Divining Intel's Future
Its core computer business is maturing, and its new businesses in communications are generating losses instead of hoped-for growth. Some of those new businesses are so woven into the company's fabric that it will not be easy, or maybe even possible, to dispose of them.
Welcome to Paul Otellini's world. The recently installed chief executive of Intel Corp. has said he will erase $1 billion in expenses from the company's books, eliminating loss-making groups to focus on Intel's traditional X86 computing business.
In this environment, Intel is shopping its Xscale processor, which serves communications systems ranging from cell phones to Internet routers, according to a recent report in the San Jose Mercury News. However, several analysts pan such a move as unworkable and not in Intel's best interests.
Others have speculated Intel may be planning to sell its flash business after a reorganization that put all the elements of the group in a tidy business unit. Still other reports have emerged that Intel may sell or kill any number of marginal units, including its Dialogic group.
Clearly, no one knows just what Intel will do. Likely Otellini himself has not yet made up his mind on the sweeping reorg he has suggested will be announced in July or August.
Here are some suggestions for the embattled executive culled from our editors and from top analysts who have tracked the company for many years:
• Sell or spin off the flash unit.
• Retain Xscale, but reorganize it into an embedded division under new management that cultivates a new culture.
• Transition out of Itanium as soon as possible, replacing it with multicore, multithreaded X86 CPUs.
• Revamp the fab strategy to em- brace foundry partners that would co-develop process technology and share capacity requirements.
Goodbye flash
Perhaps the easiest part of a reorg along these lines would be selling or spinning off the NOR flash unit. The 2005 spinout of Spansion by Advanced Micro Devices Inc. was a highlight of an otherwise lackluster year in high-tech stocks. Intel could do worse than follow AMD's lead.
The flash business has standalone products, fabs and road maps--and its losses are currently increasing under Intel management. "There's no reason for them to be in this anymore," said a contrarian Wall Street analyst who asked not to be identified.
Once upon a time, Intel made flash in its N–1-node fabs as a way to amortize capital expenses. But those logic fabs now use copper interconnects, 300-mm wafers and other technologies flash doesn't need. "There's no manufacturing synergy anymore," said the Wall Streeter, and only a small sales opportunity in bundling flash with other products.
Intel recently consolidated its NOR fabrication operations, previously managed by the logic technology and manufacturing group, into a single flash operation. That gave rise to speculation that Intel would spin off the unit, as AMD did with Spansion.
But Alan Niebel, a flash memory analyst at Web-Feet Research Inc. (Monterey, Calif.), said he believes Intel will not spin out its NOR flash division, which accounted for about $2.5 billion in revenue last year.
What Intel is more likely to do, Niebel said, is eliminate some of the long-term research projects in nonvolatile-memory technology, much of which has been kept under wraps anyway.
Intel will continue to invest in the separate NAND flash business through its joint venture with Micron Technology Inc., IM Flash Technologies LLC.
"There is a huge opportunity over the next few years in creating mobile computers with solid-state storage instead of rotating media," Niebel said.
Weighing Xscale
The tricky bit is what to do with Xscale, especially in the wake of reports that Intel will sell the processor.
"Wall Street is sold on this idea. They love to fantasize about companies selling business units, but in reality it's much harder than anyone thinks," said the anonymous Wall Street insider.
Selling off the Xscale unit would be tough for many reasons. The core is made in Intel's unique process technology. "The first thing a buyer would have to do is port the design to TSMC. What a pain in the butt," said the Wall Streeter.
The Xscale core is currently used in both wireless and wired silicon products, including the Hermone cellular processor and the IXP router and storage processors. If a buyer doesn't have products in all those fields, it would have to kill or sell off the parts it does not want. But who would want a part based on a core it does not own?
Adding to the complexity, Xscale is woven into a web of intellectual property. The cellular products depend on a baseband DSP that Intel co-developed with Analog Devices Inc. In addition, Intel has analog technology tied up in both the wired and wireless Xscale variants.
What's more, Intel's license to the Xscale's underlying ARM core is a unique architectural license that allows it to design and brand ARM variations. In an ARM-dominated world, that license could command $15 million on its own, according to Will Strauss of Forward Concepts (Tempe, Ariz.). But transferring such rights could be a three-party legal nightmare.
