Three notes about today’s update:
- First, the Intel keynote I’m going to write about, along with the analyst Q&A session that followed, has the potential to go down as one of the single best keynotes I’ve covered in my career. If you want to abandon this newsletter and watch it now I won’t blame you.
- Second, if you do decide to soldier on, I’m going to include a lot of clips; it may be easier to consume all of them on the Stratechery website instead of email, so that the videos play in-line.
- Third, I will send out a second email in a few hours with a transcription of the clips. I do recommend you watch them, but I recognize that is not always possible.
Finally, this Daily Update is free to share.
On to the update:
From the Wall Street Journal:
Intel Corp.’s chief executive is fast-tracking efforts to revive the semiconductor giant with a broad plan that mixes increased outsourcing with a commitment to spend $20 billion on new factories that could help address a chip shortage.
New CEO Pat Gelsinger said Tuesday that Intel would rely more heavily on third-party chip-making partners, including for some of its most cutting-edge processors, starting in 2023. But Mr. Gelsinger, on the job a little more than a month, said Intel wasn’t abandoning its historic roots of being both a designer and manufacturer of chips and would retain most production in-house. The company, he said, also is renewing efforts to make chips for others and targeting customers such as Apple Inc. and chip rival Qualcomm Inc.
Yesterday’s keynote delivered on almost every meaningful strategy change I have asked from Intel, from building out a foundry business to re-organizing the company (it’s not a split, but close) to explicitly addressing geopolitics to embracing modularity to channeling Andy Grove.
Just as important, though, was the way in which Gelsinger delivered this news: he was transparent about how Intel had screwed up, demonstrated tremendous clarity of thought about Intel’s strategy, particularly in the Q&A, and was captivating and inspiring about why Intel’s best days were ahead of it. Of course a keynote is just a keynote — Intel has real work to do — but Gelsinger absolutely left the impression that if there is any chance of Intel delivering on his promises he will realize it.
There is also the timing: Gelsinger has only been in the job for 37 days (38 today), yet is already delivering the most sweeping overhaul of Intel’s business since Andy Grove and Gordon Moore got the company out of memory thirty-five years ago. That is, frankly, the best execution that Intel has demonstrated in years.
Gelsinger on Intel
I think the best way to approach this is to start at the top; I’ll provide clips of important points in the keynote, with a bit of commentary.
I am so excited to give you my perspective on the tremendous opportunity ahead of Intel and the plans we are putting in motion. As many of you know, I’ve spent four decades in this industry, three of those at this great company. And now, I’ve returned to my dream job as Intel CEO. I often say I grew up in the hallways of Intel, trained at the feet of its founders, and am driven and inspired by how technology has the power to improve the life of every human on the planet.
This is pitch perfect. Gelsinger very explicitly casts himself as following in the footsteps of Robert Noyce, Gordon Moore, and Andy Grove, even as he makes clear that he hasn’t been around for the last ten years. It’s an example of the exact sort of posture Intel needs to adopt: confidence rooted in the company’s industry-leading past, tempered by humility about the degree to which it has lost its way.
What Went Wrong
Let’s start with a progress report on 7 nanometers, a pivotal aspect of how we’re engineering the future, and from there I’ll discuss how we are putting our foot on the gas with our IDM model and our product roadmaps. We previously discussed how we identified and resolved the issue on 7 nanometers, which is our next generation process technology, intercepting our 2023 product roadmap. Let me go a click deeper on that today.
When Intel initially designed 7 nanometers, EUV was still a nascent technology so we developed our process to limit the use of EUV. But this also increased the process complexity. As EUV then matured and became more reliable, we experienced the domino effects of our 10 nanometer delay which pushed out 7 nanometers and ultimately put us on the wrong side of the EUV maturity curve. Today I’m pleased to share that we have fully embraced EUV, we have re-architected and simplified the 7 nanometer process flow, increasing our usage of EUV by more than 100%. We have a very strong partnership with ASML and plan to now stay on the leading edge of EUV usage are well underway. Our technology teams are moving rapidly through process maturity and as they do our confidence in 7 nanometer health and competitiveness is accelerating.
Leveraging our 7 nanometer process, we’re advancing the development of lead data center and client CPUs, starting with our Meteor Lake, our high volume 2023 client product. In fact, we expect to tape in our 7 nanometer computer tile for Meteor Lake in the second quarter of this year.
The actual news here is very bad! Gelsinger effectively announced that the 7 nanometer process is even more delayed than former CEO Bob Swan announced last summer, but actually, that’s less of a problem than it seems: it was very clear at the time that Swan was almost certainly not being entirely honest about just how behind Intel was, in part because it didn’t seem like he really understood the problem.
