (how much depends on volume) and this is not Epi grade quality which is needed for actual device growth (only useful for test wafers).

Throw in SOI cost and strain layer cost and Si becomes ~10-20% of the overall wafer fabrication cost.

Out of curiosity is the 6" # you are quoting based on P+, epi (p-), SOI, SSOI? Do you even have a clue?

6:36 PM, October 18, 2006  
Anonymous Anonymous said...

"A line running at 50% capacity takes abou the same power & maintenance & human workforce as a line at 100% capacity."

This may be true, but chemical,gas, spare part, and Si consumption is variable - this is at least 30% of overall wafer production cost.

Some examples for CMP (>12 steps in the process flow) - pads and slurry are almost as much as a factor as equipment cost when depreciated over 4 years, low T SiN precurors are >$1/gram, photoresists, Si test wafers...

6:42 PM, October 18, 2006  
Anonymous Anonymous said...

Estimate on 300mm epi wafer cost (table 2):

http://www.reed-electronics.com/SEMICONDUCTOR/article/CA420729?pubdate=06%2F01%2F2004

Yes, this is several years old, and is an estimate, but note this doesn't even include SOI yet alone SSOI...and I think is a better source than your out of thin air estimate base on 6" wafer cost...

So you are likely 10X off on your estimate... conservatively at $500 for an SSOI wafer (assuming added cost of SOI and SSOI is offset by volume discount) puts Si cost at a minimum of 15-20% of overall Si wafer production cost. (If you would like me to embarass you yet again I can send a link with full Si production cost estimates...)

So, if starting Si is a variable cost, Si alone represents a 15-20% variable cost as opposed to your +/-1% estimate pulled out of the air with no facts whatsoever.

Throw in chemicals, gases and equipment spare variable costs and suddenly >30% of finished wafer cost is no longer "fixed". If you would like me to provide some chemical costs let me know...

7:08 PM, October 18, 2006  
Anonymous Edward said...

"A 300mm bare Si p+ substrate is >$300 wafer (how much depends on volume) and this is not Epi grade quality which is needed for actual device growth (only useful for test wafers)."

True. My estimation was off about an order of magnitude. The variable cost is about 10%, not 1%. I stand corrected.

However, adding a strained silicon layer to the process, which is one extra processing step like any other layer, should add to the fixed operation cost, not variable cost.

Still, a 10% variable cost, or even a 15% one, is not significant. The semiconductor production is a fixed-cost business, off 1/7 to 1/10.

Also note that my # die/wafer and raw material price was based on 8'' (200mm) wafer size; I did not mention 300mm at all. And, IIRC, the yield of a larger wafer is about the same as the yield of a smaller one. The main factor is the area per die, not the total wafer area.

12:16 AM, October 19, 2006  
Anonymous Anonymous said...

"The point is, assuming Intel continues its operation, it will always carry 4 billion inventory at any given moment of time, at the time of its BK, these inventory worth 0, because they are unwanted."

So if Ford goes bankrupt, using Sharikou economics, the value of all those cars and trucks in showrooms is instantly zero, because they are unwanted!

You are brilliant, there is just no other way to put it. Your understanding of economics, balance sheets, benchmarking and marketing is a wonder to behold.

Ever consider a career as a skidmark in my FOTL's?

12:30 AM, October 19, 2006  
Anonymous Anonymous said...

"However, adding a strained silicon layer to the process, which is one extra processing step like any other layer, should add to the fixed operation cost, not variable cost.

Edward, Edward, Edward...you just stop. You are clearly guessing/hoping and have no idea what you are talking about:

"One step procss"
1) SSOI involves growing an oxide layer on bare Si wafer (not that expensive), growing a strained layer on a SEPARATE donor wafer (somewhat expensive), bonding the 2 wafers together (probably resonably cheap), and then cleaving ("Smart cut") the strained Si transfer layer from the donor Si wafer. An adiditional implant step is done to achieve this smart cut process. This requires a lot of additional processing equipment at the wafer supplier and also introduces more yield loss issues.

So essentially you are using 2 bare Si wafers to make every 1 SOI (or SSOI) wafer. The only thing that is not clear is whether the donor wafer can be re-used several times. The link/# above is price for a non-SOI wafer.

