As a rookie reporter, my first assignment was to follow the unexciting world of microcontrollers with the ferocity of a hound dog, and if I did a great job, the editors would let me write about Flash memory and eventually the DRAM business. This was before commercial Internet led to first a global delusionary disorder of gigantic proportions.
While tracking the business of silicon, I came to expect the release of the monthly book-to-bill ratio and the annual Semiconductor Industry Association’s annual forecasts with similar enthusiasm (and dread) as People magazine reporters anticipated the Oscars.
Analyzing the B-to-B ratios and SIA data was like reading tealeaves and trying to predict the long-term impact on overall tech business. Fast forward to today, and the news of SIA basically admitting that the 2007 semiconductor sales were going to be slower that expected (1.8% versus previously forecast 10%) didn’t even merit a sneeze on Wall Street. In fact the chip stocks rose, and ended the day up 2.9%, besting the S&P 500.
The SIA press release, in a nutshell, tells you that even though the demand for silicon keeps on increasing, the revenues and profits continue to move in the opposite direction. The Flash memory is everywhere – in our pockets, pumping tunes into our ear buds, storing our photos and capturing life on video. Macbook Pros ship with more memory than I could imagine in my rookie days as a chip reporter.
The big yawn from Wall Street to this report, according to my good buddy and veteran technology analyst Pip Coburn (the only person in my rolodex with more gray hair than me) is that there are no wild swings in the growth rate of chips. The industry is big and boring – and well a bit of a blah.
“Key backdrop is that from 1965 to 2000 the CAGR for semis was a whopping 17%, then in 2001 revenues went down 33%,” he says. SIA has forecasted a CAGR of 5.4% through 2010. “Now the assumed trend line growth is a mere 8-10%! So the wild fluctuations are done! Since they are more modest fluctuations the benefit of “playing the cycle” is much less and therefore the book-to-bill as a data point is more modest,” Pip adds.
For someone who has always looked at the world from a bottom-up perspective – chips being the tarot cards of the tech industry’s health – sometimes I wonder if there will be a new dynamic one needs to look at? I wonder if bandwidth is a metric that one needs to track more closely in order to make reasonable guesses about the future. Any thoughts?
Photo courtesy of Intel Corp.
I use the semi data religiously, and predicted more than a year ago that the 2007 estimates were too high.
Assuming the street is halfway efficient, lots of other people also likely figured this out, which explains why the market reacted to it back in August and gave a “big yawn” to the industry cheerleader’s report.
Pip is right that the industry is big and boring, which is also why there isn’t a great alternative metric for its health. In the big sense, tech is just another industry. The only growth ideas are the occasional iPod, Garmin or Wii stories that apply to smallish companies or categories within tech.
Age 60, life long tekkie, service tech., service mgr., Real Time OS developer, RT Application developer, Semi Apps. Eng., semi sales eng., semi mfr. rep.
While the B2B has always been closely watched, it doesn’t capture the true difficulty – fab capacity, fab upgrade cost, and new fab cost.
The nature of the semi cycle has always been about controlling inventory in a continuous process industry.
A semi company lives on a tight rope. Insufficient production volume and they bleed to death feeding the fab eq. upgrade requirements necessary to stay at the geometry leading edge to enable the lowest cost per bit production. Yet when production hits 95% – full capacity – everything goes on allocation, i.e. all production is build to order. In essence the entire semi industry, or at least critical segments, becomes over a very short time – usually months – a build to order business. While allocated production is great for the capital intensive semi mfr. for the customer it leads to panic ordering.
The semi mfr sees total fab utilization – great from a revenues point of view – but what happens from a customer’s availability point of view is that all insight into the true semi product availability is lost to the consumer. This condition leads to the immediate placement of double, triple, and quadruple booking of orders – a veritable avalanche of order bookings takes place.
In short order nobody has any idea of what “true” demand is – not the consumers, and certainly not the suppliers. This all takes place because of a more basic underlying cycle – the fab upgrade cycle.
These numbers are a bit dated but the concept is still correct. As everyone knows the wafer diameter, projection geometry, and yield are the items that determine the lowest cost supplier of semis. What isn’t quite so widely known is that fabs follow a five to seven year upgrade cycle.
It takes roughly two years to build, equip, and debug a new fab. The cost in year 2000 dollars for a new fab was approximately $2B .
What do you get for your $2B? You get a building that from the outside is singularly unimpressive – what appears to be a small to medium sized office building.