Here was the problem: over the last decade ASML — and ASML alone — has perfected extreme ultraviolet lithography (EUV), which is essential for making the most advanced chips, but Intel, instead of committing to ASML’s technology, tried to go its own way with its own technology and preferred supplier (Nikon) using immersion lithography. After a few years of delays the company finally got 10 nanometer working (barely), but instead of realizing it had hit the wall with immersion lithography, Intel believed it could apply its 10 nanometer learnings to 7 nanometer, which had been developed in parallel. It couldn’t.
Instead, Gelsinger made clear that Intel had to go back to the drawing board with 7 nanometer, fully committing to ASML and EUV. That is why 7 nanometer won’t appear until 2023, but at the same time, the transparency of Gelsinger’s explanation — which makes perfect sense! — is also reason to believe the new timeline.
This is also a reason to believe that Intel can re-accelerate future process shrinks; the most advanced lithography machines are a necessary component for making cutting-edge chips, but far from the the only factor. Intel retains all of its other strengths and know-how — which remember, kept the company ahead of the industry for decades — which it can bring to bear once its lithography problems are solved.
Meteor Lake features a breakthrough new x86 architecture and modular design, utilizing multiple manufacturing processes across our XPU IPs as well as Intel’s advances Foveros packaging technology. This speaks to a larger industry transition as customers needs and workloads diversify and grow in complexity. The world will move from system on a chip to system-on-package and Intel’s unquestions leadership in packaging technologies becomes even more valuable. Being able to expertly construct the best products using the best technologies is a critical differentiator for us and one that delivers enormous value for our customers.
In a system on a chip (SoC) much of the computer’s functionality is a part of a single silicon die; Apple’s current top-of-the-line A14 chip, for example, has two high-performance cores, four low-performance cores, four GPU cores, 16 neural engine cores, an image processing core, machine learning accelerators, and the secure enclave all on a single die that is manufactured using TSMC’s 5 nanometer process. Traditional computers, to go to the other extreme, have all of those different components in modular pieces, connected via a motherboard.
A System-in-Package is in the middle: not everything is on the same die, but it is all in the same package. That means that some cores could be on one process, other cores on another, but ultimately they are delivered stacked on top of or next to each other on single substrate that has the potential to offer superior performance and smaller footprints than traditional computers, with more flexibility than SoCs, and more fault tolerance (since any one die is smaller).
I wrote a bit about this approach of using “chiplets” last fall in the context of AMD acquiring Xilinx; Gelsinger’s argument is that Intel has the best packaging technology which, when combined with Intel’s vertical integration, will result in the most powerful Systems-in-package. As you will see, this is critical to understanding the headline announcements.
This is the part of the keynote that made all of the headlines; IDM means “integrated device manufacturer”, which is certainly what Intel has been for decades: the company sold chips that integrated its own designs with its own manufacturing. What is fascinating about calling the new strategy IDM 2.0 is that in many respects it is about introducing modularity into this model, both in design and in manufacturing.
Start with the actual IDM part:
That brings me to the core of today’s announcements, which is right at the heart of Intel: manufacturing. I am excited to share the next major evolution of Intel’s Integrated Device Manufacturer, or IDM model. I call it IDM 2.0. We have always been in the business of defying physics and building the future, and that’s not going to change. Intel is and will remain a leading developer of process technology, a major manufacturer of semiconductors, and the leading provider of silicon globally.
IDM 2.0 is the leadership combination of three factors: first, Intel’s internal factory network, integrated manufacturing, has been foundational to our success, enabling product optimization, improved economics, and supply resilience. As I mentioned on the January earnings call, we will continue to build the majority of our products in Intel.
Intel’s core business isn’t going anywhere. It is, though, going to be both an augmenter, and an augmentee. Start with latter:
Second, we will also expand our use of third-party foundry capacity across our portfolio to deliver the best products in every category that we participate in. Intel’s complementary and strategic use of outside foundries is an under-appreciated fact. Today we use foundries to manufacture many products, including chips for communications, connectivity, graphis, and chipsets. As we grow the business we expect our engagement with foundries to grow, in both size and scope. This includes manufacturing a range of modular tiles on advanced process technologies, including products at the core of our computer offerings, for both high end and date center segments. To further enable this aspect of our IDM 2.0 model, we are increasing our engagement with TSMC, Samsung, Global Foundries, and UMC, building on our existing long-term relationships. This will provide us with increased flexibility and scale we need to optimize our roadmaps for costs, performance, schedule, and supply, giving us a unique competitive advantage.
Gelsinger calls it “a unique competitive advantage”, I call it “paying the price for dropping the ball with EUV.” It’s good that Intel has (hopefully) corrected course with its manufacturing, but in the meantime it has high performance chips to sell. That is why for the first time it is out-sourcing the manufacturing of not just secondary items like communications chips or chipsets, but also its CPUs. Note, though, that Gelsinger refers to its CPUs as “modular tiles”: this is where Intel’s focus on — and, Gelsinger will tell you, superior technology for — Systems-in-Package will start to pay off.
