Relax, AMD Will Remain Competitive Over The Next 5 Years

| About: Advanced Micro (AMD)
This article is now exclusive for PRO subscribers.

Summary

AMD bears emphasize Intel’s competing roadmap and how it may impede AMD’s durable competitive advantage over the next five years.

However, indications on node shift, CPU redesign implies AMD can remain relevant/competitive over the next five years.

Therefore, investors may be overestimating the competency of Intel, and may need to re-examine AMD’s process/design roadmap.

From what I can gather, there’s a reasonable path to 7nm volume production at Global Foundries, implying 7nm AMD chips in 1H’19 or 2H’19.

This would surpass Intel’s timeline, and keep AMD competitive, as it introduces stop-gap technologies to keep Zen architecture relevant up until 2020.

The primary counterargument to the AMD (NASDAQ:AMD) investment thesis is the sustainability of AMD's product roadmap over the next five years. Analysts argue that AMD could struggle to keep pace with Intel (NASDAQ:INTC) over the long run, as history is on the side of Intel and not AMD.

However, it's worth noting recent delays over at Intel with regards to 10nm, as Krzanich mentioned at its 2017 investor day that volume production will be delayed to first half 2018 with limited availability in Q4'17.

It's worth fleshing out the product roadmap and whether AMD can keep within respectable striking distance to Intel.

Filling in the holes of AMD's product/fab roadmap

A key source of concern for AMD investors/analysts is the process roadmap. From what I can gather, there's confirmation that GlobalFoundries is skipping the 10nm node in favor of 7nm. Various reports suggest GloFo is targeting volume production of 7nm chips in late 2018. This would imply 7nm AMD chips hitting shelves mid-2019. This sounds aggressive, but there's a technical workaround that GloFo is pursuing to get there marginally quicker (perhaps 6 to 12 months ahead of Intel's schedule).

I'm expecting Intel to start volume production of 7nm chips by the end of 2019, and release chips within 3 months to 6 months of volume production. This is driven by the 3-year construction cadence of fab 42 located in Chandler, AZ. I'd expect the fab to be fully operational by Q4'19 (under an aggressive scenario). Worst case scenario, Intel's fab starts volume production in 2H'20, which would give AMD an impressive runway to sell 7nm chips, as Intel would need to delay launch until 2021.

We should expect higher transistor density at Intel's 7nm node, because it's likely to run on either a 3rd or 4th generation FinFET whereas AMD is running off a 2nd generation FinFET on GloFo's DUV (deep ultraviolet lithography) process. Intel's plant buildout matches an early transition to EUV (extreme ultraviolet lithography), as Fab 42 overlaps with ASML's projected 2019 EUV launch.

GloFo will use a stop-gap technology, i.e. deep ultraviolet technology, which is more expensive given triple and quadruple patterning, but is quicker to deploy. GloFo is expected to retrofit facilities by the second half of 2018 with re-use of pre-existing fab equipment, phased-in CapEx and use of existing lithography/patterning tech the main reasoning behind a quicker 7nm deployment.

AnandTech summarizes the technical implications of GlobalFoundries 7nm roadmap:

While GlobalFoundries' expectations for performance, power, area improvements look solid, it should be noted that right now the contract manufacturer is among what's become multiple companies to have confirmed their intentions to pursue a DUV-only 7 nm process technology.

DUV (deep ultra violet) in this respect is the tried and true approach, however in order to create 7 nm features it will require using triple/quadruple patterning, which greatly increases design and manufacturing costs as well as cycle times over previous-gen nodes. So, while there is a lot of interest in using EUV if it works, many of the major fabs are starting at the same place as GlobalFoundries and at least internally preparing for DUV-only, while hoping EUV will be ready when they want it.

Assuming the transistor shifts happen on schedule for both AMD and Intel, I'd imagine AMD having a slight marginal advantage with its 7nm process versus Intel's 10nm. I believe AMD can bridge its product roadmap between 2018 and 2019 with the launch of Zen 2.

I'd expect Intel to respond more effectively in 2018 with volume shipments of Cannon Lake hitting the PC channel sometime in Q2'18. During that period, AMD's perf/watt advantage will diminish, but not as badly as prior cycles.

AMD will launch second generation Zen in 2H'18, making it in time for holiday quarter. The various families across notebooks, desktops, servers paired with the APU lines that fit into those three buckets are difficult to track. But, it's worth assuming AMD will sell three product families between MPUs and CPUs using second gen. Zen architecture before moving to a 7nm core.

