The summer Intel fell behind

In the summer of 2020 Intel seemed poised for triumph. Then it all went wrong. “To put it briefly: Intel blew it,” explains Glenn O’Donnell, a research director at Forrester Research. Intel was forced to announce to the world that it’d be delaying its next major manufacturing milestone for its chips for another few years, an admission that, once again, Intel was falling behind the competition.

After years of misplaced bets, manufacturing delays, and changing leadership, the previously undisputed chip-making king has found itself faced by competition the likes of which it hasn’t faced in decades — while simultaneously finding itself at what could be the company’s nadir.

In 2020, Intel was forced to admit that it would be severely delaying its 7nm node, recently rebranded as Intel 4. This precipitated an exodus of leadership and the admission that Intel might have to face the unthinkable and outsource its own manufacturing.

Intel managed to survive its last summer hell, at the cost of being forced to dramatically rethink its business from the top down. Now it has a new CEO, a new plan, and a market hungry for more chips. It’s set ambitious goals, revealing its most detailed process roadmap and a bold promise to jump ahead and reclaim processor leadership by 2025, if it can avoid the familiar pitfalls of the past decade. The coming years will be a make-or-break moment of redemption that will either right the course — or send Intel into what might be its final tailspin.

“It wasn’t something that just happened suddenly. These things build, they build and build over time until something happens that causes the whole thing to kind of crumble,” says O’Donnell. Things have been headed in a downward trajectory for Intel for years, yet it was last summer — and the delay on 7nm — when the cracks in Intel’s facade started to look a lot more like canyons.

“For all of its history, until recently, manufacturing was one of the magic things that was going on at Intel — part of the secret sauce of Intel was the manufacturing innovations, and that’s why 7nm was such a big embarrassment,” O’Donnell says. “It was a huge hit, psychologically, across the company, but also across the market because that meant Intel no longer had that secret sauce.”

The fact that Intel still makes its own chips today is a rarity in the industry. The vast majority of tech companies are fabless, meaning that they outsource the fabrication of their chips — even if they design them themselves — to other companies. AMD, Nvidia, Qualcomm, Apple, MediaTek, and Broadcom are some of the leading examples of fabless companies today.

Intel is one of the few computer chip firms that still designs and manufactures its own hardware — it can design every part of the process to its own specifications and purposes, and it has an unmatched control over the actual manufacturing of its products. It actually owns its own fabs, and it’s the only major company still making chips in the US right now. Manufacturing has been one of Intel’s biggest strengths for decades. But the fact that the company missed its projections for 7nm, coming after years of 10nm delays that it was only just recovering from, was a huge blow.

Last year started well for Intel. The company announced its Tiger Lake chips (technically the company’s third-generation 10nm designs) which also saw the debut of the company’s long-gestating Xe GPU architecture, giving it an exciting new wave of processors with prowess in an area (graphics) that had long been one of Intel’s weaknesses. The chips offered great performance, even though — much like their own long-delayed Ice Lake predecessors — they were limited to only lower-powered laptops. The wait for 10nm and Intel’s new graphics turned out to have been worth it: the new chips outpaced AMD’s in single-core performance when they were released, and they actually lived up to the hype.

There was still more to do: Intel had only released 10nm chips for its weakest laptop products; if you were buying a thin-and-light computer, it’d be better than ever, with all the performance and power efficiency improvements that the new architecture node brought with it. But it took the company almost a full year before it released more powerful H-series Tiger Lake chips for beefier laptops designed for gaming and video editing, which are only just starting to see those benefits. And it still has yet to move its desktop chips to 10nm (to the potential detriment of this year’s chips, which tried to fuse 10nm designs on 14nm architectures to mixed results).

After years of manufacturing delays, Intel almost seemed to maybe, possibly, be heading back on track.

But when June rolled around, Intel’s troubles started back up again. Seasoned engineer Jim Keller announced that he was leaving the company for “personal reasons.” Keller is the Mr. Fixit of the chip design world: he helped Apple with its in-house A-series chips, AMD with its Zen architecture, and Tesla with its Self-Driving Computer chip. Intel brought Keller on board in 2018, and while the company didn’t publicly define his role beyond leading Intel’s silicon engineering team for “system-on-chip (SoC) development and integration,” his job was clear: as the leader of Intel’s 10,000-person semiconductor engineering team, he was to get Intel back on track.

Keller is known for short stints at companies — he often comes in, sets things in motion, and leaves before the fireworks start. But his less than two years at Intel and his abrupt departure marked one of his shortest tenures yet, and an inauspicious start to Intel’s summer.

But there was more bad news waiting for Intel just a few days later, when Apple announced at WWDC 2020 that it would be making the switch to its own Arm-based chips, away from the Intel processors it had been using for over a decade. The reasoning was similar to when Apple first switched from PowerPC to Intel in 2005: Bloomberg reported that the Mac maker weighed and measured Intel’s roadmaps, already riddled with recent delays, and found it wanting.

Apple’s Q2 2019 earnings report goes even further, straight up blaming a dip in Mac sales that quarter on processor supply issues with Intel. “We believe that our Mac revenue would have been up compared to last year without those constraints,” CEO Tim Cook noted.