Our IDM 2.0 model has an important 3rd element: today I’m announcing our plans to be a world-class foundry business and a major provider of U.S. and European-based capacity to serve customers globally. The digitization of every industry is accelerating the global demand for semiconductors at a torrid pace. But a key challenge is access to manufacturing capacity. Intel is in a unique position to rise to the occassion and meet this growing demand, while ensuring a sustainable and secure supply of semiconductors for the world. We conservatively size the foundry opportunity as a $100 billion addressable market by 2025, with most of the growth coming from leading edge computing, which is our expertise.
The majority of leading edge foundry capacity is concentrated in Asia, while the industry needs more geogrphically-balanced capacity. Intel’s advanced manufacturing scale, including operations in the U.S. and Europe is critical for the U.S. and the world, and we will significantly grow our global operations starting with increasing our U.S. and European-centric manufacturing capacity to serve the growing demand. We are committed to ensuring this capacity will support commercial customers, as well as address unique government and security requirements in the U.S.
Intel is finally going to make chips for other companies. What is fascinating about this introduction is how heavily Gelsinger leans into the geopolitics of chip production. Yes, there is a huge market, and yes, Intel is one of the only companies with the capability to pursue the cutting edge, but just as importantly it is the only one of those companies not in Asia. That, you will recall, is a big deal given the U.S. geopolitical rivalry with China; from Chips and Geopolitics:
Taiwan, you will note, is just off the coast of China. South Korea, home to Samsung, which also makes the highest end chips, although mostly for its own use, is just as close. The United States, meanwhile, is on the other side of the Pacific Ocean. There are advanced foundries in Oregon, New Mexico, and Arizona, but they are operated by Intel, and Intel makes chips for its own integrated use cases only.
No longer! Which is why Gelsinger included a strikingly similar map:
It is also notable that Gelsinger repeatedly referred to the E.U.’s need for chip production, in addition to the U.S.; the bloc pledged $150 billion earlier this month to “technological autonomy”, including increasing its share by value of semiconductors from 10% to 20%. Intel wants in.
Still, Intel has tried to be a foundry in the past, and failed miserably; that is why the details are incredibly important. Start with organizational structure:
To deliver this vision, we are establishing Intel Foundry Services, a fully vertical, standalone foundry business, led by semiconductor industry veteran Dr. Randhir Thakur, reporting directly to me. This business unit will be completely dedicated to the success of its customers, with full PnL responsibilities. This model will ensure that our foundry customers’ products will receive our utmost focus in terms of service, technology enablement, and capacity commitments.
The entire premise of Intel Problems from earlier this year is that (1) Intel needed to become a foundry to ensure it had sufficient volume to justify future investments in cutting edge technology, (2) the U.S. needed cutting edge foundries because of the China threat, but (3) I doubted that Intel could ever build a successful foundry business given the degree to which the company was integrated. That’s why I proposed Intel split, with the U.S. government as a guaranteed buyer for the foundry.
Gelsinger obviously didn’t go that far, but he is going as far as possible while keeping Intel whole. The fact that the foundry business is completely separate from the rest of Intel, with its own profit-and-loss line item and a leader that reports directly to Gelsinger is crucial. Intel Foundry Services has to be its own independent entity, and the most important job facing Gelsinger is ensuring Intel itself treats it that way, both in terms of guaranteeing capacity and access to Intel’s best IP, and otherwise keeping the rest of Intel at arm’s length.
We will be differentiated from other foundry offerings with a combination of leading-edge packaging and process technology, committed capacity in the U.S. and Europe available for customers globally, and a world-class IP portfolio that customers can choose from, including x86 cores, graphics, media, display, AI, interconnect, Fabric, and other critical foundational IP, along with ARM and RISC-V ecosystem IPs. Intel Foundry Services will provide access to silicon design servcies to help our customers seamlessly turn silicon into solutions.
I’ll forgive Gelsinger the corniness: what is important is the acknowledgment that building a foundry is about a lot more than fabs. One of the things that makes TSMC so attractive is that you are signing up for not just a manufacturer but a customer service organization that makes it easy to build chips. It is going to be very difficult for Intel to build a similar culture and library of IP, but the fact Gelsinger is talking about it at launch is critical.
As for IP specifically, Intel partners have long been concerned about the propensity of Intel to learn a little too much from its partnerships. What Gelsinger is promising is the opposite: Intel is going to make its best IP available to Intel Foundry Services, including x86 cores (alongside ARM and RISC-V). Moreover, this is another area where Intel’s focus on Systems-in-Package could be real advantage.
That x86 part is a big deal, by the way, as Gelsinger explained in the investor Q&A that followed the presentation; this was his response to a question about why this time would be different:
This isn’t Intel’s first shot at foundry. There was a big push in that direction seven years ago. What didn’t work back then and why is it going to work this time?
