Path: utzoo!utgpu!news-server.csri.toronto.edu!mailrus!uunet!samsung!uakari.primate.wisc.edu!dali!milton!uw-beaver!Teknowledge.COM!unix!garth!fouts@bozeman.ingr.com (Martin Fouts) From: fouts@bozeman.ingr.com (Martin Fouts) Newsgroups: comp.sys.super Subject: Re: Supercomputer ROI Message-ID: <292@garth.UUCP> Date: 8 May 90 19:33:45 GMT References: <201@csinc.UUCP> <253@garth.UUCP> <202@csinc.UUCP> Sender: fouts@garth.UUCP Distribution: na Organization: INTERGRAPH (APD) -- Palo Alto, CA Lines: 140 In-reply-to: rpeglar@csinc.UUCP's message of 3 May 90 14:53:35 GMT In article <202@csinc.UUCP> rpeglar@csinc.UUCP (Rob Peglar) writes: From: rpeglar@csinc.UUCP (Rob Peglar) In article <253@garth.UUCP>, fouts@bozeman.ingr.com (Martin Fouts) writes: > Let me clarify my position. I never meant 20% starting at the end of > the first year. That's why the word "reasonable" in the quote. I > would debate the assertion that other countries investors are willing > to take an 8-10% ROI on a mature company after a large investment. > This is certainly not the Japanese approach. The Japanese are willing > to go negative for a long time, but expect a large ROI for a long time > afterwards. They are interested in owning large market share. [Rob points out that Japan is willing to take smaller ROI than the US because the cost of capital is lower. I have to agree:] Right. With a 3% savings account rate, cost of capital must be very low compared to the US, so 10% ROI is still acceptable. This doesn't impact my analysis below however. > > [ My spread sheet experiment showing that 100m/y R&D over 10 > years with 10% financing followed by 5 years of good sales > at 1x markup results in 1b in expenses and an immediate > need to generate 100b per year of business just to break even > deleted. ] You're starting to see my point. The supercomputer business is sufficiently different that starting from scratch - a la ETA - cannot cope with your four constraints above. There has to be something else, an external force, to influence this. In ETA's case, 1) was more like $50 million/year. 2) was the real problem. CDC's bankers wouldn't do it. ETA's failure to be *taken public* - in other words, let the market fund you, not the banks - killed them. 3) was beginning to be reality. In early 1989, ETA was sold out of air-cooleds - they were backordered. Couldn't build 'em fast enough. Very reasonable projections showed that ETA could sell close to 100 air-cooleds in 1989, if only they could build them that fast. 4) is wrong, at least in the way I read your words. The markup necessary over cost has to be more like 4-5x for viability, 6x for good profits; not 1x like you say. For instance, the cost to build an ETA-10 CPU was (*ballpark*) $250,000. The incremental revenue from each CPU (i.e., price charged, list) was (again, ballpark) $1,000,000. That's 4x. It should have been more like 6x, i.e. $1,500,000 per CPU. However, there is and was a market perception about 1.5 million dollars; many entities (firms, universities) would spent 1.0 million, but swallowed hard at 1.5 million. No. I'm starting to refute your point. I put a 1x markup in to avoid people saying it has to be smaller. If you demand a larger markup, you need a larger share of the market to make it. The point of the exercise was to demonstrate that with the product development cycle you are describing it is not possible to make money. If you increase the markup (as you properly should to account for taxes and other overhead that I didn't describe) you need an even larger share of the market just to break even. You are seriously impeded by the short product life time of your product. Go back and do the analysis on ETA. By the time the air cooled machine was obsolete it would not have recovered half of the R&D cost.... I don't think going public would have helped, but that is harder to call. The stock market isn't really a way to finance a startup company, as the market tends to respond to the current value of the company, rather than its potential. Given the market climate at the time, a poorly received IPO might have killed ETA faster than the bankers did, but since history doesn't record alternatives, the best we can do is speculate. I think that CRI's biggest problem right now is that CRI management is too concerned with what wall street thinks. In ten years, I never saw as many CRI managers interviewed in technical rags as I've seen CRI financial interviews in the last six months. I bet the bottom line kills that company in about five to ten more years. [ Another good argument for lower cost financing in Japan deleted.] Fortunately, there are two (that I can see) ways out. One is happening now, grossly stated as the *attack of the killer micros*. I.e., use commodity parts in ingenious ways. Far less R&D cost and less time to market than "traditional" supercomputing. The other way is the path of least resistance - just decide to allow others to make/sell these machines. Go with the flow. I agree. Please note that the killer micros are a strong argument against the "old" supercomputer industry, as they are taking their performance techniques from old mainframe technology and not from supercomputer technology.... The vector unit will probably die out with the Cray (;-) > > However, the life cycle of the machines was unlikely to be long enough > to ever recover the entire cost of those first six years, let alone > make a profit. True again, if you think R&D stops. See below. > > I hope the readers are smart enough to realize that it doesn't make > sense to develop a product whose R&D time is twice its likely market > viability. (For those readers who don't believe the 5 year life > cycle, consider that CRI introduced the Cray X/MP in 1984 and the Cray > 2 in 1985. The 2 is no longer sold, and the X/MP is being replaced by > the Y/MP in the market place.) Bingo. Your profits, hopefully, are enough to carry your R&D through into the next cycle. This is why the funding for startups is a critical issue. Once "kick-started", vendors can survive. CRI is the perfect example - although, in many ways, CRI is at a crucial point in their company history. Nope. Your technology is obsolete. You don't get to carry any of the R&D over. If it wasn't obsolete, you would still be using it. You might get to take 8 years next time instead of 10 because you've finally developed a team with the necessary experience, but the history of the computer industry is even against that, since very few people stay with a project for 10 years and almost noone stays with it for 18. You are also carrying the cost burden. > The trick is to do the R&D cheaply enough or fast enough that the > product has a long enough market life time to recover costs and make > money. If you can't do that, you shouldn't be in business. True. The point is to let the market, not the bankers, decide. We agree here. I only claim that there just isn't any market for products whose R&D cycle is twice their life cycle. -- Rob Peglar Control Systems, Inc. 2675 Patton Rd., St. Paul MN 55113 ...uunet!csinc!rpeglar 612-631-7800 The posting above does not necessarily represent the policies of my employer. -- Martin Fouts UUCP: ...!pyramid!garth!fouts ARPA: apd!fouts@ingr.com PHONE: (415) 852-2310 FAX: (415) 856-9224 MAIL: 2400 Geng Road, Palo Alto, CA, 94303 If you can find an opinion in my posting, please let me know. I don't have opinions, only misconceptions.