Intel CEO: Reliance on Asia for Semiconductor Manufacturing Isn’t ‘Palatable’

Intel CEO Pat Gelsinger wants to shift the balance of power in the global semiconductor industry out of Asia. This newfound push for a more balanced global supply chain is part of Intel’s expansion into the client foundry business and the company clearly hopes to capitalize on the perception that current supply chain issues are the result of building too much capacity in the Asia-Pacific region.

“Having 80 percent of all supply in Asia simply isn’t a palatable manner for the world to have its view of the most critical technology,” Gelsinger told the BBC. ” Every smartphone, every telemedicine, every remote worker, every remote education, every autonomous vehicle, every aspect of humanity is becoming more digital. This is the heart of every aspect of human existence going forward. And the world needs a more balanced supply chain to accomplish that. We’re stepping in.”

Gelsinger also noted that Intel intends to open at least one new facility in a new European country as part of its buildout efforts over the next 3-5 years.

Intel Is Spinning the Semiconductor Shortage

According to Gelsinger, Intel’s recent decision to enter the client foundry business in earnest was driven by its desire to help the world. Seriously.

When asked why Intel would make chips for other companies and thereby “help the competition,” Gelsinger reiterated that manufacturing is a capacity game and that foundries that can’t manufacture enough chips can’t exploit a large enough economy of scale to sustain the cost of chip manufacturing. There’s real truth to that. One reason Intel conquered the server and workstation markets of the 1990s was by taking advantage of its superior economy of scale.

But Gelsinger goes on, saying: “The world needs more semiconductors, and we’re one of the few companies that have leading-edge technologies to be able to step into that void. And to keep it for our own products simply is not the right thing for the planet, for the industry, and for a globally distributed supply chain.”

That’s honestly a bit much.

Companies that ought to know better are pretending that just building more foundries is a magical way to avoid this problem in the future. We’re stuck in the middle of a semiconductor shortage this year because the pandemic disrupted buying patterns and sent demand for semiconductors skyrocketing. It takes months, even years, to bring new semiconductor manufacturing capacity online. AMD, Nvidia, Qualcomm, MediaTek, Broadcom, and other chip designers aren’t captives, exactly, but moving a chip design from one foundry to another is a 6-12 month affair.

The idea that building more capacity now will prevent or even mitigate a future calamity relies on four false assumptions. First, it assumes that companies will keep a meaningful percentage of their capacity idle most of the year to absorb emergency demand. Second, it assumes that every foundry is capable of building every chip. Neither of these things is true. Semiconductor foundries face enormous fixed costs and operate as close to full as possible. Third, it assumes companies can shift from one manufacturer to another. Moving a chip from Samsung to TSMC doesn’t just take 6-12 months, it requires the efforts of a new CPU design team, at a cost of tens to hundreds of millions of dollars. Fourth, it assumes that most foundries can build most chips and that we just need extra “capacity” to make that happen. Leading-edge fabs and smaller facilities both specialize in certain types of technology. A company that needs to build RF silicon can’t just slap their chip on the same 5nm node TSMC would use for AMD or Nvidia.

Everything in semiconductor manufacturing is slow. Teams at Intel, Nvidia, and AMD are working on products right now we won’t see in-market until 2023 or 2024. The new capacity expansions announced by TSMC, Samsung, and Intel won’t come online until several years from now, and they’ll be scheduled as close to capacity as possible once they do. If a different event causes a similar surge in semiconductor demand in 2030, we’ll find ourselves right back where we are now.

This shortage, and the ongoing US-China trade war, have kicked off a flurry of silicon nationalism. Suddenly, the EU wants 20 percent of silicon manufacturing by 2030. Samsung and TSMC have announced they’ll build facilities in the United States. Intel is building $20B worth of fabs in Oregon and looking to expand internationally. There are plenty of good reasons why the United States should consider silicon manufacturing a strategic capability, but the idea that we’re suffering now because 80 percent of silicon manufacturing is in Asia, specifically, is false. The silicon manufacturing industry is hyper-optimized. CPUs from Intel and AMD already circle the globe multiple times, plural, before you buy them off the store shelf at Newegg or Amazon. Only a far more resilient supply chain comprised of multiple companies with easily available capacity expansion (or a tremendous inventory of important components) would have made a difference here. In order to truly work, such resiliency would have to be incorporated into every stage of manufacturing, from initial mining of raw materials of growth of the silicon boule, right down to when the hardware hits the store shelf.

Intel’s plan to simultaneously revitalize its own manufacturing while opening the doors of a client foundry is a radical departure from anything the company has done before. But talk of saving the global supply chain as part of responding to this present crisis is disingenuous. Intel definitely wants to expand its own manufacturing base and catch up to its competitors, but it’s not pursuing those goals in a way that would prevent the semiconductor shortage of 2021 from happening again under similarly disruptive circumstances. Neither, as far as we can tell, is anyone else. Previous efforts to create common platforms for cross-porting designs, like the Common Platform Initiative between IBM, GlobalFoundries, and Samsung, were not continued. There is no analogous method for porting designs between TSMC and Samsung today, and none of the three companies have yet announced an intent to create one.