4Q and FY21 Financial Results and Analyst Call

January 26, 2022

Financial Statement

Income Statement

Intel’s income statement is the opposite of the WFE companies I follow. In the WFE companies things keep getting better as you move down the statement. For Intel, things get worse the further down you go. It brings to mind the image of a ball thrown in the air. Intel right now looks like the ball shortly reaching its apex. It is gathering speed downward and appears to be grasping to hold altitude.
There is significant seasonality in Intel’s business, so the comparison to the same quarter a year ago is better than sequential performance, though I still keep an eye on both. Revenue in the quarter was $20.5B compared to $20.0B a year ago. Sequentially, revenue rose $1.3B. Intel spent disproportionately more to support sales this quarter so gross margin compressed 330 bps, to 53.6%. Overhead also rose faster than revenue, leading to further compression in operating margin. There were gains on equity investments in the quarter of $359M, a small amount relative to Intel’s overall business. Intel is a business in decent, but they still make a lot of money. Net income for the period was $4.62B, down from $5.86B in the same quarter a year ago. These lower results are from across the company; worse pricing because of increased competition, less efficient operations, and more investment in R&D to start up their foundry business and catch up on technology. Share count declined slightly (0.6%) from a year ago. Fully diluted EPS was $1.13, a decline of 20% from the fourth quarter last year.
Intel’s descent started during 2021, so their full year results are flatter at the top of the income statement than the quarterly numbers. Revenue was $79.0B compared to $77.9B a year ago. Gross margin was 55.4% vs. 56.0% last year. Between higher R&D spending and other restructuring costs of the turnaround, operating expenses swelled by 22%, to $24.4B. Because of this, income before taxes declined $3.3B from 2020, to $21.7B. The company provisioned half (8% vs. 16%) as much for taxes this year as last, so the reduction in net income wasn’t nearly as large as the rest of the declines. Net income for the full year was $19.9B compared to $20.9B in 2020. EPS on 4.09B diluted shares was $4.86. Per share earnings were $4.94 in 2020. Share count declined 3.3% year-over-year. Even though Intel is in decline and navigating a major turnaround and transformation, they still make a massive amount of income. $20B per year in earnings, almost $5 per share, is formidable.

Statement of Cash Flows

Intel generated almost $30B from operations in 2021. $10B of this was depreciation and amortization, which will go back out, and then some, when we get to investing activities. Share-based compensation was $2B for the year, more than 10% of the value of net income to this not-an-expense expense. Working capital consumed $2.76B worth of cash during the year, in a mix of higher AR, higher inventories, and more prepaid supply agreements. The last item is largely advanced payments made to TSMC to secure foundry wafer supply. There is an irony that Intel has to pay this much in cash advances to the very company they are trying to compete with in their new foundry business. PP&E investment in the year was $18.7B, almost double the D&A for the year. This is investment in the company turnaround, the cost of starting a foundry business and trying to catch up on process technology. There was a net flow of $5.26B of cash into the combination of trading assets and available-for-sale securities. The company must feel good about their cash position overall to do this. Total cash used for investing activities was $25.1B. The company issued a net $2.5B worth of debt during the year. They bought back $2.4B worth of stock (gross, not net of share-based compensation) and paid $5.6B in dividends. Adding it all together and Intel’s net cash position declined by $1B, to $4.8B. The company generates so much cash that this large percent change in their cash balance doesn’t really matter. For the full year, my owner’s cash flow for Intel (which includes taking out the effects of share-based compensation) is $10.24B. Divided over their market cap of $198.8B (as of 2/1/22) is a cash flow yield of 5.2%. This isn’t as high as last quarter but still a good yield compared to other semiconductor companies. Intel is discounted because of their current situation (behind in technology and investing massively in a new business). Of the $10.24B in owner’s cash flow, management returned 80% ($8.06B) to shareholders, two-thirds in dividends and the remaining third in share repurchases.

