(FORTUNE Magazine) – Only yesterday, it seems, a choir of journalists, college professors, and management consultants was telling how the Japanese, through their superior manufacturing techniques, would subjugate the world economy by the start of the new millennium. Now, in the biggest flip-flop in recent industrial history, the Japanese--Mitsubishi, NEC, Fujitsu, and soon Sony--are quietly selling some of their treasured U.S. factories to the supposedly backward Americans, who will make Japanese products in them.

Japanese turning to Americans to manufacture Japanese products? Have Eskimos suddenly forgotten how to make igloos? Are McDonald's owners selling off their outlets because they no longer know how to flip hamburgers? One answer is that after seven years of recession and poor stock performance, the Japanese have finally begun to give up their cherished dream of keeping everything in the family. They can no longer ignore outsourcing--and outright sale of plants--as a better financial model.

There's a lot more to it. Japan's rigidity has finally caught up with it in manufacturing, at least in electronics industries that require flexible production and fast product introduction to accommodate rapidly shifting demand. An executive of a Silicon Valley diskmaker boasts that, when it comes to speed in turning out a wide variety of complex products, such as computer servers and hard drives, and changing the product mix almost daily, "we can turn on a dime. The Japanese can't do that."

Don't get the wrong idea. Sony, Panasonic, and other Japanese giants still excel at cranking out high-quality consumer electronics products--such as camcorders and TVs--by the millions. Some American manufacturers, such as Xerox, continue to feel pressure from Japanese competitors, particularly in midrange and low-end copiers and other machines. And Japanese auto manufacturers still make outstanding vehicles.

But it's a different story in industries with short product cycles and factories that must build what customers order instead of churning out products in anticipation of demand. Here Japan's great strength--repetitive manufacturing--is becoming its greatest weakness.

Listen to John Costanza, president of the John Costanza Institute of Technology in Englewood, Colo., and one of the foremost experts on manufacturing: "The Japanese are terrible at building on demand--terrible, terrible, terrible. Repetitive manufacturing works fine if you can sell everything you can build. But the day when they could tell a customer, 'I'll make a product and you'll buy it,' is gone. Building variable products to demand is the requirement for manufacturing today. So now the Japanese are saying, 'We're going to contract manufacturing by selling plants because we don't know how to change our manufacturing culture.' "

This is where a new breed of American supermanufacturers--you could call them the new Japanese--comes in. These are the so-called contract electronics manufacturers (CEMs), which are gobbling up those Japanese plants in the U.S. as well as abroad. A far cry from the grimy job shops found on the bleak back streets of suburban Detroit or Chicago, CEMs use production techniques that top those of the Japanese, in huge, surgically clean, highly automated plants employing thousands of workers. Surprisingly, some mammoth new CEM plants are located right in Silicon Valley, on some of the most expensive industrial land on earth. The CEMs now make products cheaper in San Jose than the Japanese can in Tokyo.

Through supply chains linking dozens of plants around the world, the big CEMs have become the new providers to so-called original equipment manufacturers (OEMs). Though OEM brand names appear on a vast array of products--from computer servers to PCs to cell phones--a growing proportion of these wares emanates from CEM plants. The term OEM, in fact, has become dated; today these companies could more accurately be called OBHs (original brand holders).

According to a recent report by the BancBoston Robertson Stephens investment firm, the largest brand owners using CEMs include Hewlett-Packard, which relies on ten different CEMs to make its products, Cisco (with nine CEMs), IBM (eight), Lucent (seven), and EMC, Ericsson, Motorola, and Nortel, with six each. Technology Forecasters of Alameda, Calif., estimates that 9.5% of the electronic goods sold by the world's OEMs are now put together in CEM plants, and expects this percentage to reach 17% by 2003. Japanese brand owners are just starting to get on the bandwagon.

Thus, the CEMs are the new tailors, shoemakers, butlers, and chambermaids of the brand-holder elite, whose members concentrate increasingly on R&D, product design, and marketing. The biggest contract manufacturers are Solectron of Milpitas, Calif., with projected sales this year of $13 billion, SCI Systems of Huntsville, Ala. ($8 billion), Celestica of Toronto ($6 billion), Flextronics International of San Jose ($3 billion), and Jabil Circuit of St. Petersburg ($2.8 billion). Most of these companies provide other services, ranging from product design and testing to the management of the supply chain and even repair of branded equipment in the field.

CEMs are one of the fastest-growing industrial segments. Technology Forecasters predicts that their revenues, already $60 billion a year, will hit $150 billion in 2003. The big CEMs are growing at more than 50% a year, a remarkable figure considering their size. Just how good the CEMs are is indicated by the fact that Solectron is the only company that has twice won the Baldrige award for manufacturing excellence. And CEMs excel at running global supply chains and using the Internet.

As the CEMs' stock market performance shows, they are not engaged in a marginal, low-profit activity. Since Solectron went public in 1989, its stock has soared 16,000%, vs. a 7,000% rise for Microsoft and 4,000% for Intel during the same period.