Needless to say this is MUCH more than a 1 step process and adds substantial cost over a bare Si wafer (I've seen estimates of anywhere from 50-90% more)

The ~$300 is for a bare Si test wafer, device quality Si wafer is more than that not to mention an SSOI wafer that is device quality grade.

"Still, a 10% variable cost, or even a 15% one, is not significant. The semiconductor production is a fixed-cost business, off 1/7 to 1/10."

THIS IS JUST THE SILICON! You are assuming Si is the ONLY variable cost and are completely ignoring chemical, gas and material cost which are all variable costs and will vary from 15-30% of the wafer cost.

If you lower the # to ~60% fixed cost you then may be in the right ballpark.

In any event the 1%-2% you previously claimed is just flat out wrong and completely pulled out of the air....

3:14 PM, October 19, 2006  
Anonymous Anonymous said...

Anonymous said...

"The point is, assuming Intel continues its operation, it will always carry 4 billion inventory at any given moment of time, at the time of its BK, these inventory worth 0, because they are unwanted."

So if Ford goes bankrupt, using Sharikou economics, the value of all those cars and trucks in showrooms is instantly zero, because they are unwanted!


Compare apples to apples please. A cpu is useless without a motherboard. Some of the finished processors MIGHT have some value. Are you on the look out for any socket A cpus? If you are, where can you find them? Direct from AMD? If Intel really goes bankrupt, what are the chances that their partners are quite aware of that before Intel enters into chapter 11 and have already phased out/cut down Intel related products and gone full AMD too?

If Ford goes bankrupt, would YOU buy one of their cars or trucks knowing that you may have a problem getting spare parts later?

Diss Sharikou all you like. In the event that Intel does go BK, you bet the value of any existing Intel product is likely to go way down.


You are brilliant, there is just no other way to put it. Your understanding of economics, balance sheets, benchmarking and marketing is a wonder to behold.

Ever consider a career as a skidmark in my FOTL's?


Ah, part of what Intel worshippers are good at. Lies, abuse, short sighted reasoning and faulty reasoning.

I have some Lada cars for sale. They still move. Interested?

11:01 PM, October 19, 2006  
Anonymous Edward said...

"Needless to say this is MUCH more than a 1 step process and adds substantial cost over a bare Si wafer (I've seen estimates of anywhere from 50-90% more)

The ~$300 is for a bare Si test wafer, device quality Si wafer is more than that not to mention an SSOI wafer that is device quality grade."


1) Anyone knows that no "step" is really one step, unless you are talking about an action. Yes, making a strained silicon layer takes many small steps, but in terms of the processing flow, it is just one step: make a strained silicon layer. The point is, it adds to the fixed cost, not the variable cost.

2) My estimate was not based on $300 per bulk Si wafer, but $600 per SOI wafer. Please read and understand before you BS-ting more.

4) Gas & material blah blah cost 30% of wafer price, fine, that's why I said 15% after the 10%. I guess you really have a reading problem, don't you?

5) Semiconductor chip manufecturing is a fixed-cost business, with about 1/10 variables. Whatever your BS argument is, this is just the fact. This is how the fabs run and make/lose money every day.

Okay, so you are reading challenged, lets see if you are also math challenged. Assume a 300mm SOI cost $600, and you get only 150 good dies out of one wafer. Plus 30% gas & material and another 20% purely because I pity you, how much is the variable cost per chip? Did I hear $6, or did I see a math challenged dude scratching his head?

Isn't AMD's Q3 ASP around $90? Now what is the variable cost percentage? No, it's not 1% (I never said 2%, BTW; seem you are SERIOUSLY reading challenged), but did I make a correct estimate of 10% or less?

11:06 PM, October 19, 2006  
Anonymous enumae said...

Anonymous said...

"A cpu is useless without a motherboard."

Does that mean you are assuming that in Intels $4 Billion inventory that there are no motherboards?

"...Intel enters into chapter 11 and have already phased out/cut down Intel related products and gone full AMD too?"

Hate to be a nag, but AMD can not supply the worlds demand for processors, motherboards etc...

If Intel goes down, prices will sky rocket.

