Description

AMD AND INTEL: OPPORTUNITY ABOUNDS

As the 10-year stock price chart of AMD shows, the market is almost comically bad at pricing in step changes in relative product quality between Intel and AMD, even when newly released chips from both companies have been fully benchmarked and reviewed. We'll briefly review the past decade of the Intel/AMD rivalry to highlight the opportunity available to investors who pay attention.

Chart at http://2.bp.blogspot.com/_BbvwQAOOzxw/TNhP5yrET5I/AAAAAAAAAkQ/pkXNpNV7DIs/s1600/amd-intc.JPG  

2003: AMD Introduces the Opteron

There are many similarities between the server CPU market in 2002 and 2010. Much like today, in 2002, Intel dominated the market with 90%-plus share and AMD was relegated to picking up scraps. However, in early 2003, AMD introduced the original Opteron, which handily took the performance crown away from Intel and eventually captured roughly 25% of the server market. Again much like today, one reason the Opteron was successful was that AMD took a much different architectural approach than Intel.

When the Opteron was introduced and benchmarked in April 2003, AMD's share price was languishing in the midsingle digits even though it was clear that AMD would have Intel beat for the upcoming processor generation. The following chart published on the AnandTech website on April 23, 2003 (http://www.anandtech.com/show/1099/5), showed the Opteron 244 outperforming the Intel Xeon by 13% in a CPU-bound benchmark, even though the Xeon was clocked a full gigahertz higher.

The market eventually caught on as AMD racked up market share and the stock peaked at more than $40 in early 2006.

2006: Intel Introduces the Core 2

We argue that one reason the Intel/AMD trade is so lucrative is that investors tend to extrapolate current processor generation trends far into the future. However, each processor generation represents an opportunity for step changes in relative product quality between Intel and AMD. After the NetBurst and Itanium debacle, Intel executed an amazing turnaround. The company scrapped its processor roadmap and refocused on an architecture developed by Intel's mobile group in Israel. That architecture was the basis for Intel's Pentium M notebook chips, and its descendant was released in mid-2006 as the Core 2 family of chips.

The Core 2 architecture was a massive step change improvement in the two critical metrics of price/performance and performance/watt. Benchmarks showing that Core 2 outperformed AMD by 25%-30% began appearing online in March 2006 when AMD shares were trading in the mid-$30s (e.g., http://www.anandtech.com/show/1966/1). Once the details of Core 2 were known, AMD shares were an undeniable short. Core 2, along with its successor Nehalem, allowed Intel to use its traditional advantages of scale and process technology to recapture market share and crush AMD's margins.

AMD'S BULLDOZER ARCHITECTURE

AMD has been noncompetitive in server CPUs during the past few years, but we expect that will change with the introduction of chips based on the Bulldozer architecture in mid-2011. Bulldozer will be AMD's first new architecture since the original Opteron in 2003, and much like the Opteron, AMD has again taken a very different architectural path than Intel.

We emphasize that we do not expect Bulldozer to outperform Intel's 2011 server chips to the degree that the Opteron or Core 2 did when they were introduced. We simply believe the market is not pricing in any measure of success for Bulldozer, which, given our current knowledge of the architecture, is too pessimistic. AMD has very little server market share left to lose, and this is a case where AMD will feel better once it stops getting repeatedly punched in the face by Intel.

Intel Is a Ferrari, AMD Is a 16-Passenger Van

AMD's key architectural decision with Bulldozer is to share certain chip circuitry such as the floating point unit among two integer units, which allows AMD to fit more cores within a specified silicon area and power consumption budget. Each of the two integer units in a Bulldozer module will appear as a separate core to the operating system. By comparison, Intel's chips sport cores with much less shared circuitry and a 1:1 ratio of integer and floating point units. 

Bulldozer chips are likely to face off against chips based on Intel's new Sandy Bridge architecture. Given what we currently know, we believe Sandy Bridge will win hands down against Bulldozer in single-threaded processor performance. However, AMD is focused on overall chip throughput, rather than pure performance. There are certain workloads where Bulldozer's greater core count and throughput will be an advantage. To use an analogy, Sandy Bridge is a finely tuned Ferrari, while Bulldozer is 16-passenger van. Sandy Bridge will get you where you need to go quicker, but Bulldozer will let you bring along 15 of your closest friends for the ride. Sometimes you just need the van.

The Virtualized Datacenter

One area where Bulldozer's high core count will be an advantage is virtualized environments. A majority of enterprise application workloads are integer based (thus we do not believe Bulldozer will lose much by sharing a floating point unit among two integer units) and IT administrators tend to follow a one-virtual-machine-per-core rule. Every physical server entails a fixed cost in terms of datacenter space and power consumption. Thus, the more cores that can be packed into a physical server, the more virtual machines that server can host, with the upshot that fewer physical servers will be needed. For example, a server with four CPU sockets can be outfitted with four 16-core AMD Interlagos Bulldozer chips for 64 cores versus four 8-core Intel Xeon chips for 32 cores. A physical server with Bulldozer chips can potentially host twice as many virtual machines as one with Intel chips (assuming memory and I/O are also scaled appropriately).

Virtualization is a secular growth trend in technology, and we expect Bulldozer to be especially attractive for enterprises looking to consolidate their servers with virtualization.

POTENTIAL FINANCIAL IMPACT OF BULLDOZER

Server CPUs sell at much higher prices than desktop/notebook CPUs, and accordingly have much higher margins. We estimate that Intel's server CPU gross margins are 70%-75%. In modeling the potential financial impact of Bulldozer, we make three key assumptions: (1) AMD is able to capture an incremental 10 points of server market share, resulting in an additional $225 million of quarterly revenue ($900 million annual run rate); (2) AMD realizes 65% gross margins on Bulldozer server CPUs; and (3) AMD's operating costs stay relatively flat, given the high fixed cost nature of the business. As shown in the following table, all other things equal, an incremental $225 million flowing through at a 65% gross margin should allow AMD to approximately double its operating income.

Hypothetical AMD Quarterly Results

Table at http://2.bp.blogspot.com/_BbvwQAOOzxw/TNhRo_QwhrI/AAAAAAAAAkg/1mRdm2DHBjM/s1600/amdtable2.GIF  

WE BELIEVE AMD IS A TIMELY LONG IDEA AHEAD OF BULLDOZER

Catalyst