Memory Drives the Market...Again

The industry was recently surprised by a sudden drop in fab utilization rates, which indicated production cutbacks in response to growing inventories. However, it now looks as though the industry responded quickly, meaning the production slowdown will be relatively minor and should not have major effects of investments in new capacity. While overall utilizations dropped again in 4Q06, leading-edge facilities remained comfortable at ~90%. We expect overall utilization rate to level out at the 85% range in the first half of 2007, and then begin a slow upward climb. Leading-edge utilization rates will stay in the low 90% range — the level that encourages continued investment in new facilities. These signs all point to a 2007 for the equipment markets, which is flat in the annual comparison, but not significantly down from 2006. The quarterly picture is a different matter, however.

The big uncertainty is what will happen in the memory markets. Or, to put it another way, when will all its aggressive investment in new facilities finally catch up with skyrocketing unit demand? The answer is that it will probably happen sooner than we would like. If it happens, is there enough strength in investments from other segments to come to the rescue of our industry?

Putting things into perspective, if we look at the two hottest segments of the memory markets, DRAM and NAND flash, we see a picture emerging for 2007 that is vastly different from 2006. In DRAM, 2006 saw revenue growth of 36% driven by unit growth of ~18%; in 2007, the reverse will be true. Our latest forecast shows DRAM revenue up 10.5% and units up 38% for 2007. A slightly different pattern holds for NAND flash: 2006 revenue is up 7% and units up 47%. In 2007, we anticipate similar unit growth of 49%, with a slight revenue decline of -1%. Supply has clearly caught up with demand, and unit growth is driven only by plummeting ASPs.

In terms of capex, however, our latest projections for memory investment show a similar picture to 2006. Last year, memory spending accounted for 48% of all capex, or slightly more than $27B. In 2007, we expect similar numbers, with memory spending at almost half of total capital investment. The key question is why can the memory market, which typically accounts for around a quarter of all semiconductor revenue, drive half the capital investment?

The answer lies in the soaring unit demand for both DRAM and NAND flash chips, which can only be met by heavy investment in additional manufacturing capacity. But this is only half the picture; since memory markets are extremely competitive commodity markets, market share and market leadership goes to the company that can expand capacity to meet rising unit demand. In addition, since overall market supply and demand characteristics determine average market selling prices, the competitor who can keep his costs going down ahead of market averages will enjoy superior profitability, and can better afford continued investment. This means that in a market that is supply-limited and facing skyrocketing demand, there is a lot to be gained by being there first with the most — meaning investing heavily in new capacity. Which is exactly what's happening in the memory markets today.

At some point, however, it no longer makes sense to continue the breakneck pace of investment in new capacity if this will only drive prices down to the point where profitability is marginal. The big question facing us today is when this will happen in the memory markets.

Historically, the last time that memory investment was close to half of total capex was in 1995–96. During that period, DRAM ASPs declined more than 60% from 4Q95 to 4Q96, and memory investment as a percentage of total capex dropped dramatically in 1997.

We are already seeing signs that the supply/demand balance is tipping more toward the supply side. In both DRAM and NAND flash, we expect to see significantly faster price erosion than last year. This compares to the harsh price erosion of 1996 when DRAM unit demand rose only 5.4%. For 2007, we are predicting a 38% increase in DRAM unit demand in 2007. So the conditions are different.

We predict that memory spending as a percentage of capex will be high in 2007, and then decline to more "normal" historical levels starting in 2008. The quarterly investment picture will be front-end loaded in 2007, with declines in the second half of 2007. DRAM capex could still expand by a few percentage points based on carryover momentum from last year, but we see NAND flash investments down to 10%. We do not foresee that much if any capex is moved from DRAM to NAND, given the dire profit margin situation in the latter market and another year of steep price declines.

Overall, we expect total capital spending to be flat this year and return to growth in 2008. There is hope that investments in other device segments, logic and particularly foundry, will offset any weakness in memory. This is all based on the assumption that unit demand for DRAM and NAND flash chips will stay strong. With the amount of new memory capacity already coming online, any significant slowdown will slow capex spending.