If you follow the semiconductor market, you likely read about Intel’s Q4 earnings and Q1 guide. It wasn’t pretty, a confluence of many different things hitting at once. One thing that didn’t get talked about a lot was Intel’s long-term future. I get it- Wall Street lives off quarterly reports, and most press articles have an even shorter time horizon. And for many who don’t understand the semiconductor market, it’s hard to talk about a future with a challenging current narrative. Having been in and around semiconductors for over 30 years, I can definitively say that a real semiconductor turnaround takes 4-5 years, and I’ve never seen one happen quicker. While it’s easy to start with Intel’s quarter and guide and ignore the future, I’m going to start with a view of the future and then talk about the quarter and guide.
Let’s dive in.
Semiconductor turnarounds take time
I’ll point out first that, as humans, we’re not great at spotting turnarounds.
Remember when everyone thought Apple was headed out of business when it needed Microsoft cash to survive? Oh, and AMD.
So few people saw the potential of AMD in its numerous crises before its Ryzen announcement, and this could very well be the case for Intel. Having been a key part of one AMD turnaround, watching it almost run out of cash on my exit and covering the other AMD turnaround from the analyst sidelines, I think we can safely say turnarounds take between three to five years. The reason semiconductor turnarounds take so long is that silicon change takes time.
I like to simplify significant semiconductor company changes in a few buckets:
- Product and market: the products you develop and the markets you play
- Architecture: the fundamental method of designing IP blocks and packaging
- Manufacturing: how you build the chip
Intel is engaged in changes in all three at the same time.
All these changes take time.
Entering new markets and being successful takes time
On the product and market side, the biggest change is that Intel has expanded its TAM to go after consumer discrete GPUs, datacenter discrete GPUs, datacenter HPC SoCs, consumer AI chips and datacenter AI cards. Intel currently has incredibly low revenue share in these products and markets. NVIDIA now dominates in these areas; AMD is a player, and there’s a lot of upside for Intel if it can be successful in these markets. Roughly, you can think of the value of these new markets as the market cap of NVIDIA ($471B). Yes. That big. Intel is in its first “real” year of entering these new markets.
Entering new markets with entrenched and ferocious competitors isn’t easy. And takes time.
Completely changing the way products are architected and being competitive takes time
Intel is also in the middle of an architectural revolution, one with its IP building blocks (ie, compute, GPU, AI, I/O chiplets) and the way it packages those chiplets. I’ll focus on packaging.
Instead of creating monolithic designs where everything on the die is in one process and made at one fab, all of Intel’s future logic designs are distributed and assembled from 2D, 2.5D and 3D packages connecting chiplets. Those chiplets all don’t have to be fabbed by Intel either, as we saw with TSMC fabbing Intel GPU chiplets. While I wish we could have seen some TSMC-produced CPU chiplets, we didn’t, and it will take a few years to see what Intel could do if it didn’t have a chiplet size disadvantage.
I believe Intel has the industry lead with 3D packaging capabilities, which is the future.
Sapphire Rapids was one of Intel’s first high-volume, entirely distributed architecture products. Its modular design was one reason it was late. Intel’s first completely modular PC product is called Meteor Lake, and the company confirmed that manufacturing would ramp in 2H 2023. OEMs I talk with are excited about this product and see it solving many competitive issues with Apple. Interesting, huh?
Getting good at distributed designs isn’t easy as it requires a complete change in design, test, and packaging methodologies and technologies to ensure IP can perform competitively, be shared across other designs, and be cost-effective. For 30 years, Intel designed chips with a monolithic architecture, and all future designs will be modular. This takes time.
It’ll be interesting to see how Intel does when all its designs are distributed and its chiplets fabbed on a competitive process.
Becoming a competitive, full-service foundry with leading-edge nodes takes time
I believe the biggest strategic change at Intel since Pat Gelsinger’s arrival is the decision to build a foundry business, IFS (Intel Foundry Services), which will compete with TSMC, Samsung and GlobalFoundries. This obviously takes time and requires industry-leading nodes for the leading edge. It also requires the analog and specialty capabilities of Tower Semiconductor, which should be closing soon.
First, a foundry is a great business if you can get it right. Look at those obscene TSMC net profit margins of 47.3% of the recent quarter. Secondly, if Intel can execute IFS, it provides the scale for its own designs that could give it with PPA (performance, power, area) competitive leadership on leading-edge designs. Right now, this seems like a fantasy, but you have to dissect what would make that a reality. It starts with having competitive nodes for the leading edge. CEO Pat Gelsinger announced that it is on track or “ahead” of its commitment to deliver five nodes in 4 years.
The company disclosed that:
- Intel 4 is "manufacturing ready," and Intel 3 is "on track." Both of these nodes use EUV, which should decrease costs in a big way by removing double-pattering.
- 18A/20A: "silicon is running in fab" "of a major potential foundry customer." These two nodes utilize some pretty big breakthroughs like RibbonFET and PowerVia. The full PDK should be shared with key customers within the week.
