Ice Lake-SP Tells A Cautionary Tale For Intel Investors

Summary

• On the verge of its Unleashed event with historic Intel Foundry Services announcement, Intel's latest event featured the launch of its Ice Lake-SP server CPUs.
• Ice Lake-SP has been delayed by at least 1-2 years and continues to trail behind the competition significantly. This CPU embodies everything of the failed Intel in recent years.
• In 2019, Intel touted Aurora as the first 2021 exascale supercomputer. However, the 7nm GPU was already delayed, and the 10nm Sapphire Rapids CPU is also delayed into 2022.
• The Unleashed event was about the future, but the Ice Lake launch represents the present. Intel is not yet back.

Photo by JHVEPhoto/iStock Editorial via Getty Images

Investment Thesis

Following incoming CEO Pat Gelsinger's recent Intel (NASDAQ:INTC) Unleashed event, and the general stock rally since his appointment, Intel is looking to carry forward this momentum with the Ice Lake-SP data center launch. For example, one comment that was expressed at the Unleashed event was "Intel is back".

Make no mistake, this new launch is highly impressive, the largest generational upgrade in probably almost a decade. However, the cautionary tale with Ice Lake-SP is that it is not launching when it was supposed to, having been delayed by many quarters, even though there really is no clear reason for the delay (as explained below, 10nm doesn't seem to be the main reason).

Overall, then, the importance of the Ice Lake-SP launch (and its performance improvements which may drive a large upgrade cycle) can't be understated, but the many delays shouldn't be forgotten as quickly either. Intel is still not yet back.

An earlier version of this article was titled "Ice Lake Embodies Everything Wrong With [The Past] Intel". Perhaps that was a bit of an exaggeration: as the recent management changes indicate, Intel is still very much in transition. From that view, the launch should remind investors to not expect this transition to happen overnight. From a stock perspective, that often means there could be a pullback in the future for a more reasonable entry point.

Analysis

Investors Meetings often represent the most bullish management outlook about the mid or even long-term future of a company. As such, they can be a reasonable benchmark to compare a company's actual performance against. This article applies that approach to Intel.

Going back to Intel's 2019 investor meeting, and the event showcased several key announcements across Intel's major organizations, including the PC, process technology (7nm) and its data center roadmap. By now, the outcome of that meeting should be known to most Intel investors: while Intel actually did meet its PC and financial commitments, Intel significantly failed on the data center and process technology sides. While Intel's 7nm progress determined most of the company's stock performance in 2020, the data center delays should not be underestimated either.

For a brief refresher, Intel had promised Ice Lake-SP for Q2 of 2020. That's right, a full year earlier than the recent April 2021 launch event. (Of course, given the 10nm delays prior to 2019, arguably Ice Lake should have launched another year earlier, in Q2'19.) Intel's former chief engineering officer Murthy quite infamously said that Ice Lake-SP was planned to be a quick follow-up to the PC client Ice Lake CPU, which launched in Q3'19. However, already in 2019 it became apparent that Ice Lake would shift to the second half of 2020. For several months, Intel also kept denying further rumors of a delay into 2021, but the launch event unambiguously was titled "How Wonderful Gets Done 2021".

In short, Ice Lake-SP has quite a storied history for what ideally should have been "just another" CPU launch. This is the reason for the title of this article: Ice Lake-SP embodies everything that has gone wrong with Intel in the last multiple years.

As biggest evidence of Intel's data center fiasco, note that the 10nm fiasco itself cannot even fully account for the multi-quarter Ice Lake delays: all evidence clearly suggests that Intel's 10nm has, in fact, progressed on par or even ahead of Intel's (low) expectations since 2019 and even 2018. In other words, 10nm in all likelihood isn't even the reason for the push out of Ice Lake into 2021. In fact, in January Intel's CFO even acknowledged that he didn't expect Ice Lake to affect Intel's gross margins as much given the progress with 10nm.

As an outsider, one might reasonably expect that Intel would have been scrambling to gets its latest CPUs such as Ice Lake out ever since AMD (AMD) launched its game-changing Rome CPU in mid-2019. (Meanwhile, indeed, AMD already launched its next-gen Milan CPUs recently, although those were a rather small evolution.) Consequently, the next-gen Sapphire Rapids Xeon has also been delayed, obviously, and will likewise miss its original 2021 target.