"In the event that Intel does go BK, you bet the value of any existing Intel product is likely to go way down."

Sorry, but I think your wrong, it would be harder to get replacement parts thus raising the prices.

-------------

Edward said...

"Isn't AMD's Q3 ASP around $90? Now what is the variable cost percentage?"

Hey Edward I am not trying to interfere with your debate, but I have a questin.

Sharikou has stated repeatedly that the cost for AMD to produce a chip is about $40.

So why are you basing your variable cost on the ASP, wouldn't the variable cost be based on the cost of production?

$40 * 150 (good dies) = $6000 / 2 (30% gas & material and another 20%...) = $3000

$600 (wafer cost) * 100 / $3000 = 20%.

So could I conclude that the wafer is 20% if the variable cost?

Thanks.

11:33 AM, October 20, 2006  
Anonymous Anonymous said...

Edward, you are truly ignorant, and I warned you to stop but now I must correct you:

AMD BUYS the SSOI layer (THEY DO NOT GROW THIS LAYER THEMSELVES!). As such this is an increased cost of purchasing the substrate and is a VARIABLE cost! SOITEC does all of the bonding, strain growing, cleaving and passes this on as additional cost of substrate (vs a standard bare Si wafer) Note SSOI is DIFFERENT than embedded SiGe strain which is a part of the process flow and is truly a single step (this has been done at Intel since 90nm and IBM/AMD will be introducing this on 65mn)

2) $600 per wafer is ~1/10 of overall wafer cost meaning every wafer produced is ~$6000? Wrong - just check Sematech benchmarks; the "older technologies" (0.25,0.18,0.13) run in the 2000-3000 range. In this case Si would represent 20% of the overall wafer production cost.

3) Well there was no #3 must have trouble counting up sequential #'s...

4) Gas & material blah blah cost 30% of wafer price, fine, that's why I said 15% after the 10%. I guess you really have a reading problem, don't you?

You said 1/7-1/10 overall variable cost after you absolutlely ridiculous claim of 1/50-1/100 variable cost (that's the 1-2% original claim for the mathe impaired). 1/10 of $4000 wafer is 400 which doesn't eveb cover the cost of the SOI substrate yet alone chems/gases.... 1/7 is ~600 which ONLY accounts for the wafer itself (assuming it is only $600)

Just to give you a clue - 1 CMP pad alone runs in the 200-500/pad range, typically a 2-3 platen process with a pad life of ~1000 wafers (meaning 2-3 pads for every 1000 wafers processed. So for simple math assume 3 platen process * 3 pads @333 = $1000 for every 1000 wafers or $1/wafer

$1/wafer, well that is just ridiculously small no? Ok now multiply this by the 12 or so CMP steps, throw in cost of slurry (which ain't cheap either). Now do this caluclation for EVERY tool in the factory and you may begin to understand chemicals and gases are not a trivial cost.

5)"Semiconductor chip manufecturing is a fixed-cost business, with about 1/10 variables. Whatever your BS argument is, this is just the fact. This is how the fabs run and make/lose money every day."

One day ago you were trying to convince people it was 1-2%, now I put out your stupidity with just Silicon cost alone being >10%. Now you're back stating 10%. Why? Becasue you say so - you are still ignoring MANY other variable costs (spares, packaging, chems/gases)

I've given you several examples of variable costs, including some links, none of which you are actually able to refute and we are supposed to take your new calculation word for it.

As you don't even know anything about SSOI ("it's only an extra step"); your credibility is not quite that great...

As for you math challenge - you've completely ignored testing, dicing, packaging (which alone is a few bucks/chip), & distribution. By your original, oustanding math, a 1-2% variable cost, with $4-6 going to SOI means $400-600/chip! (cost)

Oh and your math sucks, $600 SOI cost/wafer divided by 150die wafer is $4. If you add an additional 20%+30% that's 20+30% of PROCESSED WAFER (meaning out of ~$3000-4000) cost not unprocessed Si wafer cost! meaning and additional $8-12/wafer.
I wasn't stating chem/gases is 20% of the starting si wafer cost, it's the % of overall wafer processing cost...

Please educate us some more on this whole fixed cost thing...