Thus, said the anonymous Wall Street source, selling Xscale would be "a monstrously complex exercise."
"Intel wants to make one clean transaction and call it a day. But anyone who buys the Xscale group may have a big mess to clean up," agreed Linley Gwennap, head of market watcher The Linley Group (Mountain View, Calif.).
Strauss said Intel sold about 6 million of the 60 million applications processors for cell phones that shipped last year, making it a distant fourth behind Texas Instruments, Qualcomm and Renesas. Gwennap estimated Intel sold about $30 million in IXP network processors, a category that has never taken off as expected.
"We pegged the whole net processor market last year at $170 million. That's not the kind of market Intel is interested in, even if it doubles in the next few years--and it just might," Gwennap said.
Xscale evolved from the StrongARM design Intel incidentally acquired when it bought the semiconductor business of Digital Equipment Corp. in 1998. In those dot-com boom days, then-CEO Craig Barrett was pushing the company to diversify into comms. Intel retooled the high-Mips-per-watt StrongARM for a range of wired and wireless businesses, but because of the dot-com bust and the company's missteps, it has failed to execute on any of them. "Intel is the world's largest wireless wannabe," said Strauss.
A $2 billion acquisition of DSP Communications failed to net the company the designs or design wins it sought. An integrated Xscale, DSP and Strataflash design has so far failed to gain much market traction beyond use in upcoming products from Research in Motion.
Part of the problem, Gwennap noted, is that Intel likes to rotate senior managers through different divisions to give them broad experience. But that practice often creates groups headed by managers who lack expertise in the given field. By contrast, Broadcom and, more recently, Freescale have hired execs from cell phone makers to head their cellular-chip groups, he said.
In addition, Intel often bars small-volume products in emerging businesses from access to its latest process technology, which is reserved for X86 CPUs. "They are squandering one of their greatest advantages here. They might as well send it to a Taiwan foundry," said Gwennap.
What to do
Given these dynamics, Intel should keep Xscale, but revamp how it handles the product. CEO Otellini should focus on the proven, high-volume market for cellular processors, but kill the wired comms businesses--or find a partner to help nurture them.
As Gwennap points out, Xscale could also serve as a basis for an embedded business in digital home products beyond the PC. That's a near-term, fast-moving growth market that Intel needs.
Wired comms systems are more dicey. They typically have long qualification periods, so it can take years for volume shipments to materialize--and the large equipment companies need assurances that a chip will be around for years, said Peter Middleton, a wireline-networking research analyst at Gartner.
Intel has never had a successful business strategy that supports a product with a diverse customer base, a relatively low average selling price and a long product life cycle. In recognition of that fact, Intel should create a separate embedded division, managed by executives with a proven track record in consumer, wireless and other embedded markets.
"The Intel 186 processor is still used in a wide variety of networking and industrial equipment, and the only second source is AMD. The i960 is used in printers. These are old processors, but an OEM usually doesn't change processors for a product line that still sells," said one source who asked not to be identified.
The new group will need to develop a culture of partnering that is broader and deeper than what's seen in Intel's computer divisions. It will need many software partners to fill the gaps usually handled for Intel by Microsoft in the homogenous PC world. It should also form foundry relationships so it can put both new and old versions of its Xscale and X86 products in non-Intel fabs to save money.
In the mobile space, Intel may already be laying plans to replace or at least match Xscale with X86 CPUs.
In Austin, Texas, Intel has been hiring design engineers to work on an ultralow-power processor aimed at handheld computers that would run the Windows operating system. In 2004, Intel canceled the single-core Tejas design effort after concerns about power consumption prompted a shift to dual-core designs running at lower speeds.
The Tejas design team, led by Brad Beavers, site director for Intel in Austin, was put to work on the ultramobile-PC processor. Intel's work force in Austin has increased from 500 to 800, including design teams working on the next-generation Xscale, an Intel Austin engineer said.