Here's an apt example of the knock-on effects of Intel delaying its entire data center roadmap. As some might know, one of the side effects of the 7nm delay was that the Aurora exascale supercomputer would be delayed from Q4'21 into 2022. Aurora will combine Intel's Sapphire Rapids on 10nm with Ponte Vecchio GP-GPU on 7nm. However, the latest delays on the CPU side actually provide another reason for the delay: Intel will be unable to deliver neither its CPUs or GPUs.

In short, with Intel already being so much behind, these latest delays are just the nail in the coffin with regards to Intel's data center execution. Remember that Intel already planned for years to become a data center-first or data-centric company ("the data center is the new PC"), so these execution issues should be quite concerning for investors.

Ice Lake-SP

Having spent the entire article so far beating down on Ice Lake, a few words may be in order about the actual CPU and its technology. Because for all of the bearish sentiment so far, there is actually a lot to like about this CPU. (The bearish evaluation, as discussed, is simply a function of its late launch date/almost endless delays showing a lack of execution.)

For starters, it of course leverages the same (decent) architecture as in the client Ice Lake, Tiger Lake, Lakefield and Rocket Lake CPUs. In other words, this is a tried and tested (and delayed) architecture. Equally (or even more) important for the data center, core count is significantly boosted from 28 to 40. This is responsible for the >40% performance improvement Intel claims, and confirmed by reviews, and in some cases up to nearly +80%.)

Additionally, to feed the beast, interconnect is upgraded from PCIe 3.0 to 4.0, and the amount of DRAM memory channels is increased from six to eight. (Intel has even touted that it used some additional tricks to improve memory bandwidth by more than the increase from six to eight mathematically suggests, which AnandTech confirmed in its review and considered impressive.)

Let's also remember that Intel still has several major underappreciated technology advantages. While the usefulness of AVX-512 continues to be debated by some (at least on the PC side), this technology can be broadly used in the data center. The two most common use cases are HPC (number crunching) and AI. Intel calls the latter DLBoost. In these workloads, Intel will crush AMD since the performance per core is at least 2x higher: for AVX-512 workloads, Intel's technology advantage is effectively akin to having an 80-core CPU against AMD's 64 cores. There is in fact another quite important class of workloads where Intel has gained a similar advantage with Ice Lake, which is crypto, which has received speed-ups of 2-5x or even more. Thirdly, Intel still has its Optane memory as a differentiated technology for its platform.

The slide below shows what exactly happens when Intel's 40-core Ice Lake-SP can use its AVX-512 engine (for crypto, HPC and AI workloads) against AMD's 64-core Milan (a few credits should go to Intel for benchmarking against Milan so shortly after its launch).

Note that AI has become the most important class of workloads in the data center, and Intel achieves here obscene speed-ups of over an order of magnitude against AMD. Intel said that its geomean advantage was +50% performance in a set of 20 real-world AI workloads, and in the same test suite also beat Nvidia's (NVDA) A100 by 30%.

Lastly, on the security side, Intel is achieving parity by introducing Total Memory Encryption. Intel is also expanding its other security feature (which AMD does not have), called SGX.

In summary, while Intel called Skylake-SP in 2017 one of its biggest Xeon updates ever, Ice Lake-SP is certainly no less of a significant update. That is the biggest bull case for this CPU: since Intel still has >90% market share, Ice Lake-SP represent a significant upgrade for most data centers, which are obviously on the Intel 14nm platform. There are likely going to be quite a few upgrades in data centers because of Ice Lake (and Sapphire Rapids next year), which may drive upside data center revenue. However, the delays mean that it is launching almost four full years after Skylake-SP - despite an increasingly competitive environment.

Further discussion

In the grand scheme of things, which is Intel's investment prospects, the data center (i.e. not the PC) is Intel's key business for growth. As such, one would or should expect Intel's most bleeding edge technology to launch in this vital segment. However, as Ice Lake-SP demonstrates (with its 18 months delay after client Ice Lake) this continues not to be the case.

As such, there may be two schools of thought regarding the latest Ice Lake-SP launch. On one hand, as for example expressed by leading data center outlet Serve The Home, Intel in one fell swoop has gone from being hopelessly behind, to actually being a valid competitor again. On the other hand, as this article notes, Ice Lake-SP kept being delayed quarter by quarter even though Intel kept reiterating that its 10nm process was either in line or ahead of schedule.

Ideally, Intel should have simply canceled Ice Lake-SP and make sure to launch its successor Sapphire Rapids in 2021: since Sapphire Rapids consists of much smaller ~15-core chiplets (Intel's long-anticipated disaggregated design), it should actually have both higher yield (lower cost) as well as higher performance.

As a reminder, the simple Intel thesis is that demand for its CPUs and other silicon, mainly in the data center, will keep increasing over time driven by secular trends such as cloud and 5G. So overall, AMD being ahead by 10% or even 20% in benchmarks won't be devastating for Intel; that likely won't cause too much market share declines to offset the market's (and hence Intel's) general growth.

That is exactly what Intel has delivered. Before this launch, Intel was often behind by 50% or more. (And for some workloads, Ice Lake-SP is actually ahead by a large margin, it can really depend a lot on the specific workload of a customer.) However, incoming CEO Pat Gelsinger has stated that Intel's goal is to obtain absolute leadership in every category. That is still far from the case, and Ice Lake-SP's careless delays have postponed Intel's return to leadership in the data center by at least another generation.

Financial outlook

As was noted by many at the time, Intel's January earnings were perhaps the first time Intel really acknowledged the impact of competition on its data center results (in the past, Intel would merely say it expected an increasingly competitive environment, or that competition was not more than it had expected). That validates the numerous comments in this article that Intel needed to launch Ice Lake-SP multiple quarters ago.

Intel expects the same competitive environment to continue into 2021, but as also discussed Ice Lake does improve Intel's competitiveness tremendously. As such, though, offsetting continued slow market share declines might actually be an upgrade cycle from customers moving from Intel's 14nm to its new 10nm CPUs.

Therefore, overall I remain quite neutral on the near-term data center prospects. Shown below for reference is Intel's data center performance (the red line is TTM revenue).

To guide the eye, what investors really should be looking at is the bottom of the 2020 cycle (in 20Q3), which at ~$6B is a full $1B higher than the bottom of the 2019 cycle (although the upcoming Q1 results should confirm if the bottom really has been reached). This should be indicative of quite decent growth.

Risks

As explained at length, the main risk was what didn't happen: the CPU's actual launch due to the countless delays. However, with Ice Lake-SP itself being a reasonable CPU and its successor Sapphire Rapids already sampling, this hence reduces the risk of Intel's forward-looking data center execution. Better late than never, one might say.

Additionally, looking past Sapphire Rapids, and Intel recently announced (or at least indicated) that it is planning to launch a data center CPU manufactured at TSMC (TSM) in 2023. In short, it took Intel seemingly forever to get Ice Lake (10nm) into the data center, but going forward the roadmap actually seems much more straightforward.

Investor Takeaway

On the heels of the Intel Unleashed event in late March, Intel is launching its Ice Lake-SP, hoping to further regain investor confidence with the "Intel is back" mantra. However, while the former event certainly delivered in big way regarding Intel's future, the same can't be said about the present, which is the Ice Lake-SP CPU.

Once touted by management as being a "close" follow-up to the client Ice Lake CPU, it should be noted that Intel's PC group already launched its successor Tiger Lake half a year ago. As described in this article, the delays are quite inexplicable given that they can't (fully) be ascribed to the 10nm delays.

The best way to view these delays is to look at what has changed competitively because of Intel's lack of data center execution. Instead of the recent "How Wonderful Gets Done 2021" launch event being the launch event for Sapphire Rapids, which very realistically could have overtaken AMD, Intel merely launched Ice Lake-SP. (As mentioned, this also means that Intel can't even in principle deliver the CPU for its much-touted Aurora exascale supercomputer, as the 7nm Ponte Vecchio GPU was already delayed.)

Hence, for forward-looking investors, the question is if they are willing to ignore this launch, which embodies the Intel of the past, and instead look at the announced future changes by incoming CEO Pat Gelsinger.

For this author, the answer is that as long as Sapphire Rapids isn't further delayed (which doesn't seem to be case, as it is already sampling), then Intel will have another large upgrade within just a year or less from now. I called Ice Lake Intel's largest upgrade in a decade, and I expect Sapphire Rapids to be anything but less than that as well; two such product launches in relatively quick succession should put Intel back on the map (from a long-term view).

This is also what the stock shows: the market has given Intel some time due to the CEO transition. Though, at this point considering the amount of time Intel will need to really get back to the top of its game, perhaps it would make more sense to await any pullback for an entry position.

Analyst’s Disclosure: I am/we are long INTC. 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.

Comments

[Kwan-Chen Ma]

yes, a good history telling which has been priced in. Would like to see more finance from your good articles.