Balance Sheet

Intel’s cash balance declined from a year ago (to $4.8B from $5.8B) but most of the rest of their assets grew. Total assets rose $15B, to $168B. Trading assets were almost $6B of this increase and equity investments rose $2B. Intel’s investments in turning the company around, in more advanced process technology and in their nascent foundry business, increased PP&E by nearly $7B. Working capital contributed another $2B. On the other side of the ledger, current liabilities increased $3B. Long-term debt was flat. All told, the company’s liabilities increased by a billion dollars, so book value increased $14B in the last year, to $95B. This is $23.20 per share, nearly half of where the shares are currently trading.

Divisions

Since this is the report for the full fiscal year 2021, I will comment on the twelve-month results rather than the last three months. One comment I will make on the most recent quarter is revenue in the Client Computing Group (CCG) declined by 8% from the previous quarter and operating income declined 20%. This is at least partially from seasonality but may include some share loss to AMD and others. Year over year Intel’s desktop computing group (CCG) saw revenue rise slightly but operating income declined about 3%. Their Data Center Group (DCG) had a decrease in revenue year over year, from $26.1B to $25.8B. While this is a small decline in absolute dollars, it is much more significant considering how fast the overall data center space is growing. Intel is losing share and reducing their pricing to try and compete. This is supported by the much larger drop in DCG operating income, which was 40.5% in 2020 and 27.1% in 2021. For comparison, CCG operating margin was 37.8% in 2020 and 36.2% in 2021. Intel’s competitive position in desktop PCs is much stronger than in Data Centers. Data Center customers are much savvier than consumers and corporate IT groups, so they realize the price premium Intel has historically gotten over AMD and others isn’t paid for with better performance, so they are switching away. Intel has to respond to this by cutting their prices. Intel is losing in data centers, which is where the most growth in processer demand is happening today and in the coming years. With the exception of “All others”, Intel’s other divisions produced year-over-year growth in operating income. Revenue and operating income were up in both the IoT Group and Mobileye, with the former tripling and the latter almost doubling their operating incomes. Revenue declined year-over-year for Intel’s NAND memory group. This is not surprising as it is being sold to Intel. The operating income in this group rose by 4x, but not because it is being run any better. This large increase in income is only because depreciation, the largest expense in COGS, was suspended once the sale was announced. The Programmable Solutions Group (PSG) continues to muddle along, with both revenue and operating income up somewhat. All the operating income made by these four divisions (IoTG, Mobileye, NAND, and PSG), and then some, as lost by Intel’s other ventures, those in the “all other” category. The four divisions made a combined $3.17B in operating income in 2021, and “all other” lost $5.34B during the year. If Intel did nothing but their two core computing businesses they would have been $2B better off in 2021.

Earnings Call Notes

Comments are all in reference to the non-GAAP financial figures

Pat Gelsinger (CEO) prepared remarks

• 2021 was Intel’s best revenue year ever. Pat emphasized how both their PC and data center revenues were all-time records.
• Mobileye has had 14 consecutive years of revenue growth. The company will take Mobileye public in 2022.
• Intel broke ground on their two new Arizona fabs three months ahead of schedule
• He called the 50th anniversary of the Intel 4004 processor a “sacred moment”. Quite a choice of words even for a company as arrogant as Intel.
• Shortages in substrates, components, and foundry wafers are keeping customer demand from being fully served. They expect these constraints to continue through 2022 and into 2023, though they will decline in severity.
• He said again, as he did in last quarter’s call, that Intel plans to progress five process nodes in the next four years, matching their competition at the leading edge in 2024 and regaining process leadership in 2025. They are ahead of the schedule they laid out in July 2021.
• Gelsinger spent several minutes talking about how great Intel’s product portfolio is, how they are winning over competitors, etc. I mostly ignore this commentary for two reasons. One, fast innovation is a baseline requirement in semiconductors. All companies need to do it well every quarter just to keep up. The second reason I ignore it is because Gelsinger sounds like a salesman, never saying anything to indicate things aren’t going perfectly at Intel, ahead of schedule, highest ever, etc.

