Intel Foundry Services

Intel's share price is down to ~$21 at the time of writing. At the end of last year, about 10 months ago, it was at ~$50. Things have not been going well for Intel for a while and this precipitous drop in share price is mainly a result of people only just realising, rather than because anything has changed very recently.

The most interesting part of Intel for me though is the split between the design and foundry businesses, so I wanted to jump into that in a bit more detail today.

Segments of Intel

There are five operating segments reported for Intel, along with an irritatingly unhelpful 'All other':

• Client Computing Group: designs chips for PCs, mainly desktops and notebooks
• Data Centre and AI: designs chips for data centres, communication service providers, networks and HPC customers
• Network and Edge: designs chips to run networks, moving them away from fixed-function hardware to general purpose, programmable hardware
• Mobileye: autonomous driving and driver-assist technologies
• Intel Found Services (IFS): manufactures chips, both their own as well as (hopefully) on behalf of other businesses
• All other: this is for revenue and expenses that cannot be charged to an operating segment, such as restructuring, some employee benefits, acquisitions, etc.

To give a sense of relative scale, the revenue and profit from each segment in 2023 is shown below. The percentages on the Profit column are all absurd, but that's what happens when you divide by such a small number.

'All other' is an annoying segment, accounting for an operating loss of $5-6bn without giving that much of an explanation. However it seems that it's not changing significantly.

Intel's manufacturing history

Most semiconductors 'makers' that we think of, including for example AMD and Nvidia, but also Apple and the hyperscalers, don't make anything at all. They design the chips, but then send those designs to another firm to manufacture. In most cases this is TSMC, who do not design their own chips and do not sell their own products, all they do is make chips for others. Samsung also do manufacture, but they use a lot of their capacity themselves, and they are also a competitor for a lot of the design firms and so are less trusted. I'll focus on TSMC here. Also note we're talking about leading edge chips, there are many more manufactures for 'trailing edge' (i.e. older technology) chips.

Historically, Intel has been different, they have both designed the chips in-house but they have also manufactured those designs themselves. This integrated approach for a very long time allowed them to make the best chips in the world, both because each half of the process (design and manufacture) was tailored to the other, and because Intel had the most advanced chip manufacturing capabilities in the world. Moore's Law was really a statement of how well they were able to make this work.

But in the late 2010's, Intel missed some key milestones in their ability to manufacture chips (often called 'nodes'). TSMC in contrast continued to deliver, and as a result their manufacturing overtook. Suddenly Intel's in-house manufacturing, which had been an asset, was now a liability. Anyone manufacturing with TSMC, which was basically everyone else on the leading edge, now had access to better processes and were able to design better chips.

Splitting out the foundry services

Intel hasn't split out their foundry business; they're still a single company both designing and manufacturing chips, but there has been an increasing move in that direction. First they created separate foundry P&Ls and more recently announced that the foundry would become a completely independent subsidiary.

This means that the design part of Intel can in principle choose between manufacturing with IFS or with TSMC or others, and in fact some of Intel's more recent products are manufactured by TSMC. It also means that other designers can have their chips manufactured by TSMC, assuming they're happy with the service and capabilities of that segment.

These moves have given some more insight into the economics of the foundry part of the business. Unsurprisingly this is where a lot of the capital expenditure ends up, and IFS accounted for a ~$0.5B operating loss in 2023, up from ~$0.3B in 2022. There are good reasons for these losses: Intel has to invest huge amounts of money to catch up with TSMC, but it still isn't pretty.

The goal with the split is to both create this economic clarity (no matter how unpleasant) but also to create the incentives and the structure for IFS to be successful in the future, forcing it to compete and keep up. Some of the impacts of this change that I can see are:

Advantage 1: Intel chips stay competitive

Having lost the lead on manufacturing capabilities, Intel's otherwise well designed chips were slowly becoming less and less competitive as the processes they were being manufactured on fell further and further behind the leading edge.

This is the most profitable part of this business, and that profit is what will allow them to weather the financial storm that the next few years are likely to be. Manufacturing chips with others, even at the expense of their own IFS, allows Intel to stay in the game while they sort themselves out.

Advantage 2: Higher leverage on fixed costs

Fabs are incredibly expensive to set up, but then the per unit costs of running them are relatively low. Once you've built a fab you want it producing as many chips, for as many years, as you possibly can. That effectively allows you to 'spread out' those upfront fixed costs.

But that's hard when you only have one customer (yourself). By making your fab available to others you increase the likelihood that, at any given time, all your fabs will be working flat out making something for someone.

Advantage 3: Increased competition

Intel's internal fab has always had business: regardless of what they did they would always be manufacturing Intel chips. I'm not suggesting they necessarily got complacent, but a 'fix this or the company dies' mentality might help them break through whatever's been holding back their technological processes.

Disadvantage: Reduced differentiation

The integrated solution always has the potential to be more differentiated than the modular one and in theory it was possible for Intel's internal foundry to deliver things that gave them a genuine competitive advantage.

But as is often the case, these advantages start to diminish over time. It seems unlikely Intel was ever going to win on the basis of manufacturing prowess again, but this move precludes that possibility because now that manufacturing capability has to be standardized, and available to everyone.

The national security dimension

There is another dimension to this which this split helps to clarify. The US needs, for national security, onshore leading-edge semiconductor manufacturing capabilities. TSMC is building some fabs in the US but ultimately if China invades Taiwan then TSMC doesn't exist any more. Intel is the only really viable option for truly secure, onshore manufacturing for the US (and in turn, for its allies).

The design part of Intel, which is the bit that makes money (albeit slightly less as time progress) doesn't add anything to national security. Plenty of other US firms have the design capability. Therefore the design bits both can and should survive by themselves.

The manufacturing is the part that is required for national security, and also the part that needs help. It should be treated separately, and it is IFS (not Intel as a whole) that the US government should aim to support. If it weren't for these national security concerns there's a strong argument that IFS be allowed to die, and Intel continue as a fabless company.