[Just Some Guy]

Nice rant, but accuracy, eh. A couple of points. Chiplet design is not free, while you presumably manage more yield on the chiplets, getting them wired together is a rather hideous process and is expensive and causes yield fails, and then may impact performance significantly, too. Also the extension to 40 cores is pretty much spec sheet and benchmark engineering, Intel has always been able to build all the cores anyone could want, but high core counts don't always work well in servers running complex workloads, especially on Windows environments. Licensing gets in the way, cache management, core consistency, not to even mention (cough cough) security. Yes it's all been delayed, which is both more and less significant than generally made out. I mean, if Intel had met all schedules, would the stock be at $200 by now or what? Finally there is some question about the volume of 10nm chips making it to customers, is it limited by sales or production, and is it enough to impact revenues and profits and margins, all in good directions? Somewhat yes, but I suspect (from some comments elsewhere on the Interwebs) there are still serious yield and production bottlenecks, even now, and maybe small prospect of them going away. As I always finish: it's just the same at TSM by the way or AMD stock would be $200 by now, too.

[The Real Idiot Savant]

"Note that AI has become the most important class of workloads in the data center" That's not true for any of Cloud, Hyperscaler, HPC or Enterprise DCs. It's an important emerging segment, but certainly not the most important for these yet. This hurts the credibility of all other assertions in the article.

[The Stock Stooge]

The discussion seems to be centered entirely on Intel vs AMD x86 data center chips. Seems to me that the bigger issue is M1, ASICs, Google and Amazon's chips, etc.

[The Real Idiot Savant]

@The Stock Stooge Long term and possibly even medium term yes. But the inertia resisting major change in datacenters is immense -look at how hard achieving market share even WITHIN the x86 segment is for AMD. The question is will the rate of ARM (and possibly RISC-V) alternative improvement sufficient to vault over x86 even with a fierce performance war going on in x86 that is driving massive yearly improvement there? Big hyperscalers like Amazon that can build their own alternative architectures like Graviton have many incentives to do so beyond the cost-effectiveness of home-grown silicon for the specific areas (AWS can't make all their customer instances run on ARM): like using these possible alternatives as a price cudgel on Intel and AMD for the x86 processors they do need to buy.

[The Stock Stooge]

Is $20 billion realistically going to get Intel back to leadership in foundry? I would be more bullish if they eliminated the dividend and went with $50 or $100 billion. $20 billion seems insufficient but who knows.

[wow&wow]

@The Stock Stooge

"... get Intel back to leadership in foundry?"

Get back? When did Intel have a leadership in foundry before?

Where and how will Intel be able to get enough design services engineers to be "a" successful foundry?

U.S. has lots of "computer science" engineers that aren't helpful in silicon foundry.

[Spike Capital]

@The Stock Stooge Government will come in and help Intel and American domestic semi manufacturing. Do you think US will allow non-US companies to dominate?

[RUBYRUBY3]

@Spike Capital
"Government will come in and help"
Now that is exactly what socialism is, take money from the taxpayer and give it to the govts pet projects.... I know it when I see it I'm in Canada ;>

So much for free markets too ...." not allow non US companies to dominate."
Free markets are the best of US inventions, if another pulls ahead then get your shit together and beat them back with better products don't run to the Govt for handouts/ tariffs / tax breaks/ law changes to make your crap profitable.
Free markets are brutal to bad products, and they should be

( Bit of a rant, sorry ) but I see this in Canada all the time and it pisses me off

[WOLF 57327]

The bottom line = Bullish on Intel!

[flyfishinginWY]

Hey Arne, how would you like to make a cash bet that AMD launches Genoa before Intel launches Sapphire Rapids?

[C185]

@flyfishinginWY why bet? As I pointed out below Intel has given you the date. What date has AMD given for Genoa?

Also did they show support for CXL?

[UmbertinaV]

Sapphire Rapids is not delayed at all - It's scheduled for 2023. If it's launched in 2022 it will be absolutely on time

[C185]

@UmbertinaV On the Q4 earnings call Swan stated the following, which means they will be in production of Sapphire Rapids at the end of 2021 or beginning of 2022. The same sequence occurred with Ice Lake. They achieved product qualification at the end of 2020 and in early January started ramping production and shipping. Intel's formal launch events happen after they start ramping volume and shipping.

