AMD vs. Intel: Will the Battle for Chip Supremacy Push the Rivals Together?

Email can be a killer: Just ask New York Governor Andrew Cuomo, who discovered some scandalous messages during a multinational investigation of semiconductor giant Intel.

The Intel emails, made public by Cuomo in 2009 when he was New York state attorney general, appeared to show that the company was taking a carrot-and-stick approach to ensuring that its customers used Intel processors for all, or nearly all, of the PCs that they sold. This might not rank as one of history’s most revelatory email brouhahas, but it was a vital factor in a billion-dollar settlement that would drastically change the semiconductor industry—and the wider tech world that relies on it. The effect is still felt in courtrooms and boardrooms worldwide and is, without a doubt, in the back of the minds of salespeople the world over. And it's still a driving factor behind one of today’s fiercest tech rivalries: Advanced Micro Devices (AMD) versus Intel. 

Previously, the successes and failures of AMD and Intel were of interest mainly to a niche and unlikely coalition of PC-building enthusiasts (who like to overclock tricked-out gaming rigs), IT executives (who buy workstations and servers by the truckload), lawyers (who relish litigious rivalries), and investors (who race to check their Bloomberg terminals at any hint of a scandal).

But the legal battles and quest for technical superiority of the two tech behemoths have taken on new relevance today, when PCs are enjoying a resurgence of interest from home-bound consumers, and Apple, Qualcomm, and other new entrants to the semiconductor industry are poised to shake up the dynamic.

The rivalry between Intel and AMD (detailed in our timeline below) is also entering a new era at a time when chip shortages are roiling large swathes of the global economy. Neither company is likely to claim absolute victory, but the battles won and lost along the way tell us a lot about the future of silicon. 

Multithreading the Needle 

In any telling of how AMD and Intel got into such an epic struggle in the first place, the spring of 2001 is a great place to start. That April, Intel’s engineers were immersed in one of the company’s most ambitious silicon-development projects and were likely facing intense deadline pressure—in a little more than a year’s time, Intel planned to bring multithreading to its consumer processors, a move that would help transform the capabilities of PCs and the software that runs on them.

Multithreading, which Intel calls “Hyper-Threading,” stands apart from many silicon engineering concepts in that it’s relatively simple for laypeople to understand. At its most basic level, a computer transfers demands from a piece of software to the central processing unit (CPU). These transfers are known as threads. If the software and the CPU can each handle multiple threads at once, the system’s efficiency improves greatly, and processing times can be slashed.

The concept of multithreading has been around, theoretically, since the 1950s. (The technology came close to a commercial launch in the 1990s with the never-released DEC Alpha EV8 processor.) For most of the history of consumer PCs, though, CPUs handled only a single thread at a time. You hit Enter or clicked a mouse button, the software sent an instruction to the processor, and the processor sