Dan's Take

IBM Sells Its Chip Business

• By Dan Kusnetzky
• 10/20/2014

Two Successful Models
Designing and making semiconductors is an expensive business. In the past there have been two successful business models, but this move might indicate one of the two no longer works. The two different approaches:

1. Create a lowest-common-denominator product that can be produced and sold for a profit in high volume at a relatively low cost.
2. Create a specialized, high-performance, feature-rich product that can be sold in low volume at a relatively high price and still be profitable.

If you consider those two business models, it's obvious that the Intel Corp. and AMD Inc. x86 businesses are an example of the first strategy, and the IBM Power and Z microprocessors are an example of the second.

Chip Manufacturing Models Are Fragmenting
As with other segments of the IT market, you're beginning to see a fragmentation of the business. It used to be common for both the chip design and manufacturing to be done by a single company. Now, successful microprocessor families are designed by one company, then manufactured by many competitors. ARM Ltd. is a good example of this.

ARM designs the microprocessor and a number of suppliers manufacture the chips. ARM products can be found in smartphones, tablets, cameras and even electronic control modules found inside automobiles and aircraft. ARM offers 32-bit and 64-bit designs today and is likely to offer other, more complex products in the future.

ARM claims more than 1,000 partners who manufacture microprocessors for specific applications. If you glance around the Internet a bit, you can see that this list includes suppliers such as Apple Inc., Applied Micro Circuits Corp., Atmel Corp., Broadcom Corp., Freescale Semiconductor Inc., Nvidia Corp., NXP Semiconductors, Qualcomm Technologies Inc., Samsung Electronics Co. Ltd., STMicroelectronics and Texas Instruments Inc.

It appears that IBM wants to follow this path, rather than following the lead of companies such as Intel and AMD.

Why Would IBM Sell Its Chip Business?
Manufacturing microprocessors is an extremely expensive business. While attempting to follow Moore's law and constantly reduce the size of transistors on the chips, more and more exotic techniques must be deployed. A chip foundry might cost between $1.5 billion and $2.5 billion and only be able to produce one or two generations of microprocessors. At that point, the technology used in that plant would become obsolete. Then another plant would have to be built to support the creation of the following generations of chips using the newest imaging technology.

High-volume manufacturers have been able to sustain these costs and still create profitable products. Manufacturers of lower-volume products, such as IBM, Sun Microsystems Inc., DEC and Hewlett-Packard Co. have struggled to make the chip manufacturing business profitable. Often, these companies would combine the financial results of the profitable systems manufacturing business with the sputtering microprocessor manufacturing business to present a glowing financial report for stockholders and analysts.

As stockholders and analysts expect each and every portion of a business to be profitable, it got much more difficult for manufacturers such as IBM to hide the true costs of creating one chip foundry after another.

Dan's Take: No More Chip on IBM's Shoulder
Does IBM's move mean that the Power and Z architecture systems are in trouble? Does it mean that x86 has won at last? No, it doesn't. Systems based upon these architectures have been successful over the years, and the technology has kept up with and surpassed the state of the art regularly.

IBM believes a company specializing in chip manufacturing might be a better home for its products, so it's handing over that portion of its business to Globalfoundries. It's likely to involve other manufacturers later, after the 10-year agreement comes to an end. At that point, IBM would be in a position to deploy the "let's you and him fight over IBM's business" strategy to obtain better products at lower prices.

IBM would be free to continue basic research and enhance microprocessor design without having to foot the bill for one foundry after another after another. Is this a good move? It's hard to tell at this early stage, but ARM has demonstrated that it's a workable approach. ARM designs are powering a majority of smartphones, tablets and a growing number of servers. IBM hopes to follow that path to success.