Please stop with these ridiculous 1-2% (now 10%) claims without any supporting data...its really get tiring proving you wrong...

4:31 PM, October 20, 2006  
Anonymous Edward said...

"AMD BUYS the SSOI layer (THEY DO NOT GROW THIS LAYER THEMSELVES!)."

I had not said SSOI, because I remember reading somewhere that AMD will use SSOI wafer in 45nm die to high defect ratio (which induces higher variable cos).

It is you who's been saying SSOI over and over.

"If you add an additional 20%+30% that's 20+30% of PROCESSED WAFER (meaning out of ~$3000-4000) cost not unprocessed Si wafer cost! meaning and additional $8-12/wafer."

No, the processed wafer cost depends on the volume, because it is a fixed cost. I was not caluclating the cost to process those wafers (which is fixed cost), but the material (gas, metal, ceramics) cost of those processing (which is variable).

"One day ago you were trying to convince people it was 1-2%, now I put out your stupidity with just Silicon cost alone being >10%."

If you so much wish to focus on the previous misinformation, so be it. People with 10-year-old mind set like you jump up and down when I made a wrong estimate, but the correction does't change the fact that the semiconductor industry is a fixed cost busines.

I know all these are too difficult for you to understand, and I have no desire to teach you. Just advise you not to become the manager level in the fab company - because your company will likely bleed to bankruptcy.

9:36 AM, October 21, 2006  
Anonymous Edward said...

"So why are you basing your variable cost on the ASP, wouldn't the variable cost be based on the cost of production?"

Because the cost of production is mostly fixed, and thus it varies depending on the production volume. ASP, however, is more or less relatively stable by the market.

If AMD (or Intel) could sell and have the capacity to seel infinite number of chips, its production cost could go down to $10/chip level or even lower. At that point, the business is not a fixed-cost one any more - because the variable cost now dominates.

However, right now it is still a fixed-cost dominated business. That is why the production cost per chip goes down dramatically when a fab goes from 200mm to 300mm (which uses more expensive equipments and twice the amount of materials).

9:48 AM, October 21, 2006  
Anonymous Edward said...

"$1/wafer, well that is just ridiculously small no? Ok now multiply this by the 12 or so CMP steps, throw in cost of slurry (which ain't cheap either). Now do this caluclation for EVERY tool in the factory and you may begin to understand chemicals and gases are not a trivial cost."

Why don't you finish the calculation, though? Lets assume everything in your way, that the CMP pads charge is $12 per wafer (in fact, there is no reason that the tools were not purchased with enough supplies over their depreciation lifetimes and put those charges into fixed cost - but whatever, lets have it your way just because I pity you). Lets also assume there are 30 tools, all cost as much as $12 per wafer each, amount to an extra $360 per wafer total. With 150 good dies per wafer, that is the cost/chip? $2.4? Oh well, I should've said $7.4 instead of $6 (I already added 20%, about $1/chip, in my original estimate). That's really not trivial!

Remember, that's after a lot of pitying you and having it your ways... In the end you just proved to me one thing - technicians or people with technicians capacity are, well, appropriate only as technicians work.

4:31 AM, October 23, 2006  
Anonymous Anonymous said...

Edward:

30 tools? What are you on 1um technology - there are well over 100 unique steps in a 90nm process flow. And actually 12/wafer is an overestimate (CMP is more heavily weighted toward variable costs)

And the numbers you used was <1/2 of the cost of one particular set of steps...

Look up cost of BTBAS, resist costs, spare parts, EP chemistries, etch gases, test wafers to condition pads, etch chambers, inline monitors - you are simply out of your world of knowledge.

You took one part of the cost of 12 steps (out of >100) and say it's minimal... Any idea on cost of slurry/gal and amount of usage/wafer? (Didn't think so).

How about low Temp Si precursors for SiN process? Litho photoresists? Packaging materials? Sputter targets? Low K precursors?

Now, how much is a chip cost? If you use Sharikou's $40 per you are already at >15% with your $7.4 #.

...and that's excluding ~90% of the tools in the process flow and assuming tools don't need any spare parts...

