How will the semiconductor market develop in 2011? Will we continue to enjoy the good times that we experienced in 2010, or will it be a down year? That’s a big question.
We are now at the end of 2010, which was a banner year for semiconductor growth despite the difficulties suffered by the rest of the world economy. Memories, the more erratic segment of the semiconductor market, are showing a 12/12 growth trend of 73%.
(This 12/12 measure is the most recent 12 months compared against the same 12 months one year earlier. In the absence of a full year’s worth of data, this is a reasonable substitute for year-over-year growth.)
Since the rest of the semiconductor market makes more modest moves than memories’ wild gyrations, semiconductors overall are only seeing a 12/12 growth trend of 37%. Still, this is very good news in a market whose average annual growth is 18%.
This did not come as any surprise. A year ago, we predicted memory growth that could reach or exceed 50%, driving the semiconductor market to growth of 30% or more. This was an extremely contrary opinion compared to other forecasts being released at that time.
Semiconductor consumption is growing at a healthy pace even if unemployment is high. It’s difficult to explain, but oddly enough semiconductor consumption rarely changes significantly over time.
Over the last 40 years, DRAM gigabyte consumption has undergone a significant quarterly decrease only four times due to softness in the economy: 2009, 2001 (the Internet bubble burst), 1986, and 1974. Such demand declines are very rare.
Semiconductor Boom & Bust Cycles
So if demand grows steadily, what makes the semiconductor market go through such strong ups and downs? It all comes down to a question of demand and supply. When demand meets or exceeds supply, prices stabilize, or in extremely rare cases, prices increase. When supply exceeds demand, prices collapse.
A price collapse is quite easy to understand. The largest cost component of most semiconductors is the depreciation of the capital equipment used in a wafer fab. Whether a company sells one chip or a million, its capital depreciation is always the same. Greater volumes help spread that depreciation around, effectively lowering costs.
Let’s say that a manufacturer adds a new plant designed to produce a million chips a month. Once the commitment is made, a mad dash is on to ramp the plant to the point where it actually produces those million chips. This usually takes about two years.
Once that plant starts to ramp and the equipment starts to be depreciated, more chips flow into the market. If the number of chips increases faster than demand, the manufacturer is forced to take market share away from its competition, usually by lowering the price.
The competitor responds either by trying to win the business back based upon an even lower price or by going after other customers, pushing the original manufacturer out by lowering the prices there. Each competitor responds by lowering its price—first one, then another at a very rapid pace—until the chips are selling at cost.
The game usually ends there, since foreign suppliers who sell below cost become liable for government trade sanctions. During the decline, it’s not at all unusual for memory prices to drop by 60% in a single year.
Now this is bad enough when only one company adds production capacity. But in the world of semiconductors, all manufacturers add capacity simultaneously, causing a very sudden and dramatic shift from an undercapacity to a significant overcapacity.
There are four phases in the semiconductor cycle. During a shortage, prices stabilize and manufacturers become profitable. They invest these profits in wafer fabs, avoiding steep taxes on retained earnings. Invariably, this level of investment is too high and results in an oversupply two years after the company’s original commitment to add capacity. This oversupply drives prices into a collapse, evaporating profits.
As long as manufacturers are unprofitable, they can’t expand production to meet the needs of a steadily growing market. This creates a shortage a couple of years later, and the market enters another period of stable prices and profits.
Forecasting The Future
If we understand the cycle, and we know how capital spending has played out, then it’s not a big step to produce a relatively accurate forecast. All it takes is a solid understanding of the price and cost dynamics of the semiconductor market.
Memory capital spending jumped in two phases last year: DRAM spending increased at the middle of the year, and NAND capital spending kicked into high gear one quarter later. This implies that DRAM is likely to undergo a price collapse in the middle of 2011 (two years later), and NAND should follow by a quarter.
When this occurs, DRAM and then NAND prices likely will drop to cost in a quarter or two. The year should start out well enough (despite today’s softness in PCs and mobile handsets) but will totally unravel in the second half.
Since the year will start out strong and then collapse in the second half, 2011 memory revenues will experience a relatively mild decline of 20%. This will drag down semiconductor market growth to the range of 5%.
Overall 2011 memory revenues should reach about $56 billion, down from this year’s expected $70 billion, with total semiconductor revenues rising from 2010’s $295 billion to a record-breaking level of slightly more than $310 billion.