Chip makers are packing more circuits onto their chips. The next generation will be dense with circuits just 28 billionths of a meter–28 nanometers–wide.

That will expand, yet again, the power and speed of semiconductors, and increase yet again the number of places where chips can be used. The smaller and more powerful chips will wind up adding smarts to yet more “dumb” devices and making already smart devices from cell phones to refrigerators to cars even smarter.

For the chip industry as a whole that means, once the current recession is over, more chip sales, more chip revenues, and more chip profits.

There’s just one tiny problem. Because the factories that make this generation of chips cost so much–a new 28-nanometer chip factory (called a foundry) now being built in upstate New York will cost $4.2 billion–almost no company can afford to build one.

Intel (INTC) figures that only chip companies with at least $9 billion in revenue will be able to afford the next generation of factories. That’s a very short list: Intel, of course, Samsung, Toshiba, Texas Instruments (TXN), and STMicroelectronics (STM).  Add in chip Taiwan Semiconductor Manufacturing (TSM), a contract foundary that makes chips for companies without factories. That’s the total: 6 companies in the world.

The drop to six, if Intel’s forecast is correct, just continues a decades long winnowing process that has reduced the number of serious players from 14 when circuits were 90 nanometers wide to nine at the current 45 nanometer levels. Only Intel and Samsung have firm plans to build 22 nanometer factories, the next step downward in size from 28 nanometers.

You can expect the remaining players in the semiconductor game of musical chairs to fight like a passel of six-year olds for each remaining seat. That battle will define winners and losers in the chip sector, of course, but it will also shake up sectors as diverse as solar energy and automobiles.

The number of chip companies with their own chip factories has ben shrinking for decades as companies such as Broadcom (BRCM) developed the fab-less model. The theory was that these companies could make a lot better return on their capital if they didn’t have to sink $1 billion or $2 billion or more into building a factory that would manufacture chips. Instead these companies would concentrate on design and marketing and other higher-margin activities, and leave the manufacturing to others such as Taiwan Semiconductor, Chartered Semiconductor (CHRT), or United Microelectronics (UMC).

Turns out that theory was true if you compared a very good fabless chip company to a middling chip company that tried to do it all. For example, Broadcom earned an annual 8.2% return on capital on average over the last five, not-exactly-so-great, years. Advanced Micro Devices (AMD), a company that tried to do it all–and head-to-head with Intel no less, shows an average annual return on capital of a negative 15.4% during those years.

But it turns out that chip making was a very profitable business too–if you were good at it and if you could keep your factory full. Intel, which used its factories to bludgeon Advanced Micro into submission, earned 16.1% a year on average on its capital during that five years. Taiwan Semiconductor, which gradually emerged as the clear No. 1 among contract foundries did even better with a 21.6% return on equity.

If a manufacturer couldn’t fill its factories–or at least couldn’t fill them with the best paying contracts–though, it wouldn’t make much money. Average annual return on capital at United Microelectronics was just 4.2% and at Chartered Semiconductor is was just 0.5% during the period.

It was clear enough what you had to do as a chip manufacturer to prosper as plants got more expansive and the number of competitors fell: you had to fill your factories.

Now it’s not a matter of filling your factories to prosper–it’s fill your factories or die.

That’s why we’re seeing such huge increases in capacity utilization from chip manufacturers right now. United Microelectronics, which reported that its factories were running at just 30% of capacity in April, now says that its foundries ran at 70% of capacity in the second quarter and projects that capacity utilization will hit 90% in the third quarter.

Since we know that PC sales are still falling and that no one is showing a big turn around in other industries that are heavy users of chips, such as cell phones and autos, you’ve got to wonder what’s going on. A good guess is that the company, which has lost money for the last thre quarters, is taking on any kind of work it can now to fill its factories. Later, the company will worry about what to do if customers, sitting on big inventories of chips, stop ordering. If it doesn’t fill it’s factories now, there might not be a later to worry about.

Or at least the later will look very grim from a competitive point of view. Taiwan Semiconductor, which already owns 44% of the contract manufacturing market, is set to increase its lead. The company,he only contract manufacturer to remain profitable during the downturn, announced on June 18 that it will increase capital spending to $1.9 billion, returning that budget to the 2008 level.

So what does all this mean to you as an investor?

First, I think we’re going to see one more, very short boom and bust in the sales cycle that will damage the sector’s weaker players even further. I’d expect another shakeout when inventories begin to build since production looks like it’s running ahead of demand at the contract foundries.

Second, I think that, just as it has during earlier stages in the contraction of this sector, more of the profits will go to the biggest and most efficient players. The stock I’d watch here is Taiwan Semiconductor, which recently signed its first contract to make chips for Intel. For Intel!

And third, chip makers desperate to fill their foundries will aggressively expand into new but related business. Autos certainly where the silicon content of cars will keep climbing.

But the most enticing, if early news out of Taiwan Semiconductor, is manufacturing solar cells. Hey, it’s all silicon, no? As that move gathers strength, the most efficient chip makers are going to put pressure on the least efficient solar cell manufacturers. Solar companies are going to have to learn how to manufacture really, really efficiently. The ones that don’t aren’t oging to be bailed out by their technology and subsidies.

It should make for interesting times.