Once AMD transitions to 7nm by 2H'19 (that's assuming GloFo gets everything together), and AMD finalizes the design of Gray Hawk (7nm APU) by 2H'18, AMD will have a design and process advantage when compared to Intel's 10nm CPU family Kaby Lake, and so on.

I'm expecting Zen 3 to be patterned out of the 7nm fab. It would be a shrunken down and denser package of Zen 2 as opposed to a meaningful CPU-design revamp. My guess is that Zen 2 and 3 will operate on similar architecture and would be modified to perform better at higher voltage/clock speeds. Base clocks of the current Zen family seem modest compared to prior generations, and could use better voltage optimization at higher power envelopes with better power leakage at the lower TDPs. These features can be introduced within Zen 2, and node shift to 7nm provides enough headroom to label Zen 3 a decent successor.

This gives AMD up until 2020 to launch a completely revamped architecture for its CPU lineup. This would correspondingly match management commentary of three successive Zen architectures; keyword from Mark Papermaster was "architectures," not fab process tech.

Understanding the APU roadmap

Raven Ridge is expected to launch in 2H'17, which is going to address the entry level, mid-end, and high-performance laptop market. The APUs are expected to operate on a TDP band of 4W to 35W. Raven Ridge will come equipped with both Zen and Vega on package, and it's not yet clear whether AMD will refresh Raven Ridge with a mid-cycle refresh, or will move the family to Gray Hawk.

Gray Hawk is AMD's codenamed next-generation APU (accelerated processing unit). We're expecting die shrinkage in the third-generation Zen CPU, which will drive performance/watt, and will come equipped with AMD's Navi GPU architecture, which is what follows after Vega GPU architecture. The Vega series is expected to fill in the roadmap from 2017 to 2018, and will be fully-refreshed under the usual 2-year cadence GPUs typically operate off of.

Xbox Scorpio will use a reference design of Polaris, and doesn't come with a Zen core. The performance improvement will be meaningful, but I don't expect consoles to exhibit significant performance gains up until the launch of Gray Hawk.

Gray Hawk will use a 7nm process (assuming it's ready by 2019), and is aimed at notebook and console refresh in 2019, which assumes a 6-year refresh cadence from the original launch of Xbox One and PS4 - AMD's next major semi-custom product will arrive in Q3'19, which would match holiday season notebook/console refresh.

I believe Raven Ridge will remain a notebook specific part, whereas Summit Ridge will remain a mid-to-high-end APU product for desktops. Banded Kestrel is the first desktop-specific APU part for the sub-$50 desktop CPU category (where Sempron typically operates), whereas Horned Owl will replace Summit Ridge using a 2nd-gen Zen part for the AM4 desktop platform in 2H'18.

The timing of AM4 Desktop APUs is difficult to ascertain given the codenames. But, I'd imagine Ryzen series desktop CPUs to sufficiently address the desktop market between $90 to $600 price tiers, whereas Summit Ridge APUs for desktops will remain comfortably in the enthusiast category until AMD launches an APU that's targeted at entry-level computers (likely Banded Kestrel).

Final thoughts

Given what I understand, AMD will remain competitive over the next four years as the introduction of new architecture and nodes helps to keep pace with Intel. Though there's some subjectivity, especially if GlobalFoundries is delayed by a year, then AMD will end up launching both its next-gen architecture and 7nm CPU at about the same time as Intel. This would be perceived negatively, but those concerns really won't emerge over the immediate two-year investment timeframe I'm operating under.

Though, there's plenty of negativity embedded in sell-side estimates, the bar is set extremely low for AMD going forward. Analysts aren't anticipating AMD to transition to 7nm by 2019, as that would imply a quicker 7nm ramp than Intel. However, I believe it's within realm of reasonable possibility given the technical implications discussed earlier.

I continue to reiterate my high conviction buy recommendation on AMD.

About Cho's Technology Research

Cho's Technology Research is SA's premier technology research package that gives investors and traders an additional edge when investing into tech companies. It's a trade publication, research service and an idea generator. It includes short-term trade ideas that yield 5%+, long-term small cap investment ideas that yield 25% and best-in-class data/news on tech sector themes. Subscribers get an early preview on Cho's price target/recommendation changes.

These services comes at a low annual subscription cost of $20/month. The promotional price period will end soon! Subscribers get a free two-week trial upon checkout and can cancel at any point in time. For more information click here.

Disclosure: I/we have no positions in any stocks mentioned, and no plans to initiate any positions within the next 72 hours.

I wrote this article myself, and it expresses my own opinions. I am not receiving compensation for it (other than from Seeking Alpha). I have no business relationship with any company whose stock is mentioned in this article.