David Zinsner (CFO) prepared remarks

• Commercial demand drove their Client Computing Group where customers prioritized higher-margin systems to best use their limited components caused by supply chain shortages
• Strong enterprise and government demand drove their Data Center Group in the fourth quarter. Operating profit was down because of a federal-related one-time charge that was previously disclosed and the cost of ramping 10nm. Full year DCG profitability was down because of start-up costs of new technology and investment in future products. They don’t want to say increasing competition is eroding their margins here, but I suspect it is.
• PSG revenue growth was limited by supply chain shortages. If not for these shortages, the PSG business would have seen $500M more in revenue during the quarter. This is 25% of total revenue in the year.
• In fiscal 2022, stock-based compensation and gains and losses from their ICAP portfolio will be removed from non-GAAP results.
• Also beginning in fiscal 2022, Intel is restructuring their business units:
o Client Computing, mostly the same as the current Client Computing Group, plus workstation revenue (previously in DCG?)
o Data Center and Artificial Intelligence, the Data Center Group and Programmable Solutions Group (former Altera).
o Networking and Edge, includes their pervious IoTG and their networking products previously in DCG
o Accelerated Computing and Graphics, which is all discrete graphics products
o Mobileye is unchanged
o Intel Foundry Services includes wafer and packaging
• Q1-22 guidance: revenue of $18.3B, GM of 52.0% as 10nm ramps, EPS of $0.80, and a tax rate of 15%
• The company deferred full year guidance to their upcoming investor day in February

Question and Answer

• One analyst pointed out this is the fifth quarter in a row of cloud revenue declines in DCG. Their DCG revenue growth is carried by government and enterprise. Intel is stronger in the legacy segments of the data center markets. Gelsinger didn’t answer the question, instead saying they forecast high growth in edge computing.
• Following a strong PC growth market in 2021 (15% up), 2022 will be modest, up maybe 2%
• 2021 capex was 60% equipment and 40% facilities, to give an idea of what might be coming in 2022 and beyond
• In response to a question on inventory and PC growth, they said that correcting for discontinued businesses (modem, etc.) that they matched the overall PC market for growth. The answer wasn’t convincing, likely because they are losing share in the PC market but don’t want to acknowledge that.
• Intel is making foundry revenue from their advanced packaging services today. They expect to have wafer foundry revenue from Intel 16 in 2023 with advanced nodes coming later. Though I don’t see it on their web site, I think Intel 16 is their 16 nm node, an older process technology, which is not surprising as it is easier to bring in their first foundry customers in on established processes.
• One analyst asked a pointed question about gross margin. One customer asked why their major competitor (AMD, though they weren’t named) has the same gross margin as Intel today, without having the expense of owning fabs. When will Intel be able to increase their gross margin through pricing power? Long term, foundry margins will be on the lower end and their products will be on the higher end. Their answer mostly deferred to investor day, and also said they believe their structure (containing to be an IDM) will be superior to the rest of the industry.
• Gelsinger said that the spin-out of Mobileye is part of a model they like for value creating, leaving the door wide open for further spin-offs
• They haven’t ramped a new node in their data center business in five years
• The CHIPS act will be an “accelerant” to their investment in Ohio. The size of the bill will influence how fast that site grows.

Summary

My first thought from Intel’s results for their full fiscal year 2021 and associated call with analysts is Intel is a legacy business trading on their past accomplishments. They appear to be losing market share in both data center and desktop computing, though both of those businesses continue to generate tremendous amounts of cash. Intel’s business units outside of core processing continue to light money on fire. If it’s not processors, Intel is mediocre to poor at it. One analyst pointed out the brutal truth that AMD has similar gross margins to Intel’s without the trouble of developing new fab processes and investing in and running factories. This is because TSMC is so much more efficient than Intel. TSMC can make their 50% gross margin selling wafers to AMD, and AMD can take those wafers and make 50% gross margin selling products based on those wafers to their customers. Gelsinger’s arrogance and admiration for Intel’s past accomplishments continues to be his most prominent characteristic. As I’ve said in previous posts, I don’t believe Intel will be able to catch up on process technology. They are too big and too arrogant. Their culture needs to change too much to compete with companies like TSMC and Samsung. Intel is a has-been talking about what they used to be. But, they are also one of the few companies that have advanced semiconductor manufacturing capability and the scale to continue to compete. Few companies in the world are within two nodes of the leading edge. Those are TSMC, Samsung, and Intel in logic, and Samsung, Hynix, Toshiba (Kioxia JV), and Micron in memory. There is a massive moat around this, a moat that grows larger every generation. I have a 90% plus confidence level that Intel won’t execute on their plan do advance five process nodes in four years. But they don’t have to in order to stay within the top tier of semiconductor companies that have advanced processing. They will continue to have the weakest process technology and manufacturing capability among those top six, but they will still make a lot of money. As I said last quarter, at today’s share price, I think this is being underappreciated by the market. I am more likely than not to open a starting position in Intel in the next few months, as my cash situation allows.

