oof. Headline says it all. (The rest is paywalled but when the Street says “I don’t think I’ve ever seen anything quite like this before” after your earnings, you don’t need to hear much else.)
1 Like
I saw a headline that they had missed forecasts, but what was do bad to elicit this kind of reaction? Also, they’re only down 8% this morning so it can’t have been that bad…
I wonder how much of this is the price cutting they did and how much is due to losing business to AMD NVDA…doc
1 Like
It’s not just about the data center… given the good quarter in PCs an inline quarter was not what was expected…
Gross margins have collapsed
notebook ASPs have collapsed
Server ASPs have collapsed
“Beginning of the bear thesis writ large”
A beat was expected and not delivered
Intel and AMD are a duopoly, Intel doesn’t exist in a vacuum, AMD is calling for growth into coming quarters, Intel calling for a decline…
Process technology uncertainty likely influencing customer behavior, can’t be a good thing
“I don’t think they have a plan, and the current trajectory of the business isn’t going in the right direction”
Wish I had a transcript to share…
1 Like
Investors are reeling from a shocking set of results late Thursday from Intel, whose shares were set to tumble 10% at the start of trading on Friday
“We have written the phrase ‘Worst earnings report in our history of covering this company’ on more than one occasion over the last couple of years. But this time we REALLY mean it…”
That was the comment from a team of analysts at Bernstein, led by Stacy Rasgon, and a sampling of shock waves rippling across the investment landscape after the chip maker delivered its worst results in 20 years and a grim forecast.
Intel (INTC) shares were poised to open nearly 10% lower at $27.20. A year ago, they traded around $50.
Rasgon, who cut his price target to $20 per share from $23 and held to an underperform rating, zeroed in on the company’s first-quarter outlook, calling it “astonishingly bad even vs. low expectations, with revenues and gross margins collapsing.”
Intel forecast an adjusted loss of 15 cents a share for the current quarter versus expectations it would earn 25 cents, and said revenue would sink further to $10.5 billion, versus an expected $13.93 billion, as the company warned of a contracting data-center market and a glut of inventory.
…
Warning investors to steer clear of the stock, Stephanie Link, chief investment strategist and portfolio manager at Hightower Advisors, said on Twitter that Intel was “one of ‘the’ value traps out there.” She highlighted the 1,200 basis point decline in annual margins and another 480 basis point decline expected the next quarter. “This is more than the industry’s cyclical downturn,” said Harlan.
Some pointed to Intel’s dividend, saying it was time for the company to let it go. It paid a $0.36 cash dividend for each quarter of last year, and has faithfully paid a dividend for each quarter since 2013, and paid $1.5 billion in dividends during the fourth quarter.
…
Cowen analysts were almost left speechless, entitling their research: “Couldn’t really think of a title to describe that, but here’s the note anyway.”
“1H23 should be the bottom, but how quickly and to what extent does the business and P&L recover given competitive pressures and necessary investments for the turnaround? A long road ahead,” said a team led by Matthew Ramsay, who cut their price target to $26 per share from a prior $31, keeping a market perform rating.
The bottom?
Bernstein’s Rasgon, for one, wasn’t sure: “We keep asking ourselves when things will be as bad as they can get for Intel. And we keep getting surprised.”
3 Likes
I don’t think the statements from Stacy Rasgon are accurate based on the Intel cc. Intel may be mistaken, but they believe they both gained market share in clients while also seeing record high selling prices. We will need to see AMD numbers before we can really confirm this. Intel has stated that for all of 2022, they shipped about 10% fewer CPU’s than purchased by consumers due to a clearing of inventory. I think AMD also has pretty significant inventory overhang from the heyday of the pandemic. They expect inventory reduction to continue into Q1 and gave very poor guidance for Q1 expecting everything down quite a lot. As a significant departure from the norm they chose not to give any full year 2023 guidance yet.
Intel Data center revenue was up very slightly in Q4 from Q3, but of course down significantly year over year. They also believe the overall market is down significantly, but again we need to see AMD numbers to confirm this.
Intel was down much less than I would expect from such a bad report (revenue and profits are plummeting) but the market appears to already have this priced in, or perhaps the Intel stock price is more based on the future potential than any current results. We will know more when AMD reports next week.
1 Like
Talking heads on CNBC this am said they are still having trouble making the newest narrow line chips. This problem reported last year has still not been resolved.
Trouble probably means they are getting low yield of functioning chips. And that implies they don’t have enough top margin chips available.
They promised to address this problem last year but still no success. There were reports they decided a few years ago not to invest in the latest equipment but presumably to develop their own. Apparently without success.
I would need to see the actual report to make sense of this. Intel 7 is the latest manufacturing process in volume production and there is plenty of high end high speed Raptor lake CPU’s built on this process.
The next generation Intel 4 process is supposedly qualified for production but not shipping anything until midyear. There may, or may not be yield issues but Intel has not indicated any problems. Meteor Lake is the only high volume production product on this process, and we are told will ship in H2 2023. The die size for this product is very small, so it will yield well even if the process still has a high defect density. OTOH, the follow on server products scheduled for Intel 3 in 2024 are back to larger die products and require a healthy mature process to manufacture.
The Issue I see with the Intel_4 process is it currently only runs in the small R&D factory in Oregon, and competes for wafers with all the other R&D process start-ups. The Ireland production factory for this process is still checking out equipment, and I doubt they have production output until much later this year. Even if the process is healthy and ready to run, they have no production facility to run it in.
2 Likes
Building fabs is a multiyear process. In the past, in really bad macroeconomic times, Intel reduced it to only 2-3 years by shutting down existing fabs and upgrading them. Otherwise Intel would keep on making older chips as long as the market would buy them.
