AMD's New Processors Gaining A Foothold

| About: Advanced Micro (AMD)

By Andy Ng

Advanced Micro Devices (NYSE:AMD) remains one of our favorite semiconductor companies, as we believe its new products are starting to boost its competitive position relative to Intel (NASDAQ:INTC). Although we anticipated that AMD would have begun to see significant share gains in the microprocessor market by now, our thesis that the firm's new products are competitively differentiated from Intel's appears to be mostly intact. AMD has had manufacturing issues at foundry partner GlobalFoundries, but its Fusion Llano processors have been popular with PC manufacturers in notebooks. Although reviews and benchmarks for the new Bulldozer processors have been mixed so far, they appear compelling for certain applications in the server chip segment.

In designing Bulldozer, AMD decided to focus on increasing the number of cores on the microprocessor. This entailed trade-offs, namely less of a focus on single-threaded performance (roughly equivalent to performance per core), while emphasizing total overall microprocessor throughput by packing more cores on the chip. This is quite a different approach from Intel's Sandy Bridge and upcoming Ivy Bridge processors, which are tailored for superior single-threaded performance. The benchmarks of Bulldozer have demonstrated what we expected, namely that the processors perform well in multithreaded applications that can take advantage of the large number of cores. This is important for the server CPU segment, where multithreaded workloads are prevalent.

Nonetheless, not all has been smooth for Bulldozer. The Bulldozer desktop variant has been somewhat of a disappointment, mainly because the single-threaded performance of these chips lags Intel's processors. PC applications tend to be mostly single-threaded and typically don't make use of multithreaded processes, which means that today's desktops can't really take advantage of the high number of cores offered by Bulldozer. However, our focus here is not the desktop segment but the lucrative server processor market, where the firm still has opportunities to gain share.

Overall, the reviews of Bulldozer server CPUs, which have as many as 16 cores, have been OK so far, indicating that the chips could improve AMD's competitive position in the server segment, though there are aspects that have come in below our expectations. We had expected a bigger boost in single-threaded performance, but multithreaded performance, where Bulldozer generally does well, is much more important in servers. The chip appears somewhat competitive in power efficiency, though we suspect that there is room for significant improvements in both single-threaded performance and power efficiency via software optimizations to take advantage of the new chip architecture, which the benchmarks so far have lacked. Intel's new server versions of its Sandy Bridge processors, which are scheduled to launch sometime in the first quarter of 2012, will have as many as 8 cores, so Bulldozer will continue to have the core count advantage over the competition.

Bulldozer still appears very suitable for certain applications that incorporate highly threaded workloads, particularly virtualized data center environments, a growing opportunity that may help AMD capture market share in the server CPU segment. It has been shown that a server's capacity to run virtual machines tends to scale with the number of cores on a server processor, as well as the amount of memory in a system, and AMD's new server chip has shown advantages on these fronts.

2012 Fusion Products Should Maintain Competitive Differentiation

We believe the Fusion pipeline in 2012 should allow AMD to remain differentiated from Intel. The Fusion products have been winning market share in the notebook processor segment, with Llano being particularly successful. Llano's successor will be Trinity, which will contain an enhanced Bulldozer core, code-named Piledriver, and incorporate high-end graphics capabilities based on AMD's next-generation GPUs. The chip is anticipated to come out in early 2012, possibly in the first quarter. We expect Intel's 2012 Ivy Bridge processors to have better CPU performance than Trinity. However, Trinity should hold a significant GPU performance lead over Ivy Bridge, which should allow AMD to continue its momentum in the notebook processor market.

In netbooks and low-end PCs, AMD was supposed to replace the Fusion Zacate and Ontario APUs with Krishna and Wichita. However, rumors have been circulating that Krishna and Wichita have been canceled, and that AMD will merely be updating Zacate and Ontario with performance enhancements. Zacate and Ontario are being manufacturing by TSMC, while it is believed that Krishna and Wichita were to be produced at GlobalFoundries. It appears that AMD had second thoughts about moving manufacturing of its low-end Fusion chips to GlobalFoundries and has decided to stick with TSMC, which caused the Krishna/Wichita cancellation. We are not too concerned about this segment, as the bulk of the value in the PC processor market is in the mainstream notebook and desktop segment, which is covered by Llano and Trinity. In addition, Zacate's and Ontario's CPU, GPU, and power efficiency performance have been significantly better than Intel's chips, which means that even if Intel can narrow the low-end PC processor gap in 2012, AMD should stay competitive.

