Intel throws good money after a bad chip
After spending billions of dollars, why hasn't Intel pulled the plug on Itanium yet?
SAN FRANCISCO (Business 2.0 Magazine) - Much to Wall Street's dismay, Intel is continuing to invest billions of dollars in Itanium, a chip that most of the industry has written off.
Keeping the peace
Last week, Intel (Research), Hewlett-Packard (Research), and some smaller server makers announced plans to spend $10 billion to promote Itanium-based servers, with half of that sum coming from HP, whose Integrity server line runs on Itanium. Industry observers believe that most of the remainder is coming from Intel.
This additional spending comes on top of billions of dollars in R&D expenditures that went into bringing the chip to market.
Why is Intel throwing good money after bad? Intel depends on hardware and software partners who design servers and code specifically for its chips. Much of the value of Intel's popular desktop chips comes from the fact that, say, HP PCs and Microsoft's (Research) Windows operating system run on top of them.
For those partners who have invested heavily in Itanium, killing the chip would be an unacceptable betrayal on Intel's part, leaving their Itanium customers stranded and likely to seek out new systems from rivals. Such a move would have ramifications beyond Itanium's server niche, prompting Intel's partners to punish the chipmaker by buying chips from other suppliers.
Keeping Itanium alive might seem like an awfully high price to pay for the software industry's loyalty. But with a resurgent AMD (Research) offering an increasingly attractive alternative to Intel's desktop and server chips, Intel needs to stay in partners' good graces.
Itanium was born 12 years ago out of a collaboration between Intel and HP. At the time, high-end server makers like IBM (Research), Sun (Research), and HP designed their own chips as well as their own servers. HP wanted a partner to share the costs of a new chip design, while Intel, whose chips at the time were only used in desktop PCs and cheap servers, wanted to become a supplier to the high end of the server market.
Intel and HP planned to crack that market with a processor that gulped down software code 64 bits at a time. In the 1990s, Intel chips were limited to 32 bits, which hobbled Intel-based servers' ability to compete with 64-bit servers from IBM and Sun. Intel and HP also planned to design a chip that would have a key advantage: The ability to run existing software for 32-bit Intel processors without requiring software makers to rework their code -- most notably Microsoft's ubiquitous Windows operating system.
A rough start
Itanium faced long delays in getting to market, however, and when it was introduced in 2001, its performance running 32-bit software was abysmal. It turned out that to get any performance gains out of Itanium, software makers would have to rewrite their code after all.
That gave longtime Intel rival Advanced Micro Devices an opening to beat Itanium at its own game. In 2003, AMD introduced its first Opteron chips, which ran both 32-bit and 64-bit software well.
To catch up, Intel copied AMD's approach and adapted an existing 32-bit chip line to also run 64-bit software. Intel now sells those chips, called Xeon, alongside Itanium.
IBM and Dell (Research) stopped making Itanium servers in 2005, and Sun abandoned plans to translate its Solaris operating system for Itanium systems. IBM and Sun now make servers with AMD's Opteron chips, while Dell uses Intel's hybrid 32/64-bit Xeon chips.
Itanium has also missed a major technical shift -- the introduction of multiple processing "cores" inside a single chip. Dual-core processors can run more efficiently than single-core processors, and Intel has already introduced them into its Xeon line, but the first dual-core Itanium chip isn't expected out until later this year.
Eric Ross, lead semiconductor analyst at ThinkEquity Partners, calls Itanium's situation a "death spiral," noting that by some estimates, the industry has only sold $2 billion of Itanium systems last year, a tiny slice of the $48 billion sever market.
"Why aren't Intel and HP spending $10 billion on marketing Woodcrest?" asked Ross, referring to Intel's codename for an upcoming 32/64-bit server chip which Intel claims will outperform AMD's offerings.
That's the $10 billion question.
Intel has two unattractive choices: Kill Itanium and risk its partners' wrath, or split its marketing and R&D budgets between Xeon and Itanium. Either way, AMD ends up with a strategic advantage -- and Intel with a silicon albatross.
To send a letter to the editor about this story, click here.