Facing persistent questions over its process technology roadmap, Intel unveiled its new IDM 2.0 (Integrated Device Manufacturer) strategy under the new CEO Pat Gelsinger. One of the crucial pillars of the IDM 2.0 strategy is Intel will create a new stand-alone business unit called Intel Foundry Services (IFS), which will have its own PnL model and will report directly to the CEO. Intel wants to capitalise on the increased appetite for semiconductor capacity in the wake COVID-19 pandemic and accelerated digital transformation across industries. Besides, we think the on-going trade tensions and security concerns weighed in Intel’s decision to step into merchant foundry role as most of the semiconductor manufacturing is now centred around just two countries - Taiwan and South Korea. In recent months, the US and Europe governments voiced increased support to bring semiconductor manufacturing back to their shores and likely support Intel with incentives. Intel foundry services received massive support from the industry, including Qualcomm, SiFive, ASML and others. Qualcomm’s endorsement of Intel’s foundry services is a good sign. Qualcomm is grappling with shortages and is looking for additional capacity to avoid future crisis.
Some key details of Intel foundry service business include,
- Intel will build two new fabs in Arizona with investments of $20 billion to serve both Intel and external customers.
- Customers can tap into Intel’s existing process technologies straight away, starting with the 22 nm. Intel will expand its offerings to leading-edge nodes later as they become available.
- The company will also open up its strategic 2D and 3D packaging technologies to external companies. Intel’s packaging technologies Foveros and EMIB (Embedded Multi-die Interconnected Bridge) offer the flexibility to mix and match various IPs from various sources. Intel will support third-party IPs, including Arm and RISC-V.
- Intel will also support TSMC, GlobalFoundries, UMC and Samsung Foundry IPs to allow customers to port designs across foundries.
Historically, Intel tried to open its fabs to external customers with its custom foundry program and managed to sign up customers including Spreadtrum, LG, Panasonic, Altera (later acquired by Intel), Achronix, Cisco and others. The company even licensed Arm’s Artisan physical IP to make headway in the Arm-based foundry market, dominated by TSMC and Samsung. However, Intel couldn’t attract high volume customers such as Apple, Qualcomm and MediaTek and gave up by 2018. Most recently, intel manufactured its own 4G modems for Apple iPhones, but Intel exited from the modem market.
The leading-edge foundry market has become a two-horse race between TSMC and Samsung. Intel’s foray into foundry services puts it in direct competition with the leading-edge foundries TSMC and Samsung. However, Intel will contract TSMC for its advanced CPU and data centre product manufacturing in 2023. In the technology industry, co-opetition - companies that compete in one area also partner in another - is a common practice. The world needs more capacity and Intel is offering help. Intel’s effort will not solve the on-going capacity issues. Intel will provide geographic diversity and more choice to the industry in the future. We think Intel’s foundry offering plan has more chances of success this time as governments also appear keen on supply chain security and continuity in the wake of on-going shortages and trade war.
While it is too early, we will have to wait for how Intel will address some of the pertinent questions in the future.
- This effort requires significant financial investments. Intel will not pursue this in its current situation without the right incentives as the company has enough challenges on its hands, including driving its CPU roadmap and accelerating 7 nm introduction.
- Intel is starting with a zero foundry service revenue now. In comparison, TSMC had over $45 billion revenue from its foundry business in 2020 and is likely to grow at 15 percent over the next 4-5 years. Also, TSMC had over 42 percent operating margin on its foundry business in 2020. None of TSMC’s foundry peers has this level of impressive financials. As Intel wants to be the leader in every market it competes in, can the company outperform TSMC in 5-10 years?
- Intel’s long-term commitment to its foundry business will be essential for non-Intel companies to decide whether to go with Intel. Intel needs to give a detailed product roadmap, investment and IP commitment to convince external chip companies to fill Intel’s fabs. Given the high stakes involved, Intel needs to look into this aspect deeper. Intel’s renewed focus on execution will help demonstrate its credibility in this area.
- Can co-petition work for Intel? TSMC doesn’t compete with its customers but Samsung does and is still thriving. Intel hasn’t outlined whether it will license TSMC or Samsung’s process technology IPs to manufacture its own or third-party chips in its fabs. Intel is expected deliver 7 nm process technology in 2023 and by that time TSMC will have moved to 2 nm. This dynamic could limit Intel's ability to capture high quality foundry business.
- Can Intel match the Capex investment of TSMC and Samsung and how this increased Capex will affect Intel’s ability to invest in its chip design R&D? TSMC will spend $25-$28 bn in 2021 alone.
- Will Intel eventually group all its fabs, assembly and packaging units into its foundry services business, not just the two new Arizona plants?
Sravan Kundojjala (@Skundojjala)