Over ten years ago we read a really interesting piece of financial research from Jeffries analyst Mark Lapacis. This note contended that as the cost of building leading edge semiconductors fabs increased steadily, the number of companies capable of keeping up with Moore’s Law would diminish. Even in 2012, this was not exactly news, but few others really understood just how far reaching this conclusion wold end up.

Most intriguingly the team at Jeffries quantified the point at which companies would exit the market for leading edge semis manufacturing. Their rough rule of thumb held that a company’s annual revenue had to be double the cost of building a latest-generation fab, below that it became increasingly challenging to stay in the game. And once companies broke through 0.8x revenue/leading-edge fab they always exited. The team had a range of historical data to prove this point. In 2012, there were still people who remembered when TI, Panasonic, UMC and many other companies had once had leading edge fabs in their time. By 2012 all of them were out of the race. At the time, the question of the day was whether TSMC would be able to stay in the mix, that’s how long ago this note was written.

Of course, today we know how that story turned out. But we come back to this note frequently, and have updated the data to show where the industry stands now.

How to read this graph. The dotted black lines toward the bottom show the estimated cost of building a leading edge fab (the lower line) and a line showing double that number (the upper line). Our thesis is that companies whose annual revenue fall between those two lines are at risk of falling off the Moore’s Law treadmill. This is evidenced by a number of companies at the bottom who have all long since given up building leading edge fabs.

Not surprisingly, the only companies still operating in the comfort zone are Intel, Samsung and TSMC.

It is interesting to see how close TSMC came to falling off the curve. Only in hindsight can we say they were destined to lead, but for many years it was an open question as to whether they could stay in the business. Their big break came with the 28nm node, but they had to really struggle to get to that point.

Samsung is also interesting. Note that here we using revenue for Samsung Electronics’ DS Segment which includes memory, LSI, foundry and a few other items. We think there is a low likelihood of the broader Samsung group falling off the Moore’s Law curve due to a lack of revenue. If we plotted all of Samsung Electronics’ revenue the line would be literally off the chart. However, we do think it is telling that the semis side of Samsung is falling into the danger zone. We often struggle to understand Samsung’ investment decisions around semis. They seem to exhibit a high degree of indecisiveness. For example, last year they first said they were going to invest heavily in memory capacity, roiling the whole memory complex, only to reverse their decision a few months later. Or their Exynos mobile applications processor, a product they keep making even while the handset group eschews their internal part for more Qualcomm. And perhaps the most pertinent example – what exactly Samsung plans to do with their foundry business? Why does it always seem to be a little behind on advancing its manufacturing process? Of course the answer to those questions come down to company and chaebol internal decision making processes, but this graph puts those questions into context. The Samsung group struggles with its semiconductor business because it is so capital intensive and the return is on the border line where companies with fewer resources would consider exiting. To be clear, we do not think Samsung will give up, but the semis group will remain under pressure as long as the economics are this borderline.

This brings us to Intel. As we have mentioned in the past a key question for Intel is the future of Intel Foundry Services (IFS). This graph highlights the problem. For years, Intel operated well above the danger zone with revenue well above double the cost of building a leading edge fab. In recent years, they seem to be headed right into that zone of death. The Street (whose consensus numbers drive the 2024 and 2025 revenue figures above) thinks they will turn the corner this year and start heading away, but this is by no means guaranteed. And even if Intel can turn revenue in the right direction, they face a serious challenge. Assuming they complete their “five nodes in four years”, how will they be able to afford the sixth node, the seventh, the tenth? We would argue that the only way they can stay on the Moore’s Law treadmill is for IFS to succeed and drive enough revenue to keep the company in the fight. If we extrapolate out all these numbers a few years, Intel risks falling back into the zone by 2028 or so. Obviously a lot can change before then, but those forces work in both positive and negative ways. Our contention is the fate of IFS cannot be known until the end of the decade. By then IFS will need to be generating $5 billion in revenue at a minimum to keep the whole thig moving. Next week Intel will host an event showcasing their plans for IFS. We imagine this will highlight their technical achievements and aspirations. However, the ultimate question is one of culture – can Intel create a customer service culture capable of attracting external customers. This kind of cultural shift takes years to play out, and by the time they can achieve it they will need it if they want to stay in this game.