At least one former Intel engineer corroborates those frustrations: outspoken principal engineer François Piednoël told PC Gamer that the switch was at least partially due to poor quality assurance with the company’s Skylake chips.

But the switch to Arm chips wasn’t just an announcement that Apple was fed up with Intel’s delays — it was a declaration to the world from one of tech’s most influential companies that Intel’s chips just weren’t up to snuff anymore. “When we make bold changes, it’s for one simple yet powerful reason: so we can make much better products,” Cook said at the announcement — to Apple, leaving Intel behind was a necessary choice to let the Mac leap forward. And sure enough, Apple’s first M1 chips exceeded expectations on every front, upending our concept of laptop performance when they arrived later last year.

But all that was still a prelude to Intel’s biggest looming disaster: an announcement on its Q2 2020 earnings call that the company had identified a “defect mode” in its 7nm process, derailing progress by at least a year. The next-generation chip, originally set for a late 2021 release and a crucial part of Intel’s strategy to get its production back on track and to catch up to other silicon producers like TSMC, were now delayed until early 2023.

And that’s a big problem for Intel. As Natalie Enright Jerger, a professor in computer engineering and leading researcher in computer architectures explains, there are two main ways that chipmakers can improve performance: new designs, like putting more cores or new AI accelerators onto chips; and adding more transistors, which primarily come from new and smaller architectures. “Without being able to shrink the node they’re losing out on one of the two key drivers of performance. They’re still able to innovate … but they’re not able to get that additional win of more transistors that AMD has.”

AMD is also a beneficiary of TSMC’s more advanced production: its Ryzen chips have been on 7nm since 2019, and it could be poised to leapfrog Intel again when its 5nm Zen 4 chips ship in 2022.

Intel is effectively competing with one hand tied behind its back. The fact that it’s keeping up at all is a testament to the advantages its in-house production offers, and the innovations in things like its unique Foveros packaging technologies — all of which is the aforementioned “secret sauce”— but the continued delays to new architectures show it’s the kind of trick that Intel can only keep up for so long.

The fallout from the summer’s events came quickly. Shortly after the 7nm news hit, Venkata “Murthy” Renduchintala, the company’s former hardware chief, left his position. And Intel’s former CEO Bob Swan — previously Intel’s chief financial officer before taking the top spot — was replaced by Pat Gelsinger, a former Intel engineer best known for helping pioneer some of Intel’s earliest innovations in the 1980s and serving as the company’s first chief technology officer. Gelsinger was a hire meant to help right the ship, a sign to both employees and the world that Intel wanted to get serious about chips and play to its strengths again.

Gelsinger has wasted no time, already announcing a new “IDM 2.0” strategy for Intel that will see the company outsource some of its production to other players like TSMC and Samsung, work to establish Intel as a chipmaker for other companies through a new Intel Foundry Service business, and create an ambitious goal to return Intel to the forefront of cutting-edge chip designs by 2025.

The newly announced roadmap — the most detailed look at its future products that Intel has shared in some time — and the renaming of its technology nodes to better contextualize them with competitors is part of that strategy; a major declaration from the company that it has a plan to make its way back.

In a comment to The Verge, Intel executive vice president Michelle Johnston Holthaus said that “We have a solid foundation in place, and an exciting, strategic vision ahead of us,” highlighting the company’s new investments and roadmap.

But it’ll be a challenge to get there. Intel — the company once known for being the untouchable leader of CPU making — is undeniably behind. While some of its new announcements, like the $20 billion it is investing into new factories in Arizona, sound impressive, they’re bets that will take years to pay off. They also pale in comparison to competitors like Samsung, which announced a decade-long $116 billion investment in semiconductor production in 2019, or TSMC, which has announced plans to invest over $100 billion in expanding capacity over the next three years.

Even Intel’s new roadmap is not guaranteed for the future. There are a lot of ambitious goals there, including launching its long-delayed 7nm Intel 4 chips in 2023, and next-generation transistor architecture with Intel 20A in 2024. But Intel is coming off its recent 10nm and 7nm struggles and will need to make sure that its aggressive annual pace it’s setting here doesn’t hit similar snags that could cascade through the PC industry. And competitors like Samsung and TSMC are already gearing up their next generation of chips with higher transistor density and more advanced technologies, meaning that newly reinvigorated competitors like AMD and Qualcomm will be waiting to pick up any of Intel’s slack. As O’Donnell explains, “‘They’ve got to leapfrog now — simply catching up is insufficient.”

Last summer was a wake-up call for Intel, a clear sign that the years of delays and missed opportunities were a buck that it could no longer continue to pass. The changes that Intel’s made in the year since are the right ones: bringing back experienced leaders, hardware designers, and focusing on getting its semiconductor business back on track. Intel will have to regain trust and rebuild its reputation as a leader that OEMs and customers can rely on and show that it can still compete on the bleeding edge of chips. The last year was a good first step, and the ball’s in Intel’s court. But now, it needs to deliver.