We have been getting asked a lot lately about the state of Intel. Since this topic seems to be of great fascination for so much of the semiconductor world, we thought it might help to put our thoughts down here.
Intel currently faces a series of sequential problems. Sequential in the sense that each problem has to be solved before they can truly address the next one. The company is working on all of them, but they are inter-dependent in such a way to confound the company's prospects until they are each resolved.
The first problem the company needs to solve is its manufacturing process. The company is built around its IDM model, with internally controlled fabs. And as much as they have moved some of their products to TSMC, 70% or so of their revenue still comes from their own fabs. The company fell off the Moore's Law path several years ago, and is now racing to catch up. Advancing 5 nodes in 4 years is the official slogan. This question is existential to the company and if they cannot right the cart, it faces a very difficult future. As far as we can tell from the outside, they are on track to reach that goal, but still have over a year to go before they could actually get these into production. Achieving this is incredibly challenging, but if we had to make a guess today we think they can at least reach some level which allows their products to be competitive again, albeit overtaking TSMC any time soon looks less unlikely.
Then we reach the second problem. Even if their manufacturing process reaches a 'good enough' level, then they have to compete with their products. Here they face a litany of difficulties. Top of that list is AMD which has been executing like the proverbial machine. AMD's latest CPU portfolio looks very compelling to customers. They have brought a large dose of innovation to the CPU market, such as their chiplet architecture and associated packaging. AMD argues that their products today are more performative and offer a better Total Cost of Ownership (TCO) than Intel, and that lead will only advance while Intel gets its manufacturing house in order, which again is not likely until deep into 2024 at the earliest. On top of that, the market itself is changing. In the data center, customers are moving away from CPU-centric systems to heterogenous compute comprised of CPUs, GPUs and accelerators. Intel has GPUs and AI accelerators, but they are not lighting the world on fire, to put it diplomatically. Our sense is that Intel is struggling so hard just to survive that they are not advancing their roadmap in a way that adapts to these new realities. And we are being polite, we speak with many who think the situation is much worse. Put simply, when Intel entered its current funk in the middle of the last decade, they had one CPU competitor, today they face a dozen. A revived Intel can face off all these challenges. They still have a solid position in the PC market and long-established relationships with the whole ecosystem, but they need products which can stand on there own.
And then we reach the third problem - they have to keep all of this going. That means investing heavily in advancing their manufacturing process. Even if they can achieve some version of 5 nodes in 4 years, they have to keep moving beyond that. The economics of Moore's Law are punishing, only companies with massive revenues can sustain the pace of investment required. This problem is deepened by the fact that Intel's fabs produce so much less than TSMC. In order for Intel to sustain the R&D pace it needs to fill up its fab with more than just their own products. This is a matter of compounding. Part of the reason TSMC has become the leader in semis manufacturing is that they produce so much volume, which means they learn faster than everyone else, a critical component of all of this. So in order for Intel to sustain itself in the future it has to build Intel Foundry Services (IFS) into a legitimate Foundry competitor. Customers will only move to IFS if they think it is competitive with TSMC, which is part of the reason we see this as a sequential series of problems. And while there are constant rumors that fabless companies are going to sign with IFS, we do not think any of that will become material until Intel has proven its manufacturing process is both competitive and sustainable. Our view is that IFS cannot contribute material revenue until the end of the decade.
At heart, the company's biggest challenge is cultural. The organization has to come to grips with the fact that the world has changed and the company needs to change with it. Over the years Intel created immense blindspots for itself by overly focusing inward, losing sight of the fact that they had long since ceased to be the giant they once were. So we cringed on their recent analyst update when they claimed that IFS is "The world's second largest foundry" because they included their internal products in that calculation. This is akin to taking your hot older sibling to prom, strictly speaking it is true but it does not really convey the message you think it does.
This may all look insurmountable, but there are a few factors working in the company's favor. They have immense internal talent, and from what we can tell they seem to be engaged and energized. Beyond that, they do not have to execute perfectly on all of these. For manufacturing, they do not need to exceed TSMC, they just have to get close enough that process is not working against their products as it does today. They do not need to be the second largest foundry in the world (excluding their own product), they just need to fill a few fabs with external customers.
Tough, but not impossible.
Photo by Long Truong on Unsplash
No comments:
Post a Comment