-Smooth Hughes (no INTC position)

5 Likes

Intel won’t execute on their plan do advance five process nodes in four years.
This is a bit of a cheat on their part. Using the old naming convention, the five nodes are:
10nm++ (now Intel 7), 7nm (now intel 4), 7nm+ (now Intel 3), 5nm (now Intel 20A), and finally 5nm+ (now Intel 18A). It basically is 3 years from 10nm to 7nm, and then 2 years from 7nm to 5nm.

To be fair though, Intel 10++ (or Intel 7) is clearly equivalent if not better than TSMC N7. TSMC N3 continues to slip and no longer contains GAA (gate all around). Intel picked up some GAA technology from IBM after the global foundries crash, which should help them accelerate that transition. TSMC just started shipping N4, with N3 early next year. Intel should ship their N4 later this year, and N3 equivalent later next year (2023). I see the gap as less than a year, and with AMD products on 2 year old TSMC technology Intel should be very competitive with AMD.

Historically TSMC has started up new processes using small die size iPhone CPU’s, which has allowed them to make money while the process still has a high defect density. From what I can tell the Intel Meteor lake chiplet being used for the Intel 4 start-up is still a fairly large die. One of the first things Gelsinger did last year was to double the R&D wafer starts. Accelerating the process start-up schedules is going to be very expensive, and the main reason for the grim gross margin forecast.

I think the plan is solid, but it is going take time and the huge expense adds risk.
Alan

3 Likes

Thank you for the insight on Intel’s process roadmap Alan. I didn’t know the information you shared comparing the TSMC processes to those on the Intel roadmap. I am dubious on their goal to succeed in such massive undertaking, following more than a decade of losing their capability in manufacturing. I don’t think what is lost in that decade can be regained in a few years. But maybe I am underestimating Gelsinger’s ability to turn Intel’s culture and ability to execute around quickly. I have watched my own company gain process leadership, lose it, and then get it back again. It took us upwards of a decade to get our edge back in manufacturing after we lost it. But maybe we are slow learners compared to Intel. :slight_smile:

Smooth

It is easy to lose sight of the fact that Intel spends more on R&D than AMD+TSMC+NVDA because they don’t appear to have much to show for it.
The processes we attach numbers to (Intel 4, TSMC N7, etc…) are an integrated collection of more basic technology elements. Left to their own devices, engineers will tend to underestimate the risk of any given technology element, and as a result can crater a whole program with just one bad decision. The Intel 7 (10nm) process has the tightest M1 pitch in the world printed uses DUV steppers. It is also the only one in the world done with Cobalt instead of copper. Really cool technology; two year delay in the program. They could have made the thing 5% more expensive and saved countless headaches.

The original Intel 7nm (Intel 4) process had 5 EUV layers. EUV layers are expensive. TSMC uses more than 10 on their leading edge processes. When Intel got into trouble about a year and a half ago, Intel increased the EUV layer count to 10. This greatly reduces the risk in the program, but also increases the wafer cost. This is a hit to future gross margins, but improves the process schedules.

Intel has invested heavily in the tip of the spear for technologies many years out. This is why I am somewhat confident that given improved decision making and risk management they will be more successful in hitting schedules.

You are spot on about Intel just burning money with acquisitions. It appears this is something Swan was pretty good at fixing as it seems both Altera and Mobileye are being successful. Now if Gelsinger can fix the internal decision making and execution machine they might just have something.
Alan

4 Likes

I think the failure in terms of process node tech we hear about is partly one of marketing, hence the new names. The sophistication of process nodes has long been detached from ‘nanometers’ having any real meaning.