This is incredibly positive news. I can’t tell you what good news this is. As long as it keeps being good news.
From a customer standpoint, Intel seems to be doing well, too:
- MediaTek will buy $4B of production using Intel 3 over the node’s lifetime
- It added a “leading cloud, edge and datacenter solutions provider” as a customer for Intel.
- It has 43 test chips in flight.
- Engagements in 7/10 of largest foundry customers
I have a lot more to say about what it would take Intel to have a competitive foundry, but neither you nor I have 10,000 words. Net-net, getting into the fab business and being good at the leading edge, lagging edge, and specialty takes time.
Now onto the short-term pain.
What was announced for Q4 2022
For Q4, overall revenue was down 32% year over year, gross margins were off 14.5%, and non-GAAP EPS was 92%, substantially missing investor expectations. The revenue and profit declines were shared between its PC and datacenter division, the revenue and profit dollar drivers of the company, but the datacenter group did beat investor expectations.
The PC group explanation, revenue down 36% and profit down 82% is straight forward, and I wasn’t surprised.
My channel checks indicated months ago that demand was soft in both consumer and commercial markets. At this time, both OEMs and ODMs limit the amount of WIP inventory they hold and therefore don’t order.
I also heard that Intel was doing many “meet comp” deals versus AMD, which drove down ASPs. With 39.6% overall margins and 3-4 points lower without an accounting change, that seems to make sense. Until we get to ASPs, where Intel claimed it had “record… up 11% YoY,” which I can’t foot to the margins. Intel doesn’t reveal ASP by form factor, so it’s hard to put together all the pieces. I believe that Intel took unit share from AMD, but AMD meet or won revenue share. There’s little incentive for AMD to accept any losses here as long as it stays relevant in markets to ODMs, OEMs and the channel. When you own your own fabs and financially it makes sense where any profit above cost works, it makes sense as long as it doesn’t erode brand. Intel 7 is more expensive than 10nm until it gets to great yield, so COGS were impacted.
The datacenter and AI group had a similar financial story with revenue declining 33% and profit down 84%. While the group beat investor expectations, I was surprised the revenue was only $4.3B. I expected a lot more as Sapphire Rapids had been shipping in such high volume and there’s always pent-up demand for Intel’s new products.
In my call with Pat Gelsinger before the call and in the earnings deck, Intel alluded to “TAM contraction.” In a call Monday with DCAI leadership, China kept coming up as “soft.” This “TAM contraction” was new for me as Dell Tech, HPE, and Lenovo continued to rack up respectable server growth numbers, and the hyperscalers didn’t appear to be shrinking back. Intel also cited “competitive pressure”, likely AMD and maybe Ampere Computing, which wasn’t a surprise. AMD and Ampere are doing quite well with the hyperscalers. I am sure we will learn more as AMD announces today.
The Network Edge Group, Mobileye, graphics, and IFS all had record revenues but we’re enough to offset the PC and DCAI group.
Q1 2023 guidance
The Q1 guide didn’t surprise me. Intel provided as equally as dismal Q1 forecast as Q4 actuals and based on uncertainty, did not provide for full year guidance. I don’t blame the company, particularly with all the market uncertainty. The company guided Q1 -40% year over year, gross margins at 39%, down 14% year over year and an EPS loss of $.15, down 117%.
The Q1 guide comes from a confluence of challenges, starting with continued PC demand declines and PC inventory burn off, even larger than Q4. The company stated that server market size will decline, particularly in enterprise and China. I think AMD will continue to be a challenge for Intel in 2023 in the server space.
If nothing else, I hope I’ve made the case that semiconductor turnarounds take time, there’s gigantic upside to Intel succeeding at this one, and that it’s really-really good news that Intel is on track to deliver 5 nodes in 4 years. While there is a lot of execution to come and there is still a lot of risk, the “5 in 4” mantra is the most important thing investors can be focusing on. It’s not that design doesn’t matter- it does, but it starts with “5 in 4” and gives Intel a chance to turn this boat around.
Out of what may seem like me speaking out of both sides of my mouth, I need to acknowledge the train wreck Q4 and Q1 guide. It’s ugly; I’m not going to lie or try to sugarcoat it. What’s important to recognize is that Intel is in this challenging situation by a confluence of issues, including macro-economic issues and investments it chooses to pursue and it isn’t as simple to blame it on a short-term, self-inflicted wound. Don’t get me wrong, there were many of self-inflicted, but they happened before CEO Pat Gelsinger even came on the scene. Before Gelsinger arrived, Intel had a two-year process lead, collected 90% of the PC and server profit dollars and then blew it.
Now it’s up to Gelsinger to complete the turnaround, and by my 5-year clock, he has until January 31, 2026. There is a scenario on that day that says Intel has a technology lead (or parity at a minimum), which it opens up to IFS customers, Arizona and maybe Ohio capacity is online, and the semi market swings back, driven better economic conditions. I can’t wait to see what happens!