Is this the Intel Inflection?
What would it take to turn around the stodgy giant?
Disclaimer: Please note that this newsletter is not investment advice. The information provided is for informational purposes only and should not be considered investment advice. You should consult with a financial advisor or other professional to determine your investment goals and risk tolerance. Any investment decisions you make are solely your responsibility.
I had this tweet a few months ago, and I think this is the most kickback I have ever gotten on something.
We are at that point with Intel that no matter what they say, the gut reaction is to scream it’s screwed forever. In comment sections and Twitter threads worldwide, Intel can do no right.
I am starting to warm up to the Intel inflection. Financial aspects are bottoming out, and there is a line of sight to a narrative change. Let’s talk today about the path forward at Intel. For the first time in a very long time, I think Intel might not be screwed. Does it outperform the semiconductor index? I have no idea. But I like the setup and want to call it like I see it.
Let’s start with the big incremental news. Intel said will be at the high end of their guidance this quarter. This was from the recent TD conference last week.
So in the real short term, so as we look at our quarter because some may have a concern as to whether this somehow is impacting how the first quarter or second quarter is shaping up. And we don't see it. In fact, actually, when you look at how things are trending, linearity has been great this quarter for us, which is obviously good in terms of how that will play out in terms of the revenue side but also good in terms of kind of cash collection.
We're shipping out a lot earlier in the quarter and that helps in the fact that we generally get the cash off the receivables within the same quarter. And we think, as we kind of look at how things are playing out for the quarter, that we will end up in the upper half of the range that we gave. So we gave a range of kind of $12 billion plus or minus $500 million. And we think we're going to track at $12 billion to $12.5 billion as we close out the second quarter.
What’s more, they have the dinkiest but real accelerator. Gaudi allegedly is better than A100, which is much worse than H100. But given the insane demand at Nvidia, Intel is getting some demand. They are growing Gaudi rapidly from a small revenue base.
We also have Gaudi, our accelerator offering and the performance of Gaudi is quite good relative to competitive offerings, not only in accelerators but also in the GPU space. We've seen pretty good pickup on our Gaudi line. In fact, I think when you look at our pipeline, just in the last 90 days, it's gone up by like 2.5x. Now still a smaller base that we're talking about but we are seeing momentum even in our business associated with AI on the accelerator front.
Intel also has a GPU and CPU hybrid. But so does Nvidia. Intel is the distant third place in this space for GPUs.
Gaudi, of course, initially. We'll have our GPU product, Falcon Shores, which integrates our accelerator and GPU offering into 1 product. That will be out in '25. We have our high-performance compute product, Ponte Vecchio, already in the marketplace. That's doing quite well.
They think they gained share, likely from steep discounts in the client segment against AMD.
And I think that -- the fact that we were able to kind of eke out actually a little -- even a share gain in Q1 does show that we're starting to turn the corner. Now I don't think that for the year that will be the story. There'll still be some share erosion this year. But the fact that the product is doing well in the marketplace, I think it's a good proof point that we are starting to turn the corner in the data center space. Emerald Rapids will be out in the back half of the year. And then we start to get into some products that are really competitive with others in the marketplace.
The most interesting part of this story, however, is gross margins. Remember, they have been making inventory adjustments every quarter. Now that they seemed to have turned the corner on revenue, the incremental gross margin will be much better than expected.
As revenue starts to come back, we get much better fall through and we get margin expansion. So a lot of it is just that. We just need to see the business come back. But there are a number of things that we're driving to improve gross margins incrementally beyond that.
They also think they will announce an 18A customer, which everyone believes will be Nvidia.
And so our hope is that by the end of the year, we'll be able to announce 18a customer that I think will validate 18a's performance and our ability to achieve that in the -- under the 5-node, 4-year calendar. The cost on 18a, just if you look at it kind of strip out like scale and all those kind of things, it's going to be extremely cost competitive with the other players in the marketplace.
But we do think just in terms of what we can -- in terms of delivering on the transistor technology in terms of delivering on what we call -- what [ Bradley ] has talked about is gate all around. We think that we will have a very competitive product offering on 18a.
From this article, it seems like Nvidia is seriously considering it. The 18A process, in theory, should be better than TSMC’s, but we are a while away from confirming this.
