Get ready for the big squeeze

For the chipmakers, the heady days of 1995 seem so far away. That year, they were on a roll and it seemed nothing could go wrong. The Internet had suddenly mutated from a research curiosity into the beginnings of a technology revolution. In August, Netscape, a company founded a little over a year earlier, stunned the world with its runaway initial public offering (IPO): shares, launched onto the market at $28, more than doubled in value to $75 - close to a record for a stock's first-day gain.

The semiconductor business basked in the reflected glory. After all, you could not run the Internet without their wares. Without silicon, the information superhighway wasn't even a dirt road for data. The problem was that the chip business wasn't quite ready for the rapid growth. The 1991 slowdown was not the worst the industry had ever seen - it was, at least, not a recession like the self-inflicted catastrophe of 1985 - but it convinced many companies that they spent too much on putting fab capacity in place. They were not going to make that mistake again: they kept capital expenditure on a tight leash all the way through 1994.

It worked well for the chipmakers. The average selling price (ASP) for silicon chips during the first half of 1980s stayed stubbornly around the $1 mark. By the end of the decade, the ASP had risen to around $1.50. Then the worldwide recession hit, dragging the chip business down with it. In a market that expects the same function to be cheaper year-on-year, it is hard enough to keep prices up. When the customer knows you are drowning in product, they do not show much mercy: the price plummets, taking the industry's ASP with it (see 'Fill that fab', over page). According to the figures provided by the World Semiconductor Trade Statistics group, the ASP slid back to $1.25 by the end of 1990. As the world economy recovered, so did prices. The chipmakers did not rush to add extra capacity for the growth in unit shipments that came with the recovery. Whereas the fab owners had spent around 18 per cent of sales on new production equipment in the years running up to 1990, they slashed capital expenditure and kept cutting all the way to the beginning of 1994. The average fell well below 15 per cent in an industry that was, up to that point, spending more than 20 per cent of sales. A few percentage points either way does not sound a lot, but it makes a massive difference to how much fab capacity gets installed each year.

Spotting the surge

It was not until the beginning of 1994 that the chipmakers realised that demand was surging. Even lagging process technologies, with geometries bigger than 0.7µm - the leading edge was then around 0.4-0.5µm - were seeing fab utilisation levels rise beyond 80 per cent and head towards 90 per cent. By the end of the year, the fabs were effectively sold out: anybody not running at greater than 95 per cent utilisation just wasn't trying.

With computer suppliers eager to get computers and modems into the hands of consumers as they got online prices naturally soared. The $1.50 level was a distant memory. The ASP pushed through $2. And on it went. By the end of 1995, the ASP had hit a peak of $3.25.

Chipmakers were desperate to get their hands on wafers - desperate enough to gazump the competition. Some found that wafers they had ordered from their foundry had been shipped to a competitor willing to pay more. Banks told fabless clients - those companies buying silicon from third-party foundries - that they needed their own fabs to guarantee a source of supply. Reluctantly, companies such as Cirrus Logic and Xilinx went into joint ventures with other chipmakers. In that environment, owning a 10 per cent stake in a fab was better than owning nothing at all, even though it was not clear how such plants would operate with so many bosses.

And then the music stopped.

In what is a familiar story in the semiconductor business, the box builders found they had themselves overdone it. In order to ensure that they got some parts, they had double, sometimes triple-ordered. To their horror, all of the orders turned up just as it became clear that the Internet economy was going to grow a bit more slowly than the bubble-maniacs of 1995 had predicted.

Having been pulled to the summit of December 1995, the ASP was pushed over the edge onto the scariest part of its rollercoaster ride. Forget $3.25. By 1997, the level had slumped to a little over $2. And it was not going to stop there. Although a mini-boom in 1999 and the larger bull market of 2000 dragged the ASP up, it was not to last long. Since 2001, the ASP bounced around on a downward curve, back to below $1.50 - just as it was at the beginning of the 1990s.

Despite six consecutive years of growth in unit shipments, if not the dollar value of the total semiconductor market, ASP has stubbornly refused to grow much at all, other than a brief surge in 2004, with all its gains cancelled by ballooning inventory in 2005. Yet some analysts say 2008 may be different. It may be the time that prices, once again, start to head north.

If 2008 does see growth, it will be the fourth consecutive year for the total value of the semiconductor market to increase. Malcolm Penn, president of Future Horizons. notes that the last time that happened was in the period leading up to the 1995 boom.

