Battle of the Business Models

Global Foundries is one of those confounding businesses that seem to pop up every so often in technology. Everyone agrees they are critical to the supply chain, but no one is terribly excited about supporting them or investing in them. The US has become incredibly insecure about securing its supply chain for electronics. From a policy perspective this make Global Foundries with its domestic plants and years of experience producing chips, a critical vendor. But from a commercial perspective, their lack of leading manufacturing processes relegates them into a lower tier. All too often in these situations, commercial reality trumps grand ideas of strategy, leaving companies to fade away. Will that happen to Global Foundries?

A few months back we wrote about GloFo’s future not as a foundry, but as an analog chip company. This is not an exact comparison. The analog companies like ST Micro and Texas Instruments have direct relationships with end customers and thus much higher margins. But ultimately GF is serving the same market, making the same products. We think this is what trips up the policy people – they see GloFo as a semiconductor manufacturer and dream about how to make it competitive with TSMC and their advanced processes supplying chips for everyone’s AI fever dreams. This is never going to happen. Instead, GloFo is a maker of mixed signal analog chips just like TI and ADI. This is an important market, it is just not as glamorous as iPhones and ChatGPT.

Moreover, there is no easy way for Global Foundries to move up the value chain and start engaging end customers directly, That considerably narrows the company’s options. As far as we can tell, their best move is to become the world’s best manufacturer of analog chips, manufacturing chips for other people.

Wait. Don’t they do that already? And hasn’t most of the world already moved to fabless/foundry model? Yes and yes, but there is a big difference here. While the digital semiconductor industry shed their fabs and moved to a fabless model 20 years ago, the analog sector has not made that same transition. TI and ST Micro still manufacture about 80% of their own chips, Infineon and On about 70%, Renesas and NXP 40%. By comparison, the six biggest digital companies, with the exception of Intel, manufacture approximately 0%. A lot of GF’s business today is with digital companies who need the occasional analog part such as Qualcomm’s RF products, and less with the big analog companies.

GF’s future rests on its ability to bring the fabless revolution to analog.

Admittedly, this will not be easy. There are many good reasons why analog did not make this transition. Analog manufacturing is less about getting every bug out of the design as in digital and more about fine tuning parts to meet customer specifications. (Yes, we are oversimplifying.) This does not lend itself to economies of scale the way that digital fabs do.

Another big reason the analog sector did not adopt fabless business models was that the cost of these fabs are much lower. Global Foundries is spending about $2 billion in capex this year, TSMC is spending $40 billion. Analog plants also typically stay in use much longer with huge volumes still being done on very old processes.

Another factor in this reluctance is that the insertion of a foundry to the manufacturing process creates “margin stacking”, the foundry has to make a profit which cuts into the profitability of analog companies. We would argue that at some point the cost advantage of GF could eliminate such margin stacking. As a long time friend and reader regularly tells us – the bank does not accept gross margin percentages, it only accepts gross profit dollars.

That being said, the benefits of a fabless model still hold. GF can amortize its R&D and Capex across multiple customers, while the analog companies have only themselves to fill their fabs. With time, GF could theoretically gain sufficient scale to give it a meaningful cost advantage over internal, or IDM, manufacturing lines.

This could take a long time and poses significant execution risk. However, there is a looming shift in the industry that could catalyze this transition much more quickly. Today, many of the leading analog IDMs run most of their production on 200mm wafers. By comparison, the digital companies transitioned to 300mm years ago. The larger wafers offer much better economics in the form of more die per wafer, and thus higher throughput at the fabs. But converting to 300mm lines is a big expense, essentially requiring companies to build entirely new fabs which are going to cost more than their old, fully-depreciated fabs did. As a result companies are reluctant to make the transition, especially in today’s high-interest rate, high inventory environment. ADI, Microchip and On only run 200mm lines, Renesas and ST Micro produce less than 25% of their chips on 300mm, and only Texas Instruments are above 40%. Eventually, the big analog companies are going to have to transition to 300mm, especially when industry giant TI is almost half way there already.

This will be an important opportunity for GF to convince a few of those IDMs to outsource some of their production, and even a small amount of that will drive significant operating leverage to GF’s financials.

Leave a Reply