On the server CPU front, AMD will continue to add cores to its processors, with Terramar and Sepang coming out some time in 2012. Terramar will have as many as 20 Piledriver cores, while Sepang will have up to 10, as the firm will continue to target server applications with highly threaded workloads. Given that Piledriver is an improved version of Bulldozer, we believe there may be opportunities for AMD to improve the single-threaded performance of these chips. Intel's latest server CPU is based on the 2011 Sandy Bridge architecture, but delays have caused the firm to push back the launch to the first quarter of 2012. We expect Intel's CPU to take the cake in terms of single-threaded performance, but the large number of cores in AMD's server chips should make AMD's Bulldozer and Piledriver products very suitable for certain applications.

We Expect Continued Fusion Momentum, But No Share Gains From Bulldozer

Our fair value estimate for AMD is $11 per share. We project continued success for Fusion in the PC segment, but we no longer expect share gains from Bulldozer in the server processor market.

Given Fusion's far-superior graphics technology, we think it will be at least several years before Intel can close the gap on that front. As a result, we believe AMD has opportunities to hold on to the share gains that it achieves during our forecast period. Although we now assume that Bulldozer will no longer drive share gains in the server processor market segment, our forecast of 3% revenue market share is very low and probably a trough. Therefore, any market share increases in the server CPU market would represent upside to our fair value estimate. In addition, we estimate that average selling prices in the desktop and notebook processor market will rise slightly in our forecast, driven by the shift from stand-alone GPUs to on-chip graphics.

We expect gross margins to trend higher over time, primarily because of AMD's Fusion processors. We estimate that these chips have gross margins at or slightly above 50%, though we project there will be some slight declines toward the end of our forecast period. In the past several years, AMD's gross margins have tended to fluctuate around 40%. However, we project that a lack of a competitive server processor product somewhat offsets the margin gains Fusion provides. The main assumption and key risk in our forecast is that AMD's manufacturing problems at GlobalFoundries can be solved in the next couple of quarters, so that the firm's market share gains in the notebook segment show up on the top line by early or mid-2012.

In our bull scenario, we forecast market share gains for both Bulldozer and Fusion as well as improved profitability, resulting in a fair value of $16 per share. When we assume minimal share gains for Fusion and Bulldozer in our bear case, our fair value comes out at $4.50 per share.

We expect AMD's recent momentum in the notebook processor segment to continue, as Fusion has proved to be a compelling product and the upcoming launch of Llano successor Trinity should further benefit the firm. However, we have tempered our expectations for the Bulldozer server processors, as it appears the chips won't perform as well as we had anticipated. This resulted in a reduction to our fair value estimate for AMD.

AMD Is Gaining Material Share in Notebooks

Our opinion of AMD so far has been based on the firm's 2011 product cycle, consisting of its new Fusion lineup, targeted at PCs, and its latest Bulldozer processors, primarily targeted at servers. While we believe both of these product groups provide AMD with competitive differentiation, we think only one of these really needs to be successful for AMD to see a substantial boost to its top and bottom lines. So far, the Fusion chips have fared very well in the notebook processor segment, an area where AMD has traditionally had limited success and therefore has plenty of opportunities for substantial market share gains.

Fusion processors contain a central processing unit and a high-end graphics processing unit on a single piece of silicon, which AMD calls an accelerated processing unit. The Fusion line consists of Ontario and Zacate for netbooks and low-end notebooks and desktops and Llano for mainstream notebooks and desktops. These APUs allow the firm to combine its core microprocessor capabilities with its cutting-edge technology expertise in graphics. Intel's Sandy Bridge and upcoming Ivy Bridge processors also have a combined CPU/GPU, but its graphics capabilities can't match those of AMD. As a result, while Intel can offer superior CPU performance, AMD easily has the upper hand in GPU technologies. The latest CPU/GPU generation is another factor to consider when choosing a processor, both for PC makers designing new computers and consumers purchasing PCs. We consider AMD's ability to offer high-end graphics in its Fusion chips to be an advantage, as it probably tilts the decision-making process in the firm's favor and away from the sole consideration of CPU speeds, which have traditionally been Intel's stronghold.

