Earlier this week Intel hosted an investor event to provide an update on their key strategic goals. They said a lot of the right things, but they also proposed a timeline far enough in the future to leave investors uncertain, with the stock closing essentially unchanged. We want to give the new management team the benefit of the doubt, but we remain fairly cautious about their outlook.
Intel has three key things they want/need to accomplish:
- Close the gap on the manufacturing process – they are several years behind TSMC and Samsung
- Open up their manufacturing plants (fabs) to third party customers – create a foundry business
- Invest in new technologies and regain their technology advantage.
On the call they had updates on all three, in reverse order.
On that last point, they unveiled several new ways of building chips. We will not get into the technical details (we do not fully understand the technical details), but put simply this is classic Intel innovation. Lots of good ideas, pushing the envelope on what can be built in silicon. As will be a recurring theme, these new technologies will come out over the next few years.
However, the first two goals are the most important, and they are intertwined. Intel really needs to be able to provide foundry services to others. This is only way they can maintain the manufacturing scale sufficient to afford the ongoing investment required to be in this business – with new chip plants costing ~$20 billion each. As we have described in detail, for Intel to become a foundry provider they need a massive overhaul of their culture. They are attempting to do this by setting up a separate organization that reports directly to the CEO. It remains highly unclear how this part of Intel will work with the rest of the company, especially with the tightly knit organization that controls operations at Intel’s fabs.
On the plus side, they did announce two new customers for their foundry. At first glance, they are impressive logos – Amazon’s AWS and Qualcomm. AWS makes a lot of their own chips now and clearly has ambitions to build many more. Qualcomm is one of the largest foundry customers out there, and a one-time Intel rival. Unfortunately, the devil is in the details. AWS is signing up as a customer for Intel’s packaging factory. Intel does packaging well and has some highly sophisticated ways to put completed chips into plastic packages which can then be inserted in electronics. But packaging is not a great business, with low margins and questionable sustaining value for all those innovative ways to do it. By contrast, Qualcomm is on board for actual chip manufacturing but only for Intel’s “20A” process, which you may be able to guess, is not set to come out until 2025.
And this brings us to goal #1 – closing the manufacturing process gap. Foundry customers are going to be most interested in leading edge processes. There are plenty of lower-priced, easier options for trailing edge, so for Intel to succeed in foundry, and really for Intel to survive at all, they need to catch up with TSMC. Intel now says they will catch up to TSMC by 2024, and overtake them in 2025. This is realistic (at least more realistic in comparison to what they have been saying for the last seven years) but by no means guaranteed. TSMC is not standing still and Intel is still struggling with older nodes and a legacy of missed opportunities.
As a side note, Intel also renamed their process nodes. For years the industry has given these quantitative measurements – 10nm, 5nm, 2nm, etc. These measurements stopped being meaningful a decade ago, as the distance they are purportedly measuring no longer scale in this way. So Intel is going to start labeling things based on their own code names. This makes sense, but is also totally useless unless everyone else starts using Intel’s proprietary nomenclature, and so we are likely going to see a continuation of the ‘nm’ naming. In this light, Intel’s “20A” process, with which they plan to overtake TSMC and win Qualcomm, is effectively comparable to TSMC’s 2nm process.
So the question remains, why will customers sign up for Intel’s Foundry if the technology roadmap is still somewhat…aspirational. The answer is largely marekting – waving the flag to show support, signalling patriotism to Washington, and acquiring a low-cost option for goodwill should Intel actually achieve its goal. We suspect we will be hearing about more customers this year, but all on a somewhat tentative, long-term, low-commitment basis.
Intel also still has to deal with problems with its legacy production processes – their 10nm process which is basically two generations behind TSMC is still yielding abysmally. So Intel has to fix two legacy processes while at the same time pushing ahead on the new ones. This is a tall order.
As we noted at the outset, we want to give the new management team the benefit of the doubt. They are saying the right things and making some good high-level changes. It will be some time before we can see clearly if/how those changes are percolating through the culture. We are not optimistic.