"We will qualify Alder Lake desktop and notebook for production and begin our volume ramp in the second half of 2021 and we expect production qualification of Sapphire Rapids at the end of 2021."

[wow&wow]

"A Cautionary Tale For Intel Investors"

Why is Intel going to have 8 traditional core plus 8 small cores in stead of straight 16 traditional cores to keep things (like logic design, physical design, OS, ...) simple?

[jayn]

@wow&wow looks like a direct response to the Apple chips; however, it could also turn out that the small atom cores will be used to offload high speed datapath operations, as they are used in the p5900 family chips.

[wow&wow]

@jayn

Apple isn't the #1 CPU leader in PC and Intel isn't in smart phone business, so Intel response to Apple for what?

If Intel's process is competitive or good enough, there is no such a need for reducing die size and power consumption.

[hwmook]

@wow&wow

To reduce their crazy power consumption. Rocket lake is using twice the power of Ryzen 5000 to get the same results. That's crazy power consumption, even tigerlake also lose out in power efficency to ryzen 5000. Intel is trying to use the smaller core to conserve more power to prepare themselves for AMD+TSMC 5nm and 3nm chips.

[ULS]

@author Before I begin writing, I should mention that I have very high regard and respect for prior analyses you have written on Intel. But IMHO, this article was clearly the worst you have written on Intel lately. I found it disappointing because in prior articles you had made your theses much clearer and easier to understand. Here are some reasons why I rate your article so poorly:

#1: Crying over spilled milk usually gains neither writers nor readers anything. All prior delays of this processor are “over the dam”. They happened and they have been priced into the stock. What natters now is the future and not the past.

#2: your point that Intel has execution issues is well taken. I am hoping that Intel is addressing them now. Particularly with Gelsinger on board. Since you offer no insight at all that I could discover, on how well these issues are getting resolved, I suggest you refrain from beating a dead horse even more. Instead I suggest you speak up again, once you can offer meaningful information and conclusions on the subject of execution issues.

#3: as far as performance of this processor goes, it seems to offer a lot, according to your remarks. Since you cannot count your readers to all know as much as you do, I suggest you write a performance summary one can draw conclusions from. And not one that appear to vacillate back and forth.

#4 (my final remark): since all second-guessers who write “analyses” on Intel without sufficient knowledge available to themselves, have a very difficult job producing anything really worth reading, when in doubt, a better idea is – IMHO – to refrain totally from writing an article, until t can offer truly plausible and actionable conclusions.

And I respectfully claim that this article of yours has not reached this stage yet.

[Hamburgers]

For all the handwringing over the "late" launch, this article's discussion of Intel's inherent advantages only reinforced the idea in my mind that AMD's best days are behind it, and that Intel's valuation underappreciates this fact.

[TxSkibum]

@Hamburgers I guess their comm of expected >25% revenue for the next few years counts to u as it’s best days behind it’?

Intel is still capable to do well. But AMD will continue to do well either way, make no mistake. To deny that makes me question how much u really know about the semi industry.

[C185]

I think the 10nm issues can be pretty well attributed to Intel not using EUV at 10nm because EUV wasn't ready on the original 10nm schedule. I don't see extrapolating that to 7nm, as Intel is clearly going to use EUV for 7nm.

Also the "delay" between Ice Lake mobile and server appears quite normal. My recollection is about 18 months is the usual period between mobile/desktop and server releases. Server processors and platforms are much more complex, and require extended periods of validation and qualification.

Just to cite someone other than my recollection, here's Ashraf Eassa who used to contribute on SA. (Note his prediction was correct on Broadwell-EP. It did launch on March 31st 2016, 18 months after the PC client skus.)

www.fool.com/...
[While Intel cites its manufacturing advantage over competitors as a key enabler of its performance and power advantage, it tends to bring its server chips to leading-edge technologies far later than its PC chips. For example, the first 22-nanometer FinFET Intel PC chip launched in April 2012, but the first 22-nanometer server processor did not arrive until September 2013 -- about a year-and-a-half gap.

Now, if we assume Intel will launch Broadwell-EP (14-nanometer server chip) in early 2016, as rumored, then this would keep the gap between first client product (around October 2014) to first high-end server product (roughly March 2016) at about a year and a half.]

