May 21, 2010 - Samsung says it will spend nearly $10B in capex just for semiconductor manufacturing, and $B overall -- nearly double its initial plans, and two-thirds higher than in 2009. Analysts tell SST what's significant inside the numbers (foundry), and what it means for the rest of the industry -- and why 2011-2012 might now look a lot different.

Hey, big spender

Overall, Samsung Electronics has pledged to spend an eye-popping 26 trillion won (US ~$22.5B) on several of its businesses (and about 10,000 new employees), as follows:

• Memory: 9T won (~$7.8B). Establishing a new Line 16 for DRAM, NAND, and next-gen memory at its Hwaseong site -- a 200,000 (300mm) wafers/month operation by 2011, the company's first new semiconductor line since Line 15 in 2005. Total phased investment will be 12T won (~$10.4B). Also, increase capacity at Line 15 to boost 30nm DDR3 DRAM to 10% of total production by end of 2010.

• System LSI: 2T won (~$1.8B). For system-on-chip solutions targeting mobile handsets and digital TVs, and to strengthen the company's foundry business.

• Displays: 7.5T won ($6.5B). For a fourth 8th-gen LCD line at the Tangjeong complex in South Chungcheong Province; Phase 2 of Line 8-2 will take an investment of 2.5T won ($2.2B), with 70,000 panels/month capacity. This is in addition to earlier announced LCD investments. Also, joint venture Samsung Mobile Display plans to develop the world's largest AMOLED display factory in the Tangjeong complex, a 5.5G (1300×1500mm) line, online by 2012 with 70,000 panels/month capacity for AMOLED TVs.

• R&D: 8T won (~$6.9B).

Inside the numbers

The key thing to note about Samsung's new semiconductor capex budget is not just the dollar figure -- which, amazingly, equals the current combined budgets of Intel ($4.9B) and TSMC ($4.8B) -- but the comparison to sales. Samsung's 2010 capex/sales ratio will be ~38% (~$9.6B capex vs. $25B sales) -- not too far from its long-term average of around 32%, notes Risto Puhakka, president of VLSI Research.

Bob Johnson, VP of research at Gartner, pointed out that what Samsung is now calling "Line 16" is believed to be its previously announced plans for a Line 17 and Line 18 -- and the expansion for its Line 15 will essentially locate a planned Line 16 on the second floor of the building.

Another interesting number: nearly $1.8B, what Samsung says it will invest in its system LSI business (i.e. its logic foundry operations). "That's very high for them," Johnson said, suggesting the company is also making a big push into that space (it already has publicly been identified as fabbing a chip for Apple's iPad). Exact figures aren't available, it's expected Samsung spend only a few hundred million dollars on that operation in the past. (Ana Hunter, VP foundry at Samsung Semiconductors, recently discussed the company's foundry strategy in a podcast with SST, calling it the company's growth engine going forward.)

Boom for suppliers

Chip equipment suppliers certainly won't complain about the extra business -- it's preferable to the alternative of slowing sales the past few years. Samsung's capex boost and new facility should bring in new business across the chain, noted Risto Puhakka, president of VLSI Research: ASML and AMAT will likely book $1B from Samsung this year, and others will benefit including Hitachi High Tech, KLA-Tencor, Lam, Nikon, Novellus, TEL, and others.

The capex surge also may have a strategic angle. Reports already had surfaced earlier this year that Samsung's efforts to obtain immersion litho equipment were, in part, designed to stretch suppliers' lead times for such tools and make them essentially unavailable to anyone else. Johnson points out that Intel primarily uses Nikon steppers, GlobalFoundries already is doing immersion litho, and TSMC "probably has orders locked in" -- but the other memory firms who would want to keep pace (Elpida, Inotera, Nanya, Micron) might have problems. "At a certain point, lead times depend on getting lenses from Zeiss," noted Bob Johnson, VP of research at Gartner.

Redrawing growth trajectories

Accelerating a cyclical upswing generally results in a sooner downslope as well -- and that's likely to happen this time, too.

Gartner's most recent capex projections (in March) showed 55% overall capital spending in 2010, and 76% for capital equipment (the gap means there's a lot of empty capacity for companies to throw tools into, Johnson notes). Samsung's announced new facility will probably push total capex closer to the equipment number, Johnson said -- but at the same time, "starts to pull things in a little more." Gartner's 2011 capex forecast was 27% (for equipment, 24%), but now that might become "more like mid- to low-teens," he said. It's highly unlikely that Samsung will spend ~$10B in capex in consecutive years, and there isn't anyone who could make up the difference, Johnson notes.

Overcapacity on the horizon?

The DRAM sector is up "tremendously" this year, largely due to improved pricing, Johnson points out. To keep their costs down, PC makers eventually will reduce the memory content per box; soon memory prices will ebb, and DRAM revenues will drop -- but memory firms will still be in the capacity building stage.

"We think [overcapacity] will happen someplace near the end of 2011, into 2012" for DRAM (end of 2012 for NAND), Johnson says.

If, however, memory pricing does not decay but instead remains stable, memory suppliers will have more money to spend -- and they will, if only just to keep up with Samsung, Puhakka said. "Clearly, there is a good chance that there will be over capacity by 2012, especially if other memory suppliers spend at similar rates," he said.

Will memory ever learn?

What will really tip the scales is how other memory firms respond (or don't). With much of the 2010 capex picture targeting process migration (2Xnm for NAND, 3Xnm for DRAM), there's a need for more complex processes (e.g., double patterning litho) that will slow throughput. So the current push in capex may not result in nearly the same surge in overall capacity, "especially since other memory suppliers are spending much less in 2010," Puhakka said.

For those who can't keep up with Samsung's cost structure, some hard decisions are coming. Fellow Korean memory firm Hynix is in the same ballpark with process technology and is pushing DDR3; Micron is a survivor, and isn't far behind in technology, Johnson said (though it "has a handful integrating Numonyx"). In NAND, the only one who could keep up with Samsung would be Toshiba.

On the other hand, "Who's supporting Powerchip and Promos?" Johnson asks. And does Elpida have the capital they need to continue to pursue investments? A few years ago these firms all opened their wallets -- Powerchip ~$2B, Hynix $5B, etc., "spending ridiculously high," Johnson said. "We won't see that repeat itself."

"Semiconductors are resetting back to where they were" before the downturn, he said. And the equipment sector is "resetting back to where they should have been." -- J.M.