"You know that we also manufacture with Samsung, and we're open to manufacturing with Intel. [Intel CEO Pat Gelsinger] has said in the past that we're evaluating the process, and we recently received the test chip results of their next-generation process, and the results look good," Huang said.
Huang also said that Nvidia’s supply chain is designed for “maximum diversity and redundancy so that we can have resilience,” and “we manufacture in as many places as we can.”
Also, let’s take a second to talk about process improvements. This is a chart from the excellent Wikichip and the excellent work of David Shor. Intel 4 has transistor density parity and is pretty much baked. We are at the “catchup” phase, at the very least. If there ever was an Intel inflection, now might be the time.
The Setup and the Light at the End of the Tunnel
They are going to come in at the high end of guidance. We still have a few weeks left in June, so there’s a chance they have more of a revenue beat in them, given the linearity comment. But frankly, I think it doesn’t matter.
Remember the Nvidia quarter, where all the inventory reserves that have been written down led to a massive gross margin beat? I think that’s possible for Intel right now.
What’s more, it does seem like they are at the peak of utilization and ramping charges, with 250 bps of the gross margin reserves this quarter going away next quarter (3Q) and under-utilization charges being the remainder.
Yes. Sure, Tim. Let me give you a little color. So on the 250 basis points that you correctly highlight that are the pre-PRQ reserves, the benefit of that is as we go into the back half of the year, that's related to Meteor Lake and Emerald. And as those ship, a lot of that reverses. In fact, in essence, we end up shipping it at 100% gross margin. So that becomes a tailwind. And then we do expect, as things improve from a demand perspective, that we will start to load the fab back up.
They guided to ~37.5% Non-GAAP gross margin, which I think will increase. What’s more, the opex reductions are still happening, so there’s a chance that Intel will pull a positive EPS number this quarter. I have them at ~.07 cents versus -.04 for guidance.
I am a bit aggressive with the modeling, assuming that YoY growth turns positive by the end of the year and that next year they are growing ~20% YoY (on a weak 2-year stack), which seems aggressive. But if you squint, I can see a path to ~2 dollars in EPS in 2024. They said they should grow sequentially throughout the year, and I’m trying to say that I think the second derivative of Intel has inflected positively for EPS and that 1Q23 will be the trough.
The reality is estimates will likely not be this smooth, but I like this setup. Numbers are too low, the second derivative is turning, and Intel's EPS is accelerating by the nature of very easy comparisons, an improving market, and cost reductions. Historically it’s a good setup for stocks.
What’s more, I also want to highlight a bit of a narrative path. Remember the Wikichip article I posted above? There’s a lot of momentum around the 20A and 18A nodes, especially as the aggressive insertion of BSPDN will mean that Intel passes TSMC in density. Importantly look at the insertion of PowerVia (BSPDN); it will beat TSMC to market and should mean that by the end of 2024 if Intel can execute they will have a denser transistor process.
I understand that there are a lot of questions about Intel’s ability to execute, but if it’s ever been time, it’s right now. At the very least, the debate and story at Intel have a forward-looking view to debating PowerVia insertion in the process. Early indications (like the paper at VLSI Japan) will be hard to read, but if the market believes PowerVia will fix the process, that’s good for the stock.
If you’re curious about BSPDN, refer to this excellent piece I wrote last year. This inflection has been a long time coming, and it’s time. I wrote about it then and have been waiting until now.
Before I leave it there, I want to also talk about something that I think is interesting and incremental. AI and Intel. Hear me out; I think AMD is not as well positioned as markets would like to believe.
Intel Could get an AI Narrative.
AMD has been rallying this year because they are perceived to be AI beneficiaries as well, given that they make GPUs like Nvidia. They are the second-best GPU maker in the world, but as I discussed, AI is more than just the hardware; it’s the software, hardware, and networking stack. I called this the three-headed hydra of Nvidia.
I want to set some expectations because I think the AMD AI story could easily be the biggest bull trap I’ve seen. It’s no secret that the Kernel and software leave something to be desired, but I found this rant from George Hotz amazing. Watch the video; it will change your mind about AMD’s chances in AI.
Direct quote: “AMD’s Kernel will never be good for machine learning.”