Thanks very much Mike. First I’d say the market is really different today. As you’ve seen there is extraordinary demand for semiconductors, strong interest on the part of U.S. and European institutions and governments. I’d also say that our first efforts were somewhat weak. We learned a lot through them but we didn’t really throw ourselves behind them. As you heard today with our major capacity investment, a separate organization, a separate PnL, reporting directly to me, we’re going after this much more aggressively.
Also, we’re bringing together the leading process and packaging technology, the best that we have to offer is going to be made available to our foundry customers, and a world-class IP portfolio of industry IP, but bringing together the Intel IP: graphics, AI, interconnect, and x86 cores. We’re putting it all on the table. Imagine if you were a major cloud service provider saying, “Boy, I have 10s of millions of cores that our running, and now I can optimize them for my business, and add some of my stuff and maybe take out some things that I don’t utilize?” This is a powerful strategy and we believe that now is the time for Intel Foundry Services and I’m commited to make this a huge success for Intel. It’s a key piece of our IDM 2.0 strategy.
Actually, this isn’t a big deal — it’s a massive deal. Intel’s unwavering commitment to being a true IDM has meant dictating processor design to its customers, realizing the profits from integrating design and manufacturing. The problem for Intel, though, and one of the reasons I have been so concerned about their volume in the long run, is that the quantity of their customers is dwindling: Amazon, Google, Microsoft, Facebook, Alibaba, and Baidu account for a huge portion of Intel’s sales, and in the long run those customers were going to want to create their own integration in the value chain — like Amazon designing its own ARM chip.
Intel is heading that threat off by giving up its integration: now Amazon or Google or Microsoft or any other big cloud provider can build their own custom chip, just as they can with ARM, but with x86. This is important because the problem facing Amazon and Graviton is that there is a massive ecosystem built around x86, which makes it hard to persuade customers to switch, even though Graviton is cheaper. Given that reality, why not just go with Intel?
Intel isn’t out of the woods, to be sure: what will the company charge for using x86 IP, and exactly how much will it let its customers mix-and-match said IP? You can bet that all of the people still dedicated to building and selling Intel chips will demand prohibitively high prices and strict restrictions; that, though, is why Intel Foundry Services being a separate organization is so critical, and it will be up to Gelsinger to prevent old Intel from scaling the wall.
There was a lot more in the keynote and Q&A, including a new partnership with IBM on future chip technologies, as well as Intel’s two new factories in Arizona, which will include dedicated capacity for Intel Foundry Services. It is worth noting that this is where Intel staying one company is beneficial: as long as the company can credibly guarantee that Intel Foundry Services will not be crowded out by Intel’s own business, sharing capex is a big advantage. I also appreciated Gelsinger’s insistence that Intel is making these structural changes without a guarantee of government assistance, either in the U.S. or in the E.U. It does seem clear, though, that Intel is banking on it.
That brings me to what was the most striking moment of the presentation; when I wrote Intel Problems, I was all but ready to give up on the company, arguing that the U.S. should not give Intel anything unless the company abandoned its model. As I noted, Gelsinger didn’t go all the way, but he went pretty far; more importantly, I finally understand why it is so many Intel employees believe he can pull this off:
I’ll say one of the themes to my young CEO tenure: execute, execute, execute. We’re bringing back the execution discipline of Intel. I call it the Grovian culture that we do what we say we will do. That we have that confidence in our execution. That our teams are fired up. That we said we’re going to do ‘x’, we’re going to 1.1x, every time that we make a commitment. That’s the Intel culture that we are bringing back. The enthusiasm we’re seeing inside of the company, the passion to execute. The commitment to leadership. That competitive zeal. And to collectively, as that emerges, we have confidence that we’re going to meet not just the existing customers, we’re going to be leaders in the market, and we’re going to satisfy the new foundry customers, because the world needs more semiconductors, and we’re going to step into that gap in a powerful and meaningful way.
This incredible answer was delivered in response to a question as to how Intel would manage to build its own business while keeping its promises to the customers of Intel Foundry Services. What if Intel fell short again, and had to choose?
This is where Gelsinger took things full circle, going back to Grove and Intel’s history. The thing about Moore’s Law is that it wasn’t a law at all: it was a choice, one that Intel made to push the entire industry forward year after year after year. That is why it was so important that Gelsinger invoked that attitude: Intel is in a huge hole right now, but they finally have the right strategy, and if this keynote is any indication, the right spirit leading the way. Now it’s time to execute, execute, execute.
This Daily Update will be available as a podcast later today. To receive it in your podcast player, visit Stratechery.
The Daily Update is intended for a single recipient, but occasional forwarding is totally fine! If you would like to order multiple subscriptions for your team with a group discount (minimum 5), please contact me directly.
Thanks for being a supporter, and have a great day!