But then again this is someone who thinks pads can be purchased from the CMP equipment supplier (HINT: equipment suppliers don't make pads or slurry - but you knew that right? If you don't believe me I can provide the names of the top equipment and pad and slurry suppliers if you'd like to verify)

You are simply out of your league and are resortung to making stuff up now. How much does a Si wafer cost in Edward-land again?

So shut up on this topic unless you are going to bring some facts to the table to back up your 1-2% variable cost statement (or is it 10%, or 15?....)

8:17 PM, October 23, 2006  
Anonymous Anonymous said...

"in fact, there is no reason that the tools were not purchased with enough supplies over their depreciation lifetimes and put those charges into fixed cost - but whatever, lets have it your way just because I pity you..."

Spoken like someone who knows how a fab runs - Yeah I'll take one 3Mil tool and can you throw in 1800 pads with that, even though you don;t make pads? Oh and some large fries with that too!

11:51 PM, October 23, 2006  
Anonymous Anonymous said...

http://www.thomaswest.com/images/pdf/LowCostofOwnershipTungstenCMP.pdf

An older estimate of total CMP costs (based on W polish process)

Equipment (fixed) = 22%
Pads and Slurry = 61%
Labor/Misc (lets blindly assume all fixed to humor Edward) = 17%

Now CMP is a bit of an extreme case when it comes to consumales, but 10% fixed cost industry? Thanks for playing, come again!

Any data to support your position yet Edward? (let me know when you've had enough)

12:28 AM, October 24, 2006  
Anonymous Edward said...

"Look up cost of BTBAS, resist costs, spare parts, EP chemistries, etch gases, test wafers to condition pads, etch chambers, inline monitors - you are simply out of your world of knowledge."

Why don't you spell out all the variable cost if you have so many facts? How much they'll sum up to? $500/wafer? $1000/wafer? That's like what, another 5% variable cost of the chip selling price?

And I am only amazed that you would put etch chambers and inline monitor to variable cost. What are you? Out of ammo and start throwing pillows?

"Yeah I'll take one 3Mil tool and can you throw in 1800 pads with that, even though you don;t make pads?"

1. I never said to buy the pads and the tool from the same people. I guess you ARE out of ammo, aren't you? I can forgive your inability to think, but I have to correct your inability to read properly.

2. So change the fixed cost of the tool to 3.6 mil to include the 1800 pads it might never finish using during its lifetime.


"Now CMP is a bit of an extreme case when it comes to consumales,"

That's why I estimated effectively 30 of such variable cost.

"but 10% fixed cost industry?"

You didn't read my comments. It depends on the total production volume & average selling price. Plus, the 10% variable cost, not fixed cost, not 2% (which I never said), is just an estimation. I admitted - if you're too dense to read - that the 1% was off an order of magnitude; the 10% might not be exactly accurate, either. It's a ballpark; it's what engineers (unlike technicians) do everyday. The idea is that at this point CPU manufacturing is dominated by fixed cost.

It seems to me that you have lots of information but no brain to analyze them. Sad.

12:51 PM, October 24, 2006  
Anonymous Anonymous said...

OK Edward, let's go through the history here:

FIRST (INCORRECT)STATEMENT (10/18)":
"AMD's Q2 gross margin was 57%, which means at that average selling price AMD only need 40% of its production to break even. Any more sale adds to profit - they could be sold as cheap as the material cost (something like $20-$30 per wafer) with no negative effect on profit"

If I read this correctly you are saying any additional wafer produced adds a material cost of $20-30 per wafer (which I think we can all see is wrong now). The Si alone is >$300. Even if you were for some reason excluding Si, chems/gases are far more than this (CMP alone is probably more that 20-30 when you add pads and slurry)

NEXT STATEMENT:
"The raw material cost is relatively cheap - AFAIK it's around $30 per wafer (6 in.) at small quantity. For 8 in. it could be a bit more expensive, but not much when buying in large quantity like AMD. Even conservatively at $60 per wafer and 150 good dies/wafer, it's only $0.4 per die."

Well you finally acknowledge this statement was wrong (you "engineering" skills on pulling #'s out of the air must have been lacking here)

NEXT RESPONSE:
"The variable cost is about 10%, not 1%. I stand corrected."