Watch the first 18 minutes, starting at the timestamp in the video above. ROCM’s kernel seems “unfixable.” The market gives AMD the benefit of the doubt: they can catch up in GPUs as they did in CPUs. But the reality might be that AMD got to freeride the x86 ecosystem, and that ride won’t be available to Radeon. Radeon has to bring its software to the fore, which is not doing the job. The AMD MI300 might be a dead on arrival product for Machine Learning, and George even goes on to talk about how A770s (Intel’s GPU) is preferable for hobbyist machine learning because of software support.
Let me repeat that. Intel’s mediocre low-end GPU is preferable to AMD’s because of better documentation and a kernel that isn’t “worthless”. I don’t think I’ve ever seen a more out-of-consensus opinion be stated, and I want to caution everyone excited about AMD’s GPU.
To be clear, this is the opinion of one guy trying to offer a hardware-agnostic software platform for Nvidia, AMD, and Intel. AMD’s software efforts are laughable, or maybe Xilnix is trying to address it, and they have not launched their solution to the market. AMD is lacking one of the important legs of AI competition, software.
Intel, funny enough, has the software support to do it. And now they are starting to ramp Gaudi (their AI accelerator) in a way that surprises me seriously. STH reported that Intel will have a super cluster of Gaudi accelerators in the 10s of thousands of chips. This is a huge narrative violation, and frankly an angle that I think Intel could be viewed as an “AI winner”, at least relative to today. Today's stock has very, if not zero, expectations of AI victory. Intel has a better shot than perceived.
To be clear, Nvidia is miles ahead of everyone. But the desire and bidding for secondary players like AMD are real, and it looks like Intel has a process inflection and a potentially better AI product than AMD. It remains to be seen, but I am also seeing a narrative inflection for Intel, which could make all the difference.
In summary of why I think the Intel inflection could be happening right now:
Numbers should improve, and EPS should accelerate
EPS should accelerate long enough to put a bottom in the stock. Meanwhile, the process inflection of BSPDN is close enough (2024) for the narrative to shift to looking forward to Intel’s PowerVia node.
Intel also seems to have better software support for AI than anyone appreciates. The AI narrative could meaningfully improve there from effectively zero appreciation today.
This all happens in a “roadmap” of narrative and earnings improvements. Trough EPS > Inflection on horizon > Potential AI accelerators from Gaudi. That’s a way forward for the stock, and we could see a meaningful rerate for the sleepy giant.
I am here to say I think this is the bottom for Intel’s stock. There is a way forward, and numbers should improve. I do not know if this means meaningful outperformance, but I do think it’s not a bad stock anymore. There are execution issues to be concerned about, but every dog has its day, and it could be Intel’s day again soon. I will keep everyone updated.
BSPDN Beneficiaries
First, the company I am most surprised about, but it makes quite a bit of sense, is DISCO (6146). The original process people discussed for BSPDN involved a flipped wafer and bonding, but the core mechanism is grinding and revealing the metal under connects. I think this will be a huge market share and consumables driver, but I cannot quantify it quite yet; shares reflect this at 52-week highs.
Another beneficiary is Applied Materials (AMAT). It’s hard to know how many metal layers they take, but the clear bet is that more metal layers are added to the stack of semiconductors, and Applied is the key player in metal layer deposition. They will take more shares in BSPDN as this increases the layer count meaningfully. It’s hard to quantify, given that every company will execute this differently, but metal layers are going up and will eventually go up at TSMC as well.
Also, a fun note - I’m going to VLSI Japan next week! That’s where the Intel BSPDN paper will be presented, and I expect to have a public post about the BSPDN paper in the coming week. If you’re going to VLSI Japan, please hit me up. I would love to meet some of my subscribers; reach out if you’re around or from Japan. Talk then!
Links & References
https://www.anandtech.com/show/18894/intel-details-powervia-tech-backside-power-on-schedule-for-2024
https://fuse.wikichip.org/news/7375/tsmc-n3-and-challenges-ahead/
https://www.networkworld.com/article/3697858/intel-looking-likely-to-manufacture-nvidia-chips.html
great post Doug -vijay
I do sometimes here some positive stuff on oneAPI and SYCL. I’m not sure how good are those but I’ve got a feeling those software stack might be better than AMD’s😃