AMD has held significant unit market share in desktop processors, typically in the upper 20s, but it has had limited success in the past in the notebook chips, as its share in that segment historically languished in the low teens. We have been forecasting that the firm would be able to take advantage of Fusion to gain share in the mobile PC front, which appears to now be coming to fruition. The competitive differentiation offered by the Fusion products, as well as significant improvements in power efficiency and battery life versus AMD's previous generation of mobile processors, has allowed the firm to capture notable market share from Intel in the notebook chip segment over the past several quarters.

Manufacturing Issues Mask Fusion's Success

The latest data clearly show that AMD is gaining momentum with Fusion in the notebook CPUs, but manufacturing problems at foundry partner GlobalFoundries have hampered the firm. Production issues have limited the supply of Llano APUs, which are targeted at mainstream notebooks and desktops. These problems have masked the success that AMD is having in notebook processors, as the firm has chosen to sacrifice shipments of desktop processors and older products in order to shift available production capacity toward mobile Llano APUs. As a result, AMD saw its desktop CPU unit market share drop to 24.1% in the third quarter from 28.9% in the second quarter, according to IDC. We estimate that if there had been no manufacturing issues and the firm's desktop share had held steady at about 28%, AMD would have seen an additional 1.6 million processor unit shipments to the desktop PC market segment during the third quarter. This estimate may be conservative, as the supply of Llano APUs for notebooks was probably affected as well, despite AMD's efforts to minimize the impact of the manufacturing problems on the availability of these chips.

AMD's decision to divert desktop CPU production capacity to maximize the supply of Fusion Llano notebook APUs was strategically correct. The firm traditionally has had a more significant presence in the desktop CPU segment, but its success has been limited in notebooks. As a result, we believe AMD is better off attempting to capture notebook CPU market share and gain mind share in that segment with PC makers and users, while temporarily ceding desktop processor share, as it should be easier for the firm to win back market share in desktops than the other way around. Nonetheless, AMD will not be able to fully enjoy the success of Fusion until the manufacturing issues at GlobalFoundries are resolved. We think the problems can be fixed within a couple of quarters after arising in the third quarter.

While Intel will roll out its new Ivy Bridge processors at the beginning of 2012, we still do not expect the on-chip graphics to be able to match Fusion's capabilities. Intel is essentially taking low-end integrated graphics technology and attempting to make it more powerful, while AMD already has high-end GPU capabilities, and we think it will take years for Intel to develop high-end graphics expertise. In addition, AMD will launch its Fusion Trinity product in 2012, which should allow it to build on its current momentum with Llano.

Bulldozer Can Still Clear the Way for Server Share Gains

AMD began shipping its Bulldozer processors, which are targeted at servers and high-end desktops, in early September, and initial benchmarks and reviews have been a mixed bag so far. Although Bulldozer doesn't excel at every benchmark, we think it has advantages in certain server applications. However, we have tempered our forecasts for Bulldozer, since the chip's overall performance has come in below our expectations.

In designing Bulldozer, AMD decided to focus on increasing the number of cores on the microprocessor. This entailed trade-offs, namely less of a focus on single-threaded performance (roughly equivalent to performance per core), while emphasizing total overall microprocessor throughput by packing more cores on the chip. This is quite a different approach from Intel's Sandy Bridge and upcoming Ivy Bridge processors, which are tailored for superior single-threaded performance. The benchmarks of Bulldozer have demonstrated what we expected, namely that the processors perform well in multithreaded applications that can take advantage of the large number of cores. This is important for the server CPU segment, where multithreaded workloads are prevalent.

Nonetheless, not all has been smooth for Bulldozer. The Bulldozer desktop variant has been somewhat of a disappointment, mainly because the single-threaded performance of these chips lags Intel's processors. PC applications tend to be mostly single-threaded and typically don't make use of multithreaded processes, which means that today's desktops can't really take advantage of the high number of cores offered by Bulldozer. However, our focus here is not the desktop segment but the lucrative server processor market, where the firm still has opportunities to gain share.

Overall, the reviews of Bulldozer server CPUs, which have as many as 16 cores, have been OK so far, indicating that the chips could improve AMD's competitive position in the server segment, though there are aspects that have come in below our expectations. We had expected a bigger boost in single-threaded performance, but multithreaded performance, where Bulldozer generally does well, is much more important in servers. The chip appears somewhat competitive in power efficiency, though we suspect that there is room for significant improvements in both single-threaded performance and power efficiency via software optimizations to take advantage of the new chip architecture, which the benchmarks so far have lacked. Intel's new server versions of its Sandy Bridge processors, which are scheduled to launch sometime in the first quarter of 2012, will have as many as 8 cores, so Bulldozer will continue to have the core count advantage over the competition.