In the wired world, Tom Steenman, general manager of Intel's infrastructure processor division, fended off speculation around the comms reorg at last week's Globalcomm trade show in Chicago. "There are a lot of people connecting unconnected dots," Steenman said, referring to various media reports Intel is selling Xscale and putting older X86 em- bedded parts on end-of-life status.
On the network processor front, Steenman said Intel remains committed to the IXP440 processor, as part of a combined "service processor and content processor suite" that would include X86 parts, particularly Xeon, in control-plane roles.
Xscale, he said, was never a processor family, but rather an embedded-core technology used in designing products for specific markets. Some products designed from Xscale, such as the PXA26x processor family for mobile devices, have been placed on end-of-life notice, though other families survive.
Similarly, older microcontrollers such as the 960 are being phased out, though Steenman said speculation that Intel is getting out of embedded markets ignores the aggressive role the company has played in board-level AdvancedTCA products. Standard X86 products, particularly in multicore implementations, will be more important in future embedded boards than dedicated processors for packet forwarding, he added.
Rumors that optical communications were on the corporate hit list arose last month, when Intel closed a Copenhagen, Denmark, facility belonging to Giga S/A, a company Intel acquired in 2000 that developed physical-layer chips for electronic dispersion compensation and adaptive equalization of fiber networks. The Giga closure related to specific product lines, Steenman said, and had nothing to do with the Fremont, Calif., facility producing tunable lasers and optical transponder modules. He called optical components "a critical technology for Intel's future."
Itanium's prospects
No discussion of loss-making businesses at Intel would be complete without reviewing the Itanium. The part, co-developed with Hewlett-Packard Co., was supposed to be a next-generation very long-instruction-word (VLIW) architecture that would supplant the X86. Instead it has been confined to a "niche of a niche" as a high-end data center CPU, said one analyst who asked not to be named.
"I don't see any reason for dumping more money in that pit," said the analyst.
However, HP has built a server business on Itanium valued at more than $1.6 billion in its 2005 fiscal year and growing at slightly more than 90 percent. A handful of other OEMs are also dabbling in Itanium, which may survive Sun Micro- systems Inc.'s Sparc as the last RISC CPU standing next to the IBM Corp. Power architecture.
Intel plans to put both Itanium and its X86 Xeon processors on a Common Systems Interface bus sometime over the next year or two as a way to leverage components between the platforms. That move would also afford Intel the opportunity, in 2009 or so, to phase out or sell off Itanium--perhaps to its only big customer, HP.
Meanwhile, AMD is positioning its multicore X86 Opterons as a single-scalable architecture that will eat into Intel's high-end server business. Intel cannot afford to let that play out. It must aggressively gear up multicore, multithreaded Xeons to stave off further incursions by AMD into the highly profitable market for big-iron servers.
Fab strategy
Finally, Intel needs to stop the craziness of being the last chip maker to go it alone in process technology and fab capacity. This is a train that leads to bankruptcy if you extrapolate into the future far enough. Here, Otellini could take a page from IBM or AMD, which have formed successful partnerships to share the costs of process development and capacity planning.
Any foundry would die to have access to an Intel process technology. The question is whether Intel can manage a partnership in the area of its crown jewels.
If Intel can find a way to migrate even some of its designs to more standard CMOS techniques for their second-generation versions, it could lower capital expenses and product costs. The move would also ease the pain of supporting all those embedded 186 products that some loyal embedded OEMs will want to buy until hell freezes over.
Intel's manufacturing might is one of its greatest strengths. Dan Hutcheson, CEO of VLSI Research Inc., said Intel will soon make more processors on its 65-nm technology than on the preceding 90-nm process, a crossover that will occur before AMD ships its first 65-nm-based product from its Dresden, Germany, fab.
But today that strength only serves the diminishing performance needs of high-end PC processors. Intel's growth depends on finding a profitable road to low-power embedded products for cell phones and consumer devices. And that means a major shift in fab strategy.
Intel has never been afraid of change. The company has explored opportunities as diverse as ASIC design and Web hosting. If it is to grow beyond today's maturing computer market, Intel must under- take even more radical changes in how it makes and markets silicon.
Like the rest of the electronics industry, we await Mr. Otellini's next move.
-- David Lammers, Loring Wirbel and Dave Bursky contributed to this story.