The CEMs' profit margins are not as impressive as those of the brand owners, averaging 4% of sales. But their return on investment, as BancBoston Robertson Stephens puts it, has been "splendid." ROI has averaged more than 20% for the big CEMs, vs. an average of only 6% to 9% for companies that make up Standard & Poor's industrials. The reason is that the CEMs' plants, operating at a high percentage of capacity, are efficient.

The growth of the CEMs has been fueled in large part by their purchases of OEM plants, for which they often bid aggressively against one another. They have turned those plants into more efficient producers by upgrading equipment and by making products for more than one client under the same roof. In 1998 alone, CEMs bought 47 manufacturing plants in the U.S., in the process keeping thousands of manufacturing jobs from leaving the country.

Abroad, CEMs run dozens of plants in places like Oulu, Finland; Hortolandia, Brazil; Guadalajara, Mexico; and Kunshan, China. Generally the CEMs make more complex products in their U.S. plants than they do at foreign sites where labor costs are lower. Because of stunning productivity improvements in U.S. manufacturing, it should be noted, labor costs have fallen to low single-digit percentages of total production costs, below the figure in some Japanese industries.

A big goal of many CEM executives, helped to no small extent by John Costanza's institute, has been to return U.S. manufacturing to its former preeminence. Many who have led this drive are Asian Americans, among them Solectron's former president, Winston Chen, a Chinese American, and its current CEO, Koichi "Ko" Nishimura, a California-born Nisei who spent his boyhood in a World War II internment camp.

Buying Japanese plants in the U.S. could put the CEMs on a new growth spurt. "There's a certain irony in this," says Flextronics CEO Michael Marks. The Japanese plant sales began in 1998, when Mitsubishi sold its cell-phone manufacturing facility in Braselton, Ga., to Solectron. In addition to taking over the manufacture of Mitsubishi cell phones sold in North America--the phones are being made in a new 100,000-square-foot leased facility that it has set up on the Mitsubishi campus--Solectron also manages printed circuitboard assembly for the phones and new-product introduction. Mitsubishi, for its part, now concentrates on advanced engineering and product development.

Solectron declined to be interviewed for this article, perhaps because of the sensitivity of the subject with both buyers and sellers. But CEO Nishimura is known to be actively encouraging Japanese OEMs to sell their plants. "He tells them how Solectron would modernize the plants, reduce their costs, and improve their antiquated distribution system," says a source close to the company. The fact that Nishimura speaks fluent Japanese doesn't hurt.

The changes CEMs make in newly acquired plants, Japanese or otherwise, can be dramatic. Says Flextronics' Marks: "All factories are different. But in nearly every case, we improve the housekeeping and change some information technology functions. We almost always invest in new production equipment."

The CEMs pick their acquisitions carefully, Marks notes: "We buy plants with appropriate geographic locations, to increase customer penetration and to give us technical capabilities we don't have." Flextronics passed on NEC's Hillsboro, Ore., telecom equipment plant, which is on the block, but is bidding on another NEC telecom production facility in Sao Paulo. A few weeks ago, Flextronics also acquired a computer server factory in Paderborn, Germany, which had been jointly owned by Fujitsu and Siemens.

Late last year another big CEM, SCI Systems, bought an NEC Computers manufacturing plant in Sacramento and shipped some of the production tools to Huntsville, Ala., where it has set up a plant that makes laptops and desktops for NEC. SCI will also handle the complete supply chain for NEC Computers in North America. SCI CEO A. Eugene Sapp hailed the acquisition as providing "the basis for a range of future initiatives between the [two] companies."

What really makes CEM bosses drool is a tsunami of Japanese plant sales expected at the end of March. That's when, according to a spokesman, Sony will announce that it plans to dispose of 22 of its 70 plants worldwide, selling some and shutting down others. The CEMs will be there to bid. They know they can always fix up Japanese plants, even if they don't meet CEM standards. And plant capacity is what the CEMs are hungry for, given their expected rate of growth.

Many other Japanese manufacturing companies are expected to follow suit after Sony's announcement. "They will be saying, 'if Sony's doing it, we should be doing it too,' " says Sheridan Tatsuno, a Harvard-educated Sansei (fourth-generation Japanese American) who consults with Japanese companies from his Northern California base in Aptos. The trend is about to accelerate because of pressure from an unexpected source. Says Tatsuno: "Even MITI [Japan's Ministry of International Trade and Industry] is now telling Japanese companies to get into e-business and to sell manufacturing plants." Yes, that's the same MITI that helped engineer Japan's assault on U.S. manufacturing companies in the 1960s and 1970s.

The sale of Japanese plants is "gathering momentum," says CEO Eugene Polistuk of Celestica. Eventually, he predicts, the selling wave will hit electronics plants in Japan itself. Because of confidentiality agreements, Polistuk says, he can't disclose whether his company is bidding on any Japanese plants. Selling the family jewels--Tatsuno's term for the plants--is a traumatic experience for older Japanese executives, many of whom made their names in manufacturing. "It's a huge loss of face," says Tatsuno, "because they literally grew up on the plant floor. But the younger executives don't seem to care who makes their products."

