The past five years were rough for Intel (INTC -1.11%), the world's largest manufacturer of x86 CPUs for PCs and data centers. It fell behind Taiwan Semiconductor Manufacturing Company (TSM -0.69%) and Samsung in the "process race" to create smaller and more advanced chips, and its constant delays and chip shortages resulted in massive market share losses to AMD (AMD -1.65%).
Intel also abandoned the mobile market by scrapping its smartphone chips and baseband modems, and it made scattershot investments in programmable chips, Internet of Things (IoT) chips, and automotive chips -- none of which resolved the company's core problems.
Intel's former CEO, Brian Krzanich, abruptly resigned in 2018. His successor, Bob Swan, focused on cutting costs and repurchasing shares instead of resolving its pressing R&D challenges. Swan even considered outsourcing most of Intel's production to TSMC -- instead of upgrading its own foundries -- before being ousted in January.
Swan's successor, Pat Gelsinger, has dismissed the idea of Intel becoming a "fabless" chipmaker like AMD and doubled down on expanding its domestic foundries. The company is even reportedly mulling a takeover of GlobalFoundries -- AMD's former foundry unit -- to accelerate those efforts. Gelsinger expects that expansion will help Intel reclaim the process lead from TSMC and regain its lost market share from AMD.
Whether or not Intel can accomplish those lofty goals remains a divisive topic. But Intel recently revised its entire road map through 2025, and that update includes some jarring changes. Let's review the biggest modifications and how they could impact Intel's growth trajectory over the next five years.
Renaming the nodes
The process race is measured in nodes. Smaller nodes, which are currently measured in nanometers, are generally considered more advanced than larger nodes because they're more power-efficient.
TSMC started mass producing 7-nanometer chips in 2018 and 5-nanometer chips in 2020. Intel started mass producing 10-nanometer chips in 2019 after several years of delays, and previously postponed its next generation of 7-nanometer chips to late 2022 and early 2023.
It initially seems like Intel is two chip generations behind TSMC. Yet Intel's 10-nanometer chips are just as dense as TSMC's 7-nanometer chips, with roughly 100 million transistors per millimeter squared. In other words, Intel's 10-nanometer chips are technically comparable to TSMC's 7-nanometer chips -- but the node sizes (which are set by each foundry instead of a unified industry standard) still put TSMC in the lead.
But with its new road map, Intel is renaming its 10+ node, also known as the 10nm Super Fin node, as its "new" 7-nanometer node. These new chips, which will launch by the end of 2021, should offer better performance than TSMC's 7-nanometer chips, but fall short of matching TSMC's 5-nanometer chips.
Intel is renaming its old 7-nanometer node, the one that had been originally postponed, as its "new" 4-nanometer node to indicate it will outperform TSMC's 5-nanometer node. Intel still plans to launch the chips in late 2022 or early 2023, but that keeps it behind TSMC, which will start mass producing its 3-nanometer chips in the second half of 2022, and will potentially launch its 2-nanometer chips in 2023.
Intel plans to gradually catch up to TSMC by equipping its new plants with more high-end extreme ultraviolet (EUV) lithography machines. It also expects to be the first chipmaker to use next-gen high-NA EUV machines, which are required to produce smaller chips beyond the 3-nanometer node. It claims to be working closely with ASML, the world's only producer of EUV and high-NA machines, to secure those orders -- but ASML also supplies the same machines to TSMC.
Intel plans to launch its first 3-nanometer chips in the second half of 2023. It expects this new node to provide 18% better performance than its previously planned 3-nanometer chips.
Intel expects to start producing 2-nanometer chips, also known as its "20A" (20 angstrom) chips, in 2024. These chips, which will replace Intel's old 5-nanometer process, could be nearly twice as dense as its 3-nanometer chips. In 2025, it will launch its 18A (1.8-nanometer) chips. We don't know much about these chips so far, but the company believes its 20A and 18A chips could help it reclaim the process lead from TSMC and Samsung by 2025.
Intel's fab teams in Arizona, Ireland, Israel, and Oregon are all preparing to produce 4-nanometer, 3-nanometer, and 20A chips. That acceleration, which will likely be supported by subsidies in the U.S. and Europe, could significantly expand its own contract chipmaking services and help fabless chipmakers reduce their overwhelming dependence on TSMC and Samsung.
But can Intel catch up to TSMC by 2025?
It's unclear whether Intel can achieve its ambitious new goals, but its new road map indicates it's dedicated to keeping pace with TSMC and Samsung by expanding its capacity and ordering more high-end machines from ASML.
I doubt Intel can reclaim the process lead by 2025, especially as TSMC aggressively ramps up its own spending to maintain its edge, but the fundamental improvements to its foundry business might help it avoid future delays and shortages, as well as reclaim some of its lost market share from AMD.