JasonDoiy
Investment Thesis
The initial investment thesis was that Intel (NASDAQ:INTC) would turn around by improving its execution under new CEO Pat Gelsinger, who joined in February 2021. Intel would regain both process and product leadership by 2024/25.
Despite much recent ado among investors about the latest product delays, those products were conceived under prior management. As such, those delays do not impact the overall thesis, which remains valid. If successful, investing in Intel now could result in outsized returns as investors are prematurely giving up on the turnaround.
In other words, the share price is being priced on near-term concerns, giving opportunities for long-term investors.
Sapphire Rapids analysis
While it is sad to see Intel’s poor execution, most recently embodied in the failed Sapphire Rapids ramp, simply put investors should not blame Pat Gelsinger for this. As a matter of fact, the initial Sapphire Rapids tape-out already powered on in Q2 2020, about 9 months before Pat Gelsinger would officially become Intel CEO. Note that it already takes about a full quarter just to manufacture the CPU blueprints that were sent to the fab, which means Intel engineers sent the first iteration of Sapphire Rapids to the fab a full year before Pat Gelsinger became CEO.
Hence, it would be foolish to blame Pat Gelsinger for a CPU that already booted Windows and Linux several quarters before he became CEO.
Of course, though, as we now know due to leaks, Sapphire Rapids has since gone through 12 so-called steppings (iterations), reportedly fixing 500 bugs in the process.
So given the time length involved with developing a new CPU, the reality is that, as other contributors have pointed out, there might be some more pain in the near-term as Intel works its way through its roadmap.
In any case, Pat Gelsinger actually didn’t wait until the latest delays to take action, as he basically fired/replaced former data center executive Navin Shenoy in June 2021, which is the same month as when Intel announced the initial Sapphire Rapids slip to 2022. Hence, Pat Gelsinger swiftly held the person who was actually responsible for the Sapphire Rapids mess accountable.
For completeness, Bob Swan had announced in Q3 2020 that initial production shipments were planned for Q4 2021, which ultimately became Q1 2022: a one quarter delay. However, the catch is that this only pertained to one SKU, with the high volume SKUs apparently containing a security bug that needed to be fixed and caused those SKUs to be further delayed. As such, although there are rumors that the general launch has been delayed until Q1’23, Intel has said that the broader ramp will occur in the second half of 2022.
Ultimately, this means that Pat Gelsinger’s influence will only gradually become more visible as time goes on. For example, when Intel announced Granite Rapids at the February investor meeting, Pat Gelsinger noted how the CPU had been completely redefined, changing basically everything expect the name: it was part of a new platform, it was ported forward to Intel 3, and the CPU core itself even got a new architecture. As a compromise, the CPU did get pushed out further in time, launching in the first half of 2024 (if it launches on schedule, at least). Granite Rapids will power on in Q3.
Looking even further out, Diamond Rapids was originally scheduled for 2024, according to a roadmap that leaked in November 2020. It has not been officially announced by Intel yet, but given Intel’s target for an annual cadence it is likely targeted for H1 2025 on the 18A node. In that case, it would achieve a clear process advantage against AMD’s 3nm Zen 5. Nevertheless, this means even Diamond Rapids was already (in name) on the roadmap before Pat Gelsinger joined.
Meteor Lake
Another recent report concerned Meteor Lake. It stated that the N3 tile for Meteor Lake would not go into production until Q4 2023, which would put its launch firmly into 2024. However, this rumor was actually already debunked by Pat Gelsinger himself several days prior to its publication, as he had confirmed on the earnings call that Meteor Lake would launch in 2023.
Moreover, at the recent HotChips conference, Intel announced that the GPU is actually manufactured on N5, so Meteor Lake doesn’t even use any N3 at all.
Still, a few nuances may be noted. Although Pat Gelsinger said it would launch in 2023, he didn’t specify first or second half of 2023. As I already discussed a few months ago, Meteor Lake was first scheduled to launch in the first half of 2023 with a GPU with 192EUs (execution units). However, a recent leaked internal roadmap slide put Meteor Lake in the second half of 2023 with a GPU of 128EUs. Strictly speaking, a second half of 2023 launch would still signify a delay for this CPU compared to prior Intel statements for an early 2023 launch.
In any case, just as with Sapphire Rapids, Pat Gelsinger could not have influenced Meteor Lake’s development much if at all, as in May 2021 former PC executive Gregory Bryant announced that Meteor Lake had been taped in.
