Components > Handset Components Blog

Intel's IFS Holds Promise but Challenges Galore

by Sravan Kundojjala | 2月 21, 2022

At its investor meeting last week, Intel outlined its strategy, financial targets, IDM 2.0 foundry strategy, new reporting segments, and product roadmap to reassure investors that Intel will turn around and come back strongly in 2025-26. However, Intel acknowledged there would be a lot of pain along the way with heavy CapEx investments, market share losses and lower gross margins. We analysed Intel's investor meeting 2022 in a separate report.

This blog post will focus on IFS (Intel Foundry Services) business and its prospects in the global foundry market landscape. As part of its IDM 2.0 strategy, Intel created the IFS business to capitalise on the sentiment of geopolitical battles and supply chain security. Semiconductor shortages further pushed the US and European policymakers to warm up to Intel's messaging on the need to reduce dependency on Taiwan and South Korea, where most leading-edge semiconductor manufacturing is concentrated today. Intel's pitch has been clear – the current concentration of Asia-centric leading-edge manufacturing is risky, and Intel will step up to build a western alternative to TSMC and Samsung.

Building a brand-new foundry business is no trivial task in the current environment, especially given the increased CapEx requirements. Most foundries dropped out of the process node race, and only TSMC and Samsung have been able to transition to 5 nm and below process technologies. Process technology, in particular, had always been Intel's strength, but the company ceded the lead to TSMC and Samsung as a lack of timely EUV tool adoption held back Intel.

As part of its IDM 2.0 strategy, Intel will adopt ASML's EUV tools and hopes to catch up with its foundry rivals by 2024 and achieve leadership by 2025. In addition to geopolitical motives, Intel sees a growth opportunity in the foundry business. Intel covets TSMC's impressive financials (53% gross margin and 38% operating margin in 4Q21). In addition, Intel's core CCG (Client Computing) and data center businesses have reached maturity in terms of growth rates, and the foundry offers an attractive opportunity to expand. TSMC, for example, expects its revenue to grow 15-20 percent long-term.

Intel, an IDM (integrated device manufacturer), manufactures 80 percent of its product volume in-house. Intel's foundry partners include TSMC, UMC and Samsung. Intel usually keeps high-value PC and data center CPU manufacturing in-house while outsourcing Wi-Fi/Bluetooth, GPU, networking, MobileEye, chipset etc. to third party foundries.

Intel needs to catch up with TSMC first and build more capacity to continue this strategy. Intel has seen a high utilisation rate of its fabs in the last four years, driven by strong demand for its products, especially PC.

IDMs, for example, account for 15 percent of TSMC's total revenue ($8.5 billion in 2021). Intel will increase its outsourcing from third parties in the next five years as its HPC/GPU business grows from $1 billion in 2022 to $10 billion in 2026. Intel is also likely to test TSMC's N3 for its client CPUs. Subsequently, we think Intel's contribution to TSMC will peak in 2024 before ramping up in-sourcing again from 2025.

Scaling IFS will not happen overnight. Intel understands this. In addition, the company knows it needs external help. Unsurprisingly, Intel has been in the market for an acquisition (rumoured to have considered GlobalFoundries in 2H 2021) to boost its foundry credentials and customer relationships. Intel zeroed in on Tower Semiconductors. Tower, a speciality technology foundry (CIS, PMIC, RFFE, MEMS, fingerprint sensor, microLED displays etc.), grew 19 percent in 2021 to $1.5 b in 2021. Tower has 200+ customers and two decades of foundry experience. However, the company stuck at around $1.3 b revenue from the past six years (2021 an exception).

Tower's speciality technology business is a minor player compared to GlobalFoundries, TSMC and UMC's. TSMC, for example, has $13 b revenue from mature nodes (50%+ from speciality) and spends $3+ b CapEx on speciality technologies. UMC and GloFo, on the other hand, get up to 70 percent of their revenue from long-term agreements (LTA). It is not easy to dislodge customers from those established rivals overnight and scale Tower's business. Tower, however, seems to have some exciting technologies such as (Silicon photonics and microLED display). TSMC, for example, is investing in microLED displays for AR/VR applications, and Apple may use TSMC. Tower could be a competitive alternative if AR/VR emerges as a significant opportunity.

Tower has an impressive list of customers, including Nuvoton (25%), Qorvo, Skyworks, Broadcom, On Semi, Infineon, InPhi (Marvell), Analog Devices (Maxim), Intel etc. With Intel onboard, there will be no shortage of CapEx for Tower. However, CapEx alone cannot scale the business. We see Tower as a modestly growing asset, but it can be more profitable under Intel.

