Intel Xeon: One Step Forward, Two Steps Backwards
Contributor since 2020 / 9798 Followers
Summary:
- Intel’s recent data center webinar was well received by investors, as the stock rallied upwards. My own analysis of the event is more sceptical.
- While Intel moved one step forward with 18A Clearwater Forest in 2025 (an unquestioned leadership product), it also moved two steps backwards.
- Previously rumored to feature 334 or even 512 cores, these higher core count Sierra Forest-AP parts have been canceled in favor of the unambitious 144-core Sierra Forest-SP.
- Diamond Rapids, the 18A P-core counterpart of Clearwater Forest, once scheduled for 2024, is now missing in the 2025 roadmap, implying a delay to 2026.
- Despite these additional Xeon setbacks, investors may remain bullish given (1) the process roadmap remaining on track, (2) the low valuation/stock price, and (3) the three emerging businesses.
JHVEPhoto/iStock Editorial via Getty Images
Investment thesis
Intel (INTC) recently held its highly anticipated data center webinar, providing a roadmap update. However, contrary to consensus (as the stock saw a noticeable rally), the update was actually underwhelming, showing a lack of ambition to regain “unquestioned leadership”, which could lead one to wonder if the turnaround will actually be completed in 2025.
While this is not a complete blow to the thesis, it is another setback in a major business which Intel can hardly afford. Nevertheless, the stock price (even after the latest rally) remains markedly below where it was when the CEO transition occurred, so to some extent an incremental delay in the turnaround may already be priced in.
Overall, I remain bullish (mostly due to the stock price and process roadmap remaining on track), but investors should not forget Intel’s execution mishaps too quickly. For example, surprisingly, no one during the Q&A asked about the P-core sibling to the E-core Clearwater Forest product on 18A.
Background
The first in the series of investors webinars was held in January (delayed from December), where Intel made the case for a sustained 300Mu PC TAM, and also teased its upcoming Lunar Lake CPU for early 2025.
My previous coverage on Intel was about some roadmap changes in a different business segment, data center GPUs: Intel Announces New Major Product Delay, But Should Remain Competitive (NASDAQ:INTC).
One step forward, two steps backwards
The headline item from the roadmap update was Clearwater Forest, the second-gen CPU in the new Forest swim lane of E-core (previously called Atom) Xeon products, adjacent to the existing swim lane of P-core Xeons.
Clearwater Forest is scheduled for 2025 on the 18A process. This implies that Intel will keep a similar approach as the upcoming Sierra Forest on Intel 3 which will see an unusually short period between client (Meteor Lake on Intel 4) and server launch. Intel provided a narrowed launch schedule for the first half of 2024 (read: Q2’24), followed “shortly” afterwards by Granite Rapids.
Since TSMC’s (TSM) N2 process is only scheduled to go into production in Q4’25 (three years after N3), this means there is a high chance main competitor AMD (AMD) may not have a competitive part against Clearwater Forest until 2027, since just like N3 in 2023, the first year of a process ramp is usually reserved mostly for Apple (AAPL).
While this is one step forward, which is perhaps what investors focused on when deciding to bid up the stock, this is still neglecting the two steps backwards.
First, while it is perhaps reassuring that both Granite Rapids and Sierra Forest are hitting their engineering milestones with a firm 2024 launch, the main issue is that Sierra Forest is a part with a lack of ambition, which will be unable to achieve unquestioned leadership.
The reason for this is, as Intel revealed, it will have a core count of just 144, only slightly above AMD’s 2023 Bergamo part, which as a Zen 4 derivative will actually have hyperthreading, which means it will have 256 threads, which in turn suggests it may still quite favorably compete against Sierra Forest, despite Intel’s process advantage with Intel 3.
Of course, while this is just speculation for now, the main point here is that Intel’s announcement officially confirms that the higher core count Sierra Forest parts have indeed been cancelled. While this was already rumored, even as late as a few days before the webinar, the internet was flooded with leaks suggesting there could be Sierra Forest-AP CPUs with up to 512 cores. Also note that Granite Rapids rumored to feature up to 132 P-cores.
Simply put, for a company that has been trailing in the data center since 2019, it is unclear why it would cancel a 512-core part that surely would have competed vigorously with AMD’s 128-core Bergamo or even its next-gen 256-core Turin. Investors may remember Intel already did something similar last year when it postponed Granite Rapids from 2023 to 2024.
The second step backwards is regarding the mystery of the missing Xeon. A few years ago, a roadmap until 2024 showed that Intel, at the time, planned to regain unquestioned leadership with a part called Diamond Rapids, which was likely on the 5nm (20/18A) process. However, in the new 2023-2025 roadmap Intel showed at the webinar, there was no Diamond Rapids.