Intel’s 10nm (Intel 7) is about 90-100MT/mm (million transistors per square millimeter). Compare to Samsung’s 8-11nm nodes which are between 50 and 62MT/mm2. TSMC 10nm is about 52MT/mm. In other words, Intel’s 10nm is about double the density of competitors 10nm. It is equivalent to TSMC 7nm, which is around 96MT/mm2. TSMC ‘6nm’ is really not much different from ‘7nm’ in terms of density, so that is more of a marketing coup for TSMC. Same with their N4 node, which is really not much different from their 5nm.

In other words, the process node gap is nowhere near as big as most “analysts” make out.

What is more, Intel’s biggest x86 competitor AMD does not have access to TSMCs latest nodes for about a year after it goes into full production. This is because Apple monopolizes their first year of production.

Having said that, there is a gap. TSMC has its 5nm node which is around 170MT/mm2, 70% more dense than Intel 7 aka “10nm”.

However, up to this point no desktop/laptop/server chip from their biggest competitor AMD used 5nm.

In other words, at this moment, there is no node advantage for shipping AMD products and there won’t be until late 2022. The just released Milan-X is still on TSMC 7nm+.

Meanwhile Intel seems to be firing on all cylinders as far as architecture, at least in the client space. Tiger Lake, and now Alder Lake, are extremely competitive and by most measures superior to AMD products in the client space. The Alder Lake chips in particular have regained the top slot on desktop, and look poised to do the same on laptops for all of 2022. Desktop will also have the Raptor Lake update to Alder Lake in Q3/Q4. So overall, I expect Intel to regain market share throughout the year in the client compute space - desktop/laptop - and that will show in Q1/Q2/Q3 2022 market share reports is my bet.

Server is their weak spot, and won’t be addressed until 2H with ‘Sapphire Rapids’. So I expect server to continue to bleed until then. That is reportedly a very strong server contender, and will probably re-assert Intel’s lead there. The danger, is that it may only re-assert itself for a quarter, as AMDs new architecture chips on TSMC 5nm should be coming online in Q4 2022/Q1 2023.

Now here is the crux of the issue for Intel. Their “Intel 4” aka “7nm” is reportedly better than TSMC 5nm with ~15% greater density (specifically 200Mt/mm2 vs 170MT/mm2), and due out in late 2022 for server then early 2023 for client. TSMC 5nm is what will be used by AMD with next gen ‘Zen 4’ based desktop/laptop/server parts.

In summary I think Intel will dig into AMD in the client space, bleed some more in the server space, for Q1-Q3. They’re also likely to take a chunk out of both AMD and Nvidia market share in the GPU space, and I suspect it will be a significant chunk.

That leaves Q4 2022/Q1 2023. This is where I think the tables may turn, with Intel gaining in the server space and potentially losing again in the client space as AMDs Zen 4 / TSMC 5nm desktop parts arrive. Much of that depends on Intel getting its Intel 4 node online for server and then client systems in early 2023. Assuming they do, then they should regain their client space in mid 2023.

After that, AMD is not likely to have access to TSMCs 3nm node until 2024. In fact, Apple is likely to use 5nm this year for the next gen iPhone as TSMC seems to have issues with getting enough volume out of 3nm for 2022, meaning Apple would use it in 2024. This would effectively make it unavailable to AMD/Nvidia until the following year - 2024.

Assuming that happens, and assuming Intel can execute on Intel 4, they will have not only closed but slightly bettered the node available to AMD come 2023. So, it is certainly possible that Intel will have the server space throughout 2023 and have an answer to Zen 4’s 5nm node on clients by Q2 2023.

The race is pretty tight, but TSMCs flub with 3nm combined with AMD only having access to new nodes a year after TSMC goes into full production (due to Apple taking up all capacity) has opened a gap for Intel to step into and re-assert some dominance they lost a few years ago.

Of course, Intel has to stick to its schedule for this to happen. I personally have a lot more confidence that Gelsinger’s team will succeed in this area than Swan’s team.

3 Likes