Confronted by a disapproving board, Intel CEO Pat Gelsinger has retired. They brought back the former Intel CTO and company chauvinist in 2021 to revive it after years of poor execution. The company had all but forgotten how to develop processors and process technology and had seen growth opportunities sail past.
Because of the capital required, addressing process technology and other manufacturing problems was (and still is) the most important issue Intel faced. Moreover, semiconductor manufacturing is bananas. Today, a banana has appeal. Tomorrow, it is brown and must be discarded. Although timeliness is also important to products, customers will stick with a vendor even when its chips have fallen behind. The server-processor market offers proof: Xeon outsells Epyc despite the latter’s superiority.
How Manufacturing Affects Business and Industry Organization
- Business structure: Intel is an integrated device manufacturer (IDM), both designing and making products.
- Why are some companies IDMs? Other IDMs include Texas Instruments and Micron. Companies like TI don’t require the costliest, most advanced manufacturing equipment, tilting the make-versus-buy balance towards make. Moreover, control over process technology helps them deliver better-performing analog chips. Developing memory is inextricably linked to developing process technology, necessitating companies like Micron operating fabs. Further, because memory is a commodity, margins are too slim to split between separate design and manufacturing businesses.
- Why are some companies fabless? Starting in the late 1980s, the minimum efficient scale (MES) for an advanced digital fab exceeded many IDMs’ production requirements and certainly that of a startup. These IDMs shed their manufacturing and outsourced production instead, a trend that continued as the MES grew.
- Why does the fabless-foundry model dominate? In addition to the MES consideration, separating product from production enables fabless companies to shop around for the best manufacturing and shields foundries from product missteps that would leave their costly physical plant underutilized. While close collaboration between product designers and process-technology developers is theoretically harder under the fabless-foundry model, bad habits (e.g., using the fab as an EDA tool) are less likely under a formal fee-for-service arrangement between distinct entities.
- Do IDM fabs differ from foundry fabs? An IDM will optimize its fabs for throughput and utilization, such as by lengthening the manufacturing pipeline. Foundry customers require quicker turnaround times and smaller production runs, so foundries have shorter manufacturing pipelines and some slack to meet last-minute requests. Foundries will also focus on customer service, offer a variety of physical IP, and partner with third parties to provide a design-IP ecosystem.
What Gelsinger Did and Didn’t Achieve
- Operations—moving manufacturing to a discrete business unit (Intel Foundry Services, IFS) with public financial performance engendered accountability and broke some of the bad habits between how design and manufacturing interacted. Trying (again) to offer foundry services externally, prepared IFS to build a customer base big enough to support chipmaking’s now-huge MES.
- Process technology—The 5N4Y program indeed revived Intel’s process development. The company is at least nominally competitive with TSMC, compared with being stuck and falling behind when Gelsinger took over.
- Risk mitigation—the 5N4Y program and outsourcing some production to TSMC mitigated product groups’ risk. Intel 3 and Intel 4 backstopped each other, as did 20A and 18A. Backstopping everything was TSMC.
- Finance—the 5N4Y program, however, was costly. So, too, is outsourcing to TSMC while failing to fill internal fabs and building even more. In its most recent quarter, the company wrote down $3 billion of manufacturing assets, mostly for Intel 7. In addition to this noncash charge, negative free cash flow totaled $2.7 billion as the company spent $6.5 billion on capital. Gelsinger apparently didn’t plan for the situation in which the dog caught the car: technologically, each N succeeded, but there was no rationalizing the Ns to economically ramp production.
What Should Intel Do?
At Microprocessor Report, we advocated in 2011 for Intel to spin off manufacturing. In this era, the company had the best manufacturing and could generate more revenue if it could sell it to others. Yes, theoretically, it offered foundry services, but big chip companies wouldn’t rely on a competitor for wafers. Meanwhile, Intel was struggling to gain share in mobile-phone chips and maintain revenue growth. Thus, separating products and production would increase profits.
Separation still makes sense, and IFS’s formation was a necessary first step. Regardless of whether IFS stays captive or spins out, the transformation into a foundry competitive not only in technology but also in other business aspects will take billions of dollars and several years. To get the required billions, Gelsinger forged joint ventures with asset managers such as Apollo and obtained Chips Act grants. As for time, only the board could grant him that. They refused.