Big news this week in the Land of Semiconductor Finance – Global Foundries filed for its IPO. This is exciting stuff, for a very small audience maybe, but still fun. We will try do a deep analysis of its filings somewhere down the line, but here we wanted to share our first impressions.

First, some background

GF is a foundry, they physically manufacture chips designed by other companies. In this sense they are comparable to TSMC, Samsung foundry, or the nascent Intel Foundry Group. However, as we shall see there are some important differences from those industry leaders. GF began life as the manufacturing side of AMD, maker of CPUs and GPUs. But like almost every other company with internal fab capacity, AMD eventually found they could not afford to keep up in the arms race that is the corollary of Moore’s Law. To save the company, AMD split in two. The fab lines became Global Foundries and the design side continued on as the company we today call AMD. GF was acquired by the Mudabala, the sovereign wealth fund of the United Arab Emirates. Mubadala has poured a lot of money into GF, but several years ago, GF realized that even as a stand alone company they could not keep up with Moore’s Law, effectively abandoned the race, and stopped developing new manufacturing process nodes. Freezing themselves two generations back. This is what sets them apart from the leading foundries we listed above who continue to move ahead in shrinking the chips they produce. GF instead has focused on specialty products such as Silicon on Insulator (SOI) and other niche-y areas of semis manufacture.

The Financials – Warts and All

If you read through GF’s financials, first impressions are not good. No profits, declining revenues, negative gross margins.

For over a decade, those of us in semiconductor land have looked longingly at the valuation multiples the stock market afforded Internet and social media companies. “We are seeing all time record high margins,” we argued, “industry consolidation and we are literally pushing the boundaries of the laws of physics but we traded at tiny fractions of gaming and messaging companies with a few dozen employees”. Reading GF’s financials make us think their solution to that problem was to do away with the profits entirely, and thus capture those crazy Internet multiples. Is GF the first semiconductor meme stock?

Dylan Patel articulates the bear case for GF pretty well here.

That being said, dig a little deeper into their filings and that view starts to look like the proverbial half-empty glass, and there is definitely a glass half-full story here.

First, take a look at cash. Despite those ugly margins, the company has been generating cash from operations for the past three years. Yes, there is a lot going on in that Statement of Cash Flows, but this should be a clue that there are some positive forces at work.

Second, capex has been declining. This is what kills foundries – the need to keep investing billions of dollars to stay on the Moore’s Law treadmill. By stepping off GF has been able to greatly reduce its capital needs.

Third, customers. When GF and AMD split, AMD committed to using GF as its foundry for several years. Those commitments have expired and AMD has been steadily moving its business elsewhere. So GF is painted as losing its biggest customer. On the other hand, GF has attracted a lot of other customers.This argues that the company’s strategy of specialization is actually working.

Fourth, customer commitments. Not only is GF able to attract new customers, but those customers are paying up for the privilege. Customers have committed $19 billion to GF to guarantee access to future capacity. GF has essentially gotten its customers to foot the bill for its capex needs for the foreseeable future.

As Benedict Evans would say this is a narrative violation. How is GF able to gain new customers even as it abandoned advanced process nodes?

We regularly get asked by investors about the foundry selection process for chip design companies. The answer usually surprises them: the first criteria is process node, a factor which weighs heavily on the decision, but the second factor is what we loosely call “customer support”. We would argue that this is more important than price in making this decision. Customer support covers a lot of territory, but comes down to how much customization the foundry will dedicate to a single customer’s manufacturing needs in terms or mapping out production flow and engineering support to get to production and then debug the process to get to commercial yields. This is complicated with a lot going on, but for our purposes here we think it is reasonable to conclude that GF is doing customer support well.

Starting from this position, and then factoring in the very real supply constraints across the industry, building a bull case for the stock is possible. Of course there are a lot of factors at play. GF’s F1 filing is dense with many, many moving parts, but there is more going on here than good vibes and a bubbly market.

The big question for every IPO is the valuation the company will seek. Intel was rumored to be interested in buying GF, and apparently the asking price was $30 billion, which is definitely getting into meme territory. We will see what the Market thinks soon enough.

Despite all this, we come away somewhat favorably inclined to the company. The stars are aligning for them and their strategy seems to paying off, or at least moving in the right direction.

We think the biggest problem the company faces is much more long term in nature. They clearly have business lined up for many years, so in the near term they should be fine. The lack of leading edge processes does not hamper them…for now. That being said, eventually this will become a problem, because eventually other foundries will catch up. GF is not all that far ahead of Chinese fabs like SMIC. And while we not that optimistic about China’s ability to catch up with TSMC, they will be able to catch up with GF. Let’s say that takes five years. GF can do a lot of good business before then and stabilize its financials, but at some point in the future the market will start to work against them. The good news is that they have time to work that out. And as long as we are idly speculating, by the time we reach that point, we will know a lot more about how Intel’s foundry business is faring. We have said repeatedly that Intel Foundry’s prospects are severely limited by their inability to provide customer support. They are not going to solve that in five years, so it might be nice if they could acquire someone who was very good at customer support.