As with the long march up the price curve in the early 1990s, tightening capacity is the key. In principle, all but one of the factors needed to ensure at least a moderately good year are in place. Jim Feldhan, president of Semico, said the chipmakers are starting the year with good fundamentals in place.

"We are entering 2008 with inventories under control and prices in some categories are on the increase," he says.

Low inventories

Aida Jebens, senior economist with VLSI Research agrees, saying that inventories are currently at the lower end of their normal range. Inventory overhangs tend to lead to rapid declines in pricing - the cycle normally leads to sudden drops in fab utilisation which encourages vendors to reduce prices in attempts to foster demand.

With inventories under control, fab utilisation is high. Jebens claims: "The utilisation rate peaked above 95 per cent in October/November last year. It is now at a seasonal low but still above 90 per cent. That is being driven by double-digit unit shipment growth."

Penn agrees: "Units have outgrown capacity."

Jebens points to how the earnings made by chipmakers per square inch of silicon shipped have, since the end of 2004, wandered around the $35 mark - above $30, the chipmakers are generally in profitable territory. But the spending on equipment, which has hugged the $10 per square inch line for the last couple of years, she forecasts to slide to less than $9 by the end of this year.

It is having a knock-on effect on the fab-equipment suppliers. Stanley Myers, president of SEMI, says the organisation also forecasts a fall in equipment spending during 2008: "We think the decline could be 10 per cent or a little greater."

In practice, the spending cuts will not have a big effect on chip prices in 2008: the big squeeze on capacity has yet to come. Spending during 2007 was fairly robust: the announcements of big cuts did not come until late last year. "We would like to have seen capital expenditure for 2007 a little bit lower," says Penn.

He says the problem is that capital spending now does not have any bearing on the chip market for close to a year. Anything bought now will not be producing chips in any volume until early next year. 2008's budget cuts mean capacity could get very tight in 2009, but probably not before then, unless unit shipments grow quickly.

Both Future Horizons and IC Insights reckon that 2008 will see around 12 per cent growth overall, lifting the semiconductor market to around $290m, based on unit shipments increasing 10 per cent over 2007 and prices edging up by just 2 per cent.

"Capital spending as a percentage of sales is coming down," says Bill McClean, president of IC Insights, claiming that this will lead to higher market growth than has been seen since the boom of 2000. Chipmakers are cutting capital expenditure, partly as a result of their move to fab-light programmes, where much of their production, particularly for capital-intensive leading-edge processes, is outsourced to foundries.

Penn says he is not so sure about long-term spending coming down, pointing to a trend line from the late 1980s that is "as flat as a pancake": a steady 16.75 per cent of sales. However, most agree that, if the chipmakers hold to their projected spending, the increase in unit shipments will gradually use up the industry's spare capacity towards the end of the year and into 2009. 

Feldhan points out: "Manufacturing is in a good state to stabilise average selling prices. And we think capital expenditure will decline 8 per cent in 2008. That will create a reasonable state for 2009."

McClean claims the picture that is emerging is of a long-term situation of tight supply that will force the average selling price up.

"Companies are really changing their outlook on capital spending. It is setting up a collision course in the industry. They are talking about having low capital spending as a percentage of sales. A lot of companies are going to the fab-light model but that is running right into the pure-play foundries getting conservative on capital spending themselves. People are expecting the foundries to come big on capacity at the same time that they are cutting back on capacity spending.

"Look at TSMC: why are they talking about spending 20 per cent of sales? Because their revenue per wafer keeps coming down. Three years ago, the revenue was $1,500. Last year it was $1,200. They don't want to play in the $1,000-wafer marketplace. And, when TSMC announces they will spend 20 per cent of revenues on capacity, because they are a foundry, that really means 10 per cent of sales," McClean adds, noting that, because foundries do not supply chips directly to end users, they do not see the same margin that the integrated device manufacturers obtain.

"What is going to give, we believe is the IC average selling price. The IC manufacturers are serious this time.

"We believe there will be good unit volume growth and better average selling prices, that will lead to better IC market growth. But it might be at the expense of the capital equipment guys in the short term," says McClean.

Dean Freeman, vice president of reserach at Gartner, says: "2008 will be a mixed year. It's a year that should be strong but will, in fact, be weaker than we would like."

But, if the chipmakers hold their nerve, come 2009, they could finally arrange a turnaround in chip pricing by having the ability to say no to their customers. The question is: will they hold their nerve?