While the Granite Rapids delay already implied Diamond Rapids would be scheduled for 2025, the new roadmap indicates the part is now slated for 2026. Even if it launches in early 2026, this still implies a nearly two year delay compared to its original schedule, which is hardly an improvement over Sapphire Rapids.
Risks
To be sure, with two Xeons yet to launch before Diamond Rapids, there may be a slim chance that Intel simply wanted to focus on Clearwater Forest and those (other) Rapids first, so perhaps it may still add Diamond Rapids to the roadmap for 2025 in a later announcement. But as it stands, this Xeon is still missing.
In addition, note that even if Diamond Rapids launches in 2026, it may still be about a year ahead of any N2 Epyc products from AMD, and as the name implies Intel 18A should be competitive with TMSC N2 anyway.
Thirdly, while Sierra Forest may not have much more cores than Granite Rapids, it should be a much smaller, and hence much cheaper product. While there has been some debate about whether a certain wafer Intel showed was Sierra Forest, the wafer featured 480mm2 chips. Although it will also need two additional I/O chiplets, for comparison Sapphire Rapids (60 cores) comprises 1600mm2 of silicon.
Financial implications
Despite the current data center slump, Intel actually increased its TAM growth estimate. At the 2021 investor meeting, Intel was targeting DGC growth to accelerate to mid-teens growth. In the new update, Intel expects a CAGR in the 20s, which is based on the faster core count growth, and the higher ASPs Intel will charge as a result. (Intel stayed on 28 cores for 4 years due to the 10nm delay.) Overall, Intel expects the TAM to grow from $40B to $110B.
However, investors should (and perhaps must) be sceptical of these claims. Intel has been touting this ~15% CAGR for the data center since 2013, and has only managed to achieve this for one or two years, even during AMD’s absence. So to suddenly expect an incredible CPU boom in the next half a decade that will about triple the TAM is quite an extraordinary claim for which there is little (if any) evidence.
For comparison, in the 2022 investor meeting, Intel expected a TAM of $65B in 2026, meaning that despite the current slowdown in demand, Intel is quite literally doubling down on its TAM forecast.
Some part of this increased comes from AI, where Intel expects a $40B. AI is indeed and obviously a major growth driver for compute. But despite the recent ado about things like ChatGPT, AI is not new. In fact, in my very first Seeking Alpha article in 2020 (Intel To Lead The AI Revolution Over The Next Decade (NASDAQ:INTC)), I put AI forward as a major growth driver for this decade. However, in that article Intel claimed that already in 2022 the TAM would be $40B (although the slide said that 75% of AI would be at the edge).
Investor takeaway
Contrary to what seems to have been the consensus, the DCAI webinar was a let-down. Previously rumored were plans for an ambitious 334- or even 512-core Sierra Forest-AP. However, after Sandra Rivera was appointed to lead DCAI, she had spoken about roadmap simplifications, and the uninspiring merely 144-core Sierra Forest-SP that Intel has now touted seems to be the result.
Equally disappointing for those who closely follow the Intel rumor mill was the complete lack of Diamond Rapids on the 2025 roadmap, a part once scheduled for 2024. While Clearwater Forest will move to 18A in 2025 (and for all we know maybe Intel is planning a 1024-core part here), the mainstream P-core swim lane will therefor take yet another year to achieve leadership, missing Pat Gelsinger’s 2024-2025 timeline. (While Intel management has been bullish about Granite Rapids, it will have a lower core count than AMD’s original plans for Turin on N3, which is also in line with Pat Gelsinger’s 2021 comment about leadership moving “nip and tuck” for a few generations.) For those not keeping track, back around 2019, when Intel lost data center leadership, the biggest AMD bulls were expecting Intel to catch up in 2022.
In the bigger picture, combined with the recent roadmap simplification and semi-delay on the GPU side of the data center, arguably Intel’s execution in this business segment remains subpar. Nevertheless, there remain two items that offset these concerns. First, the underlying (accelerated) Moore’s Law process roadmap remains on track. Moving from Ice Lake in 2021 to Diamond Rapids in 2026 means two full nodes in five years. Secondly, the stock valuation also remains suppressed due to the current macro environment. Unless the long-term semiconductor outlook (currently expected to reach $1T around 2030) would be markedly adjusted downward, then a recovery in demand should be an additional boost for the business besides the turnaround and the three emerging growth businesses (Mobileye, GPU and foundry).
Analyst’s Disclosure: I/we have a beneficial long position in the shares of INTC either through stock ownership, options, or other derivatives. 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.
Seeking Alpha’s Disclosure: Past performance is no guarantee of future results. No recommendation or advice is being given as to whether any investment is suitable for a particular investor. Any views or opinions expressed above may not reflect those of Seeking Alpha as a whole. Seeking Alpha is not a licensed securities dealer, broker or US investment adviser or investment bank. Our analysts are third party authors that include both professional investors and individual investors who may not be licensed or certified by any institute or regulatory body.