Coming back to IFS, most of Intel's CapEx will primarily be for internal needs only for the next three years. Intel shipped 2+ m wafers in 2021 and hopes to improve its wafer capacity by slightly more than 50 percent by 2026.

We estimate that Intel will spend more than $160 b CapEx between 2022 and 2026 (inclusive). The company conservatively assumed 10 percent offset from subsidies, customer prepayments and financing partnerships (Brookfield Asset Management). However, Intel hopes its CapEx could be offset by 30 percent, thanks to abundant capital from various sources. If we assume a 20 percent CapEx offset, Intel could get $3+ billion in subsidies per year through 2026.

Intel, however, will have to contend with some limitations in the initial stages. The company will be offering limited flavours of its process nodes (Intel 16,3 and 18A) to its customers, unlike TSMC and Samsung. In addition, Intel indicated that foundry customers would have to share Intel's factories initially, at least until the company starts to have business from big customers such as Qualcomm who may demand dedicated lines. We think dedicated lines will be necessary. Samsung has had issues with customer trust in the initial days as some customers feared Samsung might prioritise its own business over customers' business. However, over time, Samsung has improved in this aspect and built multi-billion-dollar foundry businesses with customers such as Qualcomm and NVIDIA. Also, we note that Samsung started its foundry business in 2005 and landed Apple as a customer in 2007. The Apple-Samsung relationship continued until 2015 before TSMC took over. The other challenge is that Intel is seeking customer commitment years ahead of process node launch. For example, customers must commit to Intel 18A by the end of this year with prepayments, but Intel 18A won't be manufacturing ready until 2H 2024. On the other hand, TSMC's customers only commit to a new node 12-18 months ahead.

Intel is building multiple fabs and shells and will start to fill the empty IFS shells with tools from 2025 onwards. By 2026, Intel will spend as much on IFS tools as it spends on IDM tools, reaching an equilibrium. Intel hopes to leverage existing IDM infrastructure (tools, space and R&D) to improve the IFS business's profitability. Intel guided its IFS business margins will be similar to the foundry market leader's (TSMC has 50% gross margins). We think this is overoptimistic, at least until 2026. However, in the last year or two, all major foundries improved their gross margins on the strength of pricing power.

So, how much revenue IFS will contribute? Intel sees three components to its IFS business – existing IFS contracts, Tower and future leading-edge (Intel 16, 3 and 18A). IFS contributed approximately $900 in revenue in 2021 and will grow to ∼$2 b in 2023 and ∼$3 b in 2026, per Intel. If we include Tower revenue, Intel will have $3.5+ b foundry revenue in 2023 and $5+ b by 2026. Any additional revenue from potential leading-edge customers such as Qualcomm will be a bonus. IFS could end up as the sixth-largest foundry in 2023.

Intel's existing IFS customers include Amazon and the US Department of Defense. Cisco will adopt as well. Intel will manufacture 30+ test chips this year and is positioning its assets to chase the smartphone, automotive and data center foundry markets. We note that some of Intel's foundry segments (RFFE and auto) are primarily controlled by vertical players. For example, the major European automotive players manufacture most of their chips in-house (60%+). Even TSMC couldn't make headway into these markets. However, with the help of Tower, MobilEye and IFS investments, Intel could make a dent. However, it will be a long journey.

All in all, Intel put together a coherent IFS story so far with ecosystem building efforts including partnerships (EDA, IP, foundry and fabless companies), Tower acquisition a competitive roadmap (5 nodes in 4 years). In addition, Intel hopes to drive the chiplet standards with its IFS effort. Furthermore, Intel's product groups get a considerable boost with increased capacity and an accelerated process roadmap. It's still early days for IFS, and we will closely watch Intel's progress.

On a related note, Intel has been sending a signal to the market in recent weeks that it will monetise its growth assets, similar to the MobilEye IPO effort. Intel fell behind peers AMD and NVIDIA in valuation and is trying to reorganise its businesses to better valuation. In recent years, Intel divested multiple businesses, including NAND memory, 4G/5G basebands, Home Gateway, data center PMIC, RealSense camera, wearables, Sports group etc.

PSG (ex-Altera) and Networking & Edge businesses may be likely candidates for future monetisation efforts. Even IFS could be a candidate if it gains momentum. However, this SoftBank-style approach could be risky as Intel will lose its essence by detaching growth assets from its core. Therefore, Intel needs to invest in long-term rather than spin-off growth assets.

Follow me on Twitter or connect with me on LinkedIn. You can share your thoughts and comments directly if you would like to.

Sravan Kundojjala
Previous Post: NVIDIA-Arm: Now What? | Next Post: Smartphone Memory Market Posts Double-Digit Growth in 2021

Let's talk

Now you know a little about us, get in touch and tell us what your business problem is.
Inquiry / Message:

please enter captcha from left