[Tony215]

I believe the government will tell TSMC to make more fabs in the USA because they always want this is is a serious dent in Intel future fabs. Intel no longer have knowledge making advanced fab only use for older nodes

[zisdead]

@Tony215 hilarious. Yeah the US government wants Taiwan (china) to build more fabs (you could not make this up)

Keep up! that ship sailed Intel is riding off into the sunset with 2 fabs in the US and an upgrade in Ireland with most of the funds coming from government grants for the exact reason that it doesn't want any more TSMC centralised chip manufacture.

[User 52289627]

www.cnbc.com/...

[zver11]

There is a good reason for the delay of the high core count server chip. AMD design is based on chiplets. The rate of defective chips produced does not climb with core count. Because Intel is using a monolithic chip, the defect rate climbs rapidly with the number of cores. Producing a 40 core version had to wait for their 10nm fab to reduce its defect rate to the point where 40 core chip yields could be produced profitably. Note that TSMC starts their 5nm process producing extremely small chips(Apple) while the early defect rate per wafer is high(mathematically, more usable chips as cutout defects if making smaller chips). Until Intel develops an answer to infinity fabric, it will have trouble keeping up with TSMC as it will always be trying to produce large chips while battling the high rate of silicon defects in early production runs.

[hanging chad]

@zver11 Sapphire rapids is not monolithic. It uses 4 dies interconnected with EMIB.

[C185]

@zver11 the "Infinity Fabric" is way behind Intel's Foveros and EMIB interconnect technologies.

[zisdead]

@zver11 Infinity Fabric is just the twenty year old Hypertransport protocol.It is streets behind Intel :)

"HyperTransport (HT), formerly known as Lightning Data Transport (LDT), is a technology for interconnection of computer processors. It is a bidirectional serial/parallel high-bandwidth, low-latency point-to-point link that was introduced on April 2, 2001"

"Infinity Fabric (IF) is a superset of HyperTransport"

en.wikipedia.org/...

[Fooly Finance]

Considering when Intel finally launches it's dated 10nm part Sapphire rapids in 2022 it will feature up to 56 cores and lag the performance of AMD's EPYC Milan which is a 7nm part featuring 64 cores.

Worse yet for Intel, Sapphire Rapids will be launching at the same time as AMD's new 5nm Zen 4 EPYC Genoa part featuring up to 96 cores.

Third party analysis indicates Intels "new" Ice Lake servers perform more in line with AMD's EPYC Rome which launched in 2019. EPYC Milan is currently running circles around Ice Lake, crushing it in all meaningful tests by more than 50%. In fact, Intel's new release even lags Ampere's ARM chips significantly despite retailing at half the cost of an Intel server.

Intel will likely trade down or flat over the next 3-4 years as they absorb losses in market share across all of their product lines due to massive technological shortfalls while their competition continues to innovate and execute.

[C185]

@Fooly Finance I'm pretty sure your "third party analysis" are a couple of websites that don't understand real world server workloads, or what features server buyers value.

www.channelfutures.com/...
“We consider confidential computing as an extension of our general-purpose fleet,” Russinovich said. “And on top of these new Intel processors, we’re also going to be offering new general-purpose, as well as large-memory virtual machine sizes that offer better price to performance ratios than any virtual machine we’ve ever had in our fleet.”
- Microsoft Azure CTO Mark Russinovich at the April 6th Ice Lake launch event. (Note: Azure has already deployed Milan instances so Milan is included in his comparison.)

[jayn]

@Fooly Finance I didn't know AMD added Optane support in Milan. When did that happen?

[RUBYRUBY3]

@jayn
This may help ( rumor so far )
videocardz.com/...

[willcousa]

Perhaps intel is sandbagging to increase profits. Isn't that the thing that intel owners have dreamed about?

[edca99]

AMD's EPYC didn't even get more than 10% of server share after a few years of Intel's delay. Now that Ice Lake is out, and Sapphire Rapids is sampling, and with top tier CEO Pat for faster and better execution, AMD free run seems to be over.

In the GPU ground, with DG2 coming out for gamers and may have aggressive pricing to gain some ground and take away the ridiculous margin from AMD and Nvidia. And for HPC Ponte Vecchio might be killing Nvidia's A100
www.reddit.com/...

More into future, Lunar Lake seems revolutionary and exciting. Intel's back.

Disclosure: long INTC

[V_uniqueacc]

“Intel is not yet back.” yet the share price is getting closer to $70 with a very reasonable PE. Imagine the upside when INTC “does get back”. Thanks for the bullish article

[corvid]

Thank you for this information on INTC.
"The slide below shows what exactly happens..." No, it is illegible.
Do you have a link to a higher resolution version?