Handel Jones, a Ph.D. economist and semiconductor engineer who runs a Japan-oriented consulting and data-gathering firm, International Business Strategies in Los Gatos, Calif., sees nothing but black clouds over the factories of the Land of the Rising Sun. "When we project Japanese strengths and weaknesses out to 2010," says Jones, "it looks very negative for them. They remain strong in the old-style electromechanical, or repetitive, manufacturing. But when it comes to using software intelligence inside new, flexible production systems, they're very weak."

Tatsuno goes so far as to say that "Japanese manufacturing will sink in ten years--90% of it will be controlled by foreigners." That seems unlikely. But Americans have always overrated Japanese manufacturing skills. Many Americans--especially those troubadours who sang of Japanese production prowess--thought that repetitive manufacturing was the be all and end all. The emergence of the much more recent on-demand production technology changes everything.

Says Flextronics' Michael Marks: "The concept of the Japanese as great manufacturers is a very outdated view, based on a time when they were able to borrow funds at 0% and often automated their factories beyond what is economically feasible. In addition, the reputation comes from automobile manufacturing and some high-volume consumer electronics manufacturing. I have not been impressed with the factories I've seen in Japan over the years."

During the past two decades, while the Japanese stuck to their repetitive electromechanical model, Americans invested heavily in new computerized manufacturing techniques. Says George Heilmeier, until recently the CEO of Telcordia Technologies (formerly Bellcore): "The Japanese haven't bought into the concept of information systems in the factory the way some U.S. companies have, nor are they prepared for the integration of the enterprise through the intelligent use of information systems."

For a firsthand look at what Heilmeier is talking about, sit next to George J. Hall, executive vice president of HMT Technology in Fremont, Calif., near Silicon Valley. HMT is a CEM that produces the disks used in PCs and other computers. Making the disks involves complicated chemistry and physics in addition to mechanical processes. At the click of a mouse, Hall displays on his PC screen the status of each production machine on the factory floor below. He or a production supervisor can immediately spot any problem arising at any machine and send in a crew to correct it then and there.

What makes such instantaneous production tracking possible is manufacturing execution systems (MES) software from Camstar Systems in nearby Campbell, Calif.--the type of software that hasn't caught on in Japan. The new software also allows HMT to make hundreds of different types of disks on the same day. The Japanese can't move that fast. Result: HMT makes disks for $6 apiece while it costs the Japanese $8 to make them in Japan. The "H" in the company name, incidentally, stands for Hitachi, yet another Japanese company that had to sell out to an American CEM because it couldn't hack it in the U.S.

The Japanese have no choice but to embrace change, and they may well do so. But some experts on flexible manufacturing say the adjustment won't be easy. When the need for the electronic speed of Internet decision-making clashes with snail-like Japanese consensus deliberations, these observers contend, the consensus culture is hard to dislodge.

"The Japanese just can't make decisions based on incomplete information, which the new Internet economy often demands," says manufacturing expert Richard Morley, who consults with Japanese companies. It will take some time before the Japanese start making business decisions "by a few e-mails" instead of via ceaseless consensus-building meetings, says Japanese executive Keisuke Yawata. Known as "K.K." to his American friends, the U.S.-educated Yawata has served as a top executive in both Japanese and U.S. companies--NEC and LSI Logic--and is now trying to help Japanese entrepreneurs build startups.

Recognizing their new needs, some Japanese companies are now seeking help from American manufacturing experts such as John Costanza. Fujitsu, Sharp, and Hitachi are trying to introduce demand-flow manufacturing, a concept Costanza pioneered, in their U. S. plants. He and his staff have installed it in hundreds of U.S. companies, from AT&T to GE to U.S. Robotics, winning high praise from Jack Welch and other executives.

Demand-flow manufacturing works on principles diametrically opposed to those of conventional manufacturing based on scheduling and forecasts. Not only does demand flow rearrange linear production lines into semicircular cells for more efficient production, but it also does away with that mainstay of Japanese manufacturing, the just-in-time (JIT) delivery of components such as automobile seats or tires. Demand flow substitutes its own concept--raw-in-process inventory, or RIP. This calls for keeping a reasonable quantity of varied raw materials or components on hand to meet changing customer demand.

Just-in-time can get a manufacturer in trouble, says Costanza. Computer buyers can suddenly switch to buying PCs equipped with 500-megahertz microprocessors instead of 400-megahertz ones. Unless the PC maker has both types, he'll have to shut down production. RIP sees to it that different raw materials and components are on hand; JIT does not.

Sheridan Tatsuno sums it up: "John Costanza and the CEMs have totally changed the game on the Japanese." Americans teaching the Japanese that it's time for just-in-time to rest in peace? That's how far the world has turned in manufacturing.

Stories from FORTUNE's Industrial Management & Technology section may be found on www.fortune.com.