After the fab has been completed it takes 2-5 years to fill the fab with new products, and re-engineered old products. After being filled the mfr must either upgrade an old fab or find another 2-3 billion dollars to create another new fab. There is a significant financial advantage to upgrading an existent facility as approx 1/3 of the cost of any new fab consists of costs associated with facilities – vibration isolation, filtering, gas handling, waste water handling…etc. The net result is that all semi mfrs. would prefer, all things being equal, to upgrade an existing fab rather than build a new one. The difficulty is that that can’t always be done – existent production is a significant consideration when deciding to retrofit a fab facility. The financial benefits however are so compelling to upgrade an existent fab as opposed to building a new one that many companies simply kill old yet very profitable product and retrofit existing facilities.
Why do we care?
Because we are at the end of the production increases that have been obtained by retrofitting all of the old 8 inch fabs with 12 inch fab equipment.
While projection geometries continue to shrink, the time of greatest benefit has also come and passed. We are now faced with very complex projection processes and very esoteric projection “light” sources – in short we have done the lion’s share of the easy profitability enhancements to semi fabs.
Once demand increases a fab into the 90% utilization range we will again be faced with titanic cash requirements to again build new fabs.
How can we know that this demand will arise?
A brief review of some ballpark population numbers will be helpful. The numbers will be divided into two categories: significant users, and non-users of Internet services.
USERS
US – 300M
Japan – 127M
Taiwan – 22M
S Korea – 50M
Australia – 20M
Canada – 22M
EC – 300M
Total 841M
NON-USERS
China – 1,300M
India – 1,200M
Russia – 141M
Africa – 924M
Latin America 566M
Total 4131M
These are numbers from both the CIA Factbook, and the Population Reference Bureau.
Using gross estimates, that 50% of the population of the using countries – 420M – account for 2/3 of the current semi consumption – it’s an estimate. And 10% of the remaining world’s non-using population – 413M – accounts for the remaining 1/3 of the world’s semi consumption then the numbers regarding consumption are startling, and compelling.
If, over the next five years, the non-using population increases it’s consumption by an incremental 10%, then semi production will have to increase by a an incremental 1/3 just to keep up with that demand, no new applications, no change in the consuming segment’s purchases.
What we are currently experiencing is the end of the 8 inch to 12 inch upgrade production cycle. When that cycle is at an end – demand fills those 12 inch fabs – we will have chaos equal to or greater than that experienced in prior semi cycles.
Semis are no longer an esoteric product used only by some. Internet access and net/computer literacy has become the equivalent of being illiterate in our time. Inability to connect or navigate the net will be the same as being unable to read.
A quick visit to some of the net statistics sites will corroborate some of the above numbers. If you don’t like my numbers try a few of your own. Almost any way you parse it out semi demand will continue up and to the right.
Here’s a completely different take than the comments above. Rather than addressing chips, I’d like to address your closing questions:
I wonder if there will be a new dynamic one needs to look at? I wonder if bandwidth is a metric that one needs to track more closely in order to make reasonable guesses about the future. Any thoughts?
I think that if you’re looking for a potential source of shocks to the internet, the most likely candidate is regulation. Already we have CALEA, SOX, various European data retention directives, and a mess of varying laws covering VoIP from outlawing it in the third world to taxing it here in the U.S. to letting it run free in the world’s broadband leaders. Also, regulation can open or close the large markets that Mike Lesko (comments, above) is counting on.
However, another potential source of a shock to the market is natural disasters. A disaster in a key transit point (especially a port) for any key ingredient could harm the internet. A tsunami could hit LA causing a spike in the prices of everything in the U.S., an earthquake in Chile could raise the price of copper, a flood in China could release toxins that are inadequately stored and temporarily stop the flow of key computer components to market, and, of course, anything in the Persian Gulf could cause oil prices to spike.
If you’re trying to find market indicators, predictors of demand in the internet economy, I suspect that the size of annual Federal deficits in the U.S. (or predicted deficits) may correlate with rises and declines in the internet economy but I don’t have this data and I’d love to see a graph of it.
Om,
A most interesting article. I will think about your provocative point. In addition to your specific question about the book to bill ratio, are there several additional questions: (1) what is the relationship between advances in and sales of “hardware” to the the vast amount of “commerce” (I use the terms “hardware” and “commerce” quite loosely to encompass many elements of the value chains) that is enabled by that hardware; and (2)are there certain advances in “hardware” that are so transformational that they will either (a) change the economics of the industry; or (b) make possible significant things that can’t be done today?
These are clearly beyond the scope of your specific article, but I am sure you are thinking about these things too.