Lastly, it may perhaps be possible that the originator of the N3 rumor only knew about the manufacturing schedule, but didn’t know exactly what would be produced on this process node. Given the current information, it seems that Arrow Lake (instead of Meteor Lake) will launch with a TSMC-manufactured (TSM) 3nm GPU (and a 20A CPU). But since Arrow Lake is scheduled for the second half of 2024, as 20A will only start production in the first half of 2024, starting production of the GPU tile in Q4’23 (or even early 2024) seems very much in time for the Arrow Lake launch.
Moore’s Law economics
This sections serves as a brief reminder of the broader investment thesis.
First, as detailed nearly a year ago, in order to actually be a compelling investment/stock, Intel is entering new growth markets such as GPU, AI, robotaxi and foundry. This will make Intel much less reliant on the PC market over time.
Secondly, what makes Intel such a powerful and unique investment (potentially, at least) is that it is an IDM that develops its own chips and process nodes, and hence (1) it does not have to pay a foundry “tax”, as TSMC enjoys a healthy near-60% gross margin after its latest price hikes, and (2) Intel’s IDM advantage was previously compounded by the fact that it had process leadership.
Investors should know that Moore’s Law is ultimately an economic “law”: it is about the cost per transistor. Being ahead means a lower transistor cost, which provides a structural advantage over competitors, even besides power and performance.
Barring any delays, Intel will regain process leadership when 18A goes into production in Q4 2024. Note that 18A should have a leadership position even against TSMC’s N2, which is scheduled for Q4 2025, a full year later.
Since Intel has actually pulled in 18A from early 2025, it seems that 18A development is progressing very smoothly. This in turn implies that the turnaround thesis may already be derisked until 2028: TSMC’s N2 in all likelihood will not beat 18A, so the earliest timeframe for TSMC to challenge Intel again would be with its N1.4 node in 2028. But then again, as the tweet above indicates, Intel will not be standing still either.
In contrast, I have seen people estimate that Intel is currently anywhere from one to four years behind. This is, indeed, the turnaround that investors are banking on, not the near-term recession and PC collapse concerns.
Investor Takeaway
The definition of a turnaround is that the pain comes before the gain. This is indeed what analysts and investors are apparently finding out just now by observing execution issues such as the Sapphire Rapids delay or the Intel 4 delay from H1 2021 to H2 2022 production.
As noted, given the long timescales involved in process and CPU development, these delays have been obvious for two years now (although admittedly the Sapphire Rapids delay is worse than expected and Intel already saw a 20% stock decline when the 7nm delay was first announced in 2020). In any case, blaming Pat Gelsinger for these mishaps just goes too far as Sapphire Rapids was already sent to the fab and booting Windows/Linux a full year before Pat Gelsinger even became CEO. It seems Intel has been debugging the CPU for the last two years, and there’s nothing Pat Gelsinger could have changed about that. As he said, the CPU shouldn’t have had those bugs in the first place.
So, changing gears from backwards to forwards looking indications, in some of my prior coverage, especially Intel’s earnings reports, I have noted for the past six quarters how Intel’s R&D spending as well as its employee count has very noticeably increased. Intel had less than 111k employees when Pat Gelsinger joined, but this has ballooned to over 128k. Similarly, R&D spending has increased from a $13.5B to a nearly $18B run rate currently. Intel has disclosed that process development received a $1.5B increase, so that still leaves a $3B increase for other projects. For comparison, AMD’s (AMD) total R&D budget had “exploded” to just $2.8B in 2021 (although combined with the Xilinx acquisition this is now on an over $5B run rate).
So as Intel announced last October and further detailed in February, free cash flow and earnings are down for the next few years as Intel is ramping up new fabs and new process nodes as well as R&D spending. This means that if Intel remains valued based on P/E, and the multiple doesn’t change, then the stock price will go down. It will only be in the years after that, after Intel has regained process leadership in 2025, that Intel will start reaping the rewards of current investments (the financial benefits of process leadership as well as the growth from its new businesses such as GPU, robotaxi and foundry).
In summary, my thesis has been that Intel is a stock for the patient, and nothing has changed this. In fact, the most telling blow to the data center roadmap already occurred in February when Intel announced that Granite Rapids was delayed by a full year to 2024. Nevertheless, the Sapphire Rapids CPU delay that caused the current mess uses only the first of the “five nodes in four years”.