Today is different. AMD is so far ahead in both product and process that Intel’s market is shifting downwards faster, and AMD is steadily gaining share in the highest margin sectors. Hiring TSMC to make top of the line CPUs has added expenses over making them yourself. As Alan pointed out in a different thread, Intel has at least three new fabs in construction. All these things hurt Intel’s bottom line. The poor yields at TSMC on Intel’s best chips is another problem Intel used to self insulate. They’d upgrade one fab, ship very low volumes of their best chips (and charge a fortune for them) while improving yields, and when yields were better, then they’d copy that production environment in many other fabs.
Right now, Intel’s stock price is sustained by their dividends, and I believe releasing more stock is a part of their economic strategy as opposed to taking out loans. I also believe that cutting the dividend would really hurt Intel’s stock price; see what happened to AT&T. I think Intel chose to cut salaries, and to cut them more of managers, is a vote of confidence in where Intel thinks they are going. They think they will have two more years of issues and then turn things around. When (if?) profits improve they can reward the employees that stayed, avoid the issue of layoffs, and not have a ton of extra debt to manage like AMD did after Opteron.
Is Intel right in their assessment? To some extent I think yes. TSMC definitely wants to improve its SOTA yields; that will help Intel’s high end availability. New fabs will help improve their product mix. But Intel is shooting at a moving product. To me the wild card for Intel remains AMD. I think the Xilinx acquisition will open up new process enhancements for AMD. AMD has easily been beating Intel in process for like five years, and I made $150,000 investing in AMD in those years. If I hadn’t sold prematurely to satisfy my wife, it would have been over twice that.
We MUST remember this. AMD has known product improvements in hand by buying better wafers. They’d cost more, but to me they keep AMD a moving target for better products. I don’t know how long GF will refuse to upgrade, but as they become less relevant, AMD could move their controller CPUs to Samsung or TSMC and make faster CPUs and GPUs almost immediately. I suspect Samsung is working on those older process parts of the chiplet CPU. At many price points, it will make sense for AMD to maintain the product lead by paying more for smaller excimer wafers, and I think that math is leaning ever more in AMD’s favor. The simple benefit of tiny chiplets improves yields and makes this transition more likely.
I have INTC and AMD in my portfolio right now. I expect to divest INTC as soon as Monday but no later than Tuesday, as it is a pure dividend play for me. My problem is AMD’s recent gains make it fairly expensive, so I need to move into an as yet unknown stock. I’ve been investing in AMD for almost 25 years, and those investments are a huge part of why our retirement is so comfortable. But if I’d been better at knowing when to sell AMD, I could easily have a couple of million dollars more, and we could afford my wife’s dream house in Pacific Palisades.
Fool on,
Roleplayer
5 Likes
I think Intel’s problem is not missing its process goals. It is reading the wants of their customers wrong. AMD, on the other hand, seems to listen to its customers and produce what the customers say they want. I’ve been following TLBs for decades. TLBs, translation lookaside buffers, become important when you have lots of cache misses. It is hard to design tests that stress cache misses and TLB hits. But the impact on latency is huge. If you have a cache miss and a TLB hit, the CPU immediately starts reading from main memory. If you have a TLB miss as well, the CPU has to walk down the tree of main memory to find where the needed page is located.*
Right now, Intel is missing it, and AMD actually started to follow Intel. There are very few customers out there who need more desktop performance. But corporations don’t want space heaters on everyone’s desk. The “new” AMD 7000 non-X CPUs are the exact same hardware as the existing chips, but they are set up to use less power. A 10% decrease in performance for a 50% reduction in power? Lots of people want that.
Also, Intel is adding lots of specialized engines to their mainstream server chips. Are these features that some customers want? Sure. Intel is adding these features with the idea that they need to be there for everyone so that software will be written to use them. But AVX 512, in particular, the MAC (multiply and accumulate) instructions, are something customers want and need compiler support for. Look at where Intel is now. The compilers, especially Intel’s are there, and now they are taking the feature out, while AMD is now providing them. Technically AMDs latest chips provide the instructions but use two floating-point pipes. If you have extremely compact code such that most clock cycles have a 512-bit MAC, and run the code on both threads in each CPU core? The AMD CPU won’t run the code as fast as you would hope. If in the densest code MAC are every other cycle, or even one cycle out of three? Performance is fine. My experience is that I can write extremely dense linear algebra code, but compilers can’t.
On the other hand, the two most important low-level floating point codes are FFT and linear programming. (Code that optimizes a system of equations with lots of constraints.) Both of these have areas where you are dealing with a single (double-precision) vector–in the FFT it will normally be double precision complex. That’s 128 bits. You can copy some values so you have two copies in one register. It works, but so far, I’ve found I want more named registers to be available. Why named registers? Both AMD and Intel floating-point processors now have lots more renaming registers. But there are only 16 names–eight names if you are in compatibility mode.
- I have some software that stresses memory in this way. It uses a large tree in main memory. The optimization I have done is to put a node and potentially its (3) children, (9) grandchildren, and (27) great-grandchildren in one page. (It is a 2-3 tree if you know what that is.) This may “waste” memory, but it results in much lower run times. I don’t get close to 16 Gig of memory, and my new system has 32.
I tried running it on an Intel CPU once, and gave up after two hours. This was back when AMD had a two-level TLB with 512 entries on the second layer, and Intel had a single-level TLB with, I think, 32 entries. Intel has gotten better, but they are still behind AMD in memory management. (If Microsoft would make it easier to use large/huge pages, I’d have to rethink that code.
1 Like