Not sure how I've been misreading your claim of variable cost being 10% - above comment seems pretty black and white. As I've pointed out SOI (or SSOI) alone is more than 10% variable cost, putting aside tool variable costs (materials, spares, etc).

You then go on to say well it's mostly just Si as variable cost; I point out one example where fixed cost is <40% fixed and your response is now do this for every module to prove me wrong.

I find it rather funny when I have provided links and information to point out your errors your only response is do the rest of the calculation for me. When I have asked you to support your ridiculous, unsubstantiated claims your only response is I'm an engineer I just estimate. (And apparently don't use any data to generate those estimates - if this is wrong why not provide the data?)

Oh and for the record you were the idiot that said "there is no reason that the tools were not purchased with enough supplies over their depreciation lifetimes and put those charges into fixed cost" - I think even Sharikou would acknowledge the ridiculousness of this statement. I used pads as an example but I can just imagine the # of 55gal slurry totes that would mean a 4 year supply....

(oh and from an economics perspective, buying a bunch of something up front doesn't change it from a variable cost to a fixed cost. It might make it a sunk cost?)

So I'll try again any ACTUAL DATA (since you are an "engineer") to support your now 10% fixed cost business model?

I think I have pointed out by now that 10% is wrong as you have resorted to comparing variable cost to sales price (as opposed to comparing variable cost to total cost) and name calling. Even others such as enumae look to be in general, agreeing (sorry to drag yuou into this e)

And if you really think I'm a technician ask me any question you'd like about Silicon processing down to the 22nm node and we'll see (just nothing on packaging). What is your area of expertise? I'm sure I can ask you some questions which won't have easy "google" responses...

4:49 PM, October 24, 2006  
Anonymous Anonymous said...

"You didn't read my comments. It depends on the total production volume & average selling price."

% of variable cost (vs fixed cost) has nothing to do with ASP. If you say something something is 10% variable cost and 90% fixed, ASP has nothing to do with this calculation... I agree with anonymous poster above (and on the fact that the magnitude of your estimate is wrong)

4:53 PM, October 24, 2006  
Anonymous Edward said...

"I think I have pointed out by now that 10% is wrong as you have resorted to comparing variable cost to sales price (as opposed to comparing variable cost to total cost) and name calling."

I sincerely apologize for the name calling.

However, as much as you'd said 10% being wrong, and with all those links you provided, you did not give a "more correct" estimate. Kind to estimate it yourself (on AMD's volume, BTW)?

As I said, the variable cost percentage depends on volume. The "less than 10%" is calculated from ~$90 ASP and $7.4 per chip. Thus, while variable cost would not depend on ASP, the 10% does.

"And if you really think I'm a technician ask me any question you'd like about Silicon processing down to the 22nm node and we'll see (just nothing on packaging)."

I appreciate this generous offer. Let me think... How much better (power usage, switching speed, Ion vs. Ioff) it seems to perform at 22nm, versus 65nm? How much (more) quantum effects are you observing? How much smaller the transistor size will be at 22nm, versus 65nm? And, just an estimate, when will it be production ready? Hope these aren't too many, thanks in advance. ;-)

1:02 PM, October 25, 2006  
Anonymous Anonymous said...

How much better (power usage, switching speed, Ion vs. Ioff) it seems to perform at 22nm, versus 65nm? How much (more) quantum effects are you observing? How much smaller the transistor size will be at 22nm, versus 65nm? And, just an estimate, when will it be production ready?

65nm-22nm: ~1.5-2X better - depends on whether FINFET (or Intel trigate) gets implemented on 22nm node. I'm also assuming there'll be no III-V integration until at least the 15nm node. Power usage is difficult to attribite to a technology node if you mean overall chip power as design, Vcc, supply voltage and # of transistors come into play.

Transistor size is relatively fluid for 22nm; there is a lot of variation from IC manufacturer to manufacturer but it should be in the 12nm-15nm range (drawn).

Quantum effects are significant; on planar transistor technologies at 65nm it accounts for ~25% of the electrical thickness of the overall transistor electrical thickness. How much it will be on 22nm depends on how aggressively Toxe is scaled - best guess would be ~45% of electrical thickness will be quantum effect.

Production ready is obviously supplier dependent, assuming current roadmaps hold Intel will be first with logic production and this will be ~2012. This is somewhat tied to the litho readiness for EUV/next gen immersion, so error bar on this is significant.

And no, I cannot provide more estimates on variable costs - I do that for a living and cannot provide that much info as it is sensitive, the couple of examples were an attempt to show how far you were off, I can say >>25% variable.

Also you really should not use ASP. If chip costs ~$40 (we'll use Sharikou's analysis for time being); then a $10 variable cost per chip (which I think you may now concede is the very low end and unlikely to be true), would constitute 25% variable cost.

By your definition you have a <10% variable cost and you fixed costs would be ~40% and 50% other costs being unaccounted (as that is gross margin). If you want to say variable cost is 10% of the chip price that is one thing to say this is a 10% variable cost industry is entirely another.

I also encourage you not to ignore materials cost for packaging...I don;t think there are any public links for those costs but I'll look.

11:31 AM, October 26, 2006  
Anonymous Edward said...

"By your definition you have a <10% variable cost and you fixed costs would be ~40% and 50% other costs being unaccounted (as that is gross margin). If you want to say variable cost is 10% of the chip price that is one thing to say this is a 10% variable cost industry is entirely another."

Agreed. Actually by my definition the estimate was >50% gross margin, 40% fixed, and <10% variable (for AMD's volume). I used ASP as the percentage base only because (as I previously said) it is less dependent on production volume, versus cost per chip.

If we don't use ASP as base thus remove the 50% gross margin, variable cost becomes ~20% of total (fixed+variable) cost. Still I fail to see how it is possible for AMD to have variable cost >> 25%.

Lets assume a 40% is >>25%, or $16 per chip on a total cost $40/chip at AMD's current volume. That means for $80 ASP, AMD's sales volume can reduce to 37% and still remain positive gross margin! Then AMD would have been making money in 2003...

OTOH, assume variable cost is $8/chip, then AMD has to maintain 45% of its current sales volume to make gross margin positive. It seems much closer to reality to me.

I think the discrepency between your and my estimates are two-fold:
1. difference in time/technology (65/45nm in your context vs. 90nm in mine)
2. difference in volume/company (Intel in yours vs. AMD in mine)

That said, I agree that using the term "industry" was probably wrong, since it's obvious the fixed-variable ratio will be different for Intel and AMD.

BTW, thanks for your information on the 22nm prospect.

1:54 PM, October 27, 2006  
Anonymous Anonymous said...

"Still I fail to see how it is possible for AMD to have variable cost >> 25%."

For $40 cost, $10 is 25% variable cost - variable costs are significantly more than $10/chip. The #'s I provided are based only on part of one process step (actually 12) and SOI/SSOI - there are many other variable costs - my point was to show that 1-2% and 10% (4$/chip) is way off.

"since it's obvious the fixed-variable ratio will be different for Intel and AMD."

Actually not that much different (of course you will ask me to now prove this where as you can make the blanket statement without any supporting data. The ratio's a bit different between 90nm and 65nm and the substrate but not as much as you think - there is an increased processed wafer cost (and thus die cost) due to extra metal layer, additional strain steps (esp in case of AMD), increasing mix of higher cost litho tools - so increase in die/wafer (WHICH IS NOT 2X SHARIKOU!) is somewhat offset by increasing fixed (capital) and variable (chem/gas...) costs. Of course yield is the other wildcard but noone really know relative yield between AMD and Intel.

One key benefit is the silicon portion of the variable cost remains flat (although this may not be true for AMD if they are moving from SOI - SSOI). So overall fixed/variable cost ratio may go down a bit. This however is offset by reuse of equipment from one technology node to the other (can be ~60-70%) - this depends on ramp up/ramp down rates and absolute peak wafer starts for each node.

"That means for $80 ASP, AMD's sales volume can reduce to 37% and still remain positive gross margin! Then AMD would have been making money in 2003..."

Not sure I understand this comment at all - if you are comparing the 2003 sales levels to 2006 sales levels you cannot do this as the fixed cost were not the same.

Also 2003 was 200mm which is significantly different fixed/variable cost ratios than 300mm.

11:27 AM, October 28, 2006  

Post a Comment

Links to this post:

Create a Link

<< Home