Bulldozer still appears very suitable for certain applications that incorporate highly threaded workloads, particularly virtualized data center environments, a growing opportunity that may help AMD capture market share in the server CPU segment. It has been shown that a server's capacity to run virtual machines tends to scale with the number of cores on a server processor, as well as the amount of memory in a system, and AMD's new server chip has shown advantages on these fronts.

2012 Fusion Products Should Maintain Competitive Differentiation

We believe the Fusion pipeline in 2012 should allow AMD to remain differentiated from Intel. The Fusion products have been winning market share in the notebook processor segment, with Llano being particularly successful. Llano's successor will be Trinity, which will contain an enhanced Bulldozer core, code-named Piledriver, and incorporate high-end graphics capabilities based on AMD's next-generation GPUs. The chip is anticipated to come out in early 2012, possibly in the first quarter. We expect Intel's 2012 Ivy Bridge processors to have better CPU performance than Trinity. However, Trinity should hold a significant GPU performance lead over Ivy Bridge, which should allow AMD to continue its momentum in the notebook processor market.

In netbooks and low-end PCs, AMD was supposed to replace the Fusion Zacate and Ontario APUs with Krishna and Wichita. However, rumors have been circulating that Krishna and Wichita have been canceled, and that AMD will merely be updating Zacate and Ontario with performance enhancements. Zacate and Ontario are being manufacturing by TSMC, while it is believed that Krishna and Wichita were to be produced at GlobalFoundries. It appears that AMD had second thoughts about moving manufacturing of its low-end Fusion chips to GlobalFoundries and has decided to stick with TSMC, which caused the Krishna/Wichita cancellation. We are not too concerned about this segment, as the bulk of the value in the PC processor market is in the mainstream notebook and desktop segment, which is covered by Llano and Trinity. In addition, Zacate's and Ontario's CPU, GPU, and power efficiency performance have been significantly better than Intel's chips, which means that even if Intel can narrow the low-end PC processor gap in 2012, AMD should stay competitive.

On the server CPU front, AMD will continue to add cores to its processors, with Terramar and Sepang coming out some time in 2012. Terramar will have as many as 20 Piledriver cores, while Sepang will have up to 10, as the firm will continue to target server applications with highly threaded workloads. Given that Piledriver is an improved version of Bulldozer, we believe there may be opportunities for AMD to improve the single-threaded performance of these chips. Intel's latest server CPU is based on the 2011 Sandy Bridge architecture, but delays have caused the firm to push back the launch to the first quarter of 2012. We expect Intel's CPU to take the cake in terms of single-threaded performance, but the large number of cores in AMD's server chips should make AMD's Bulldozer and Piledriver products very suitable for certain applications.

We Expect Continued Fusion Momentum, but No Share Gains From Bulldozer

Our fair value estimate for AMD is $11 per share. We project continued success for Fusion in the PC segment, but we no longer expect share gains from Bulldozer in the server processor market.

Given Fusion's far-superior graphics technology, we think it will be at least several years before Intel can close the gap on that front. As a result, we believe AMD has opportunities to hold on to the share gains that it achieves during our forecast period. Although we now assume that Bulldozer will no longer drive share gains in the server processor market segment, our forecast of 3% revenue market share is very low and probably a trough. Therefore, any market share increases in the server CPU market would represent upside to our fair value estimate. In addition, we estimate that average selling prices in the desktop and notebook processor market will rise slightly in our forecast, driven by the shift from stand-alone GPUs to on-chip graphics.

We expect gross margins to trend higher over time, primarily because of AMD's Fusion processors. We estimate that these chips have gross margins at or slightly above 50%, though we project there will be some slight declines toward the end of our forecast period. In the past several years, AMD's gross margins have tended to fluctuate around 40%. However, we project that a lack of a competitive server processor product somewhat offsets the margin gains Fusion provides. The main assumption and key risk in our forecast is that AMD's manufacturing problems at GlobalFoundries can be solved in the next couple of quarters, so that the firm's market share gains in the notebook segment show up on the top line by early or mid-2012.

In our bull scenario, we forecast market share gains for both Bulldozer and Fusion as well as improved profitability, resulting in a fair value of $16 per share. When we assume minimal share gains for Fusion and Bulldozer in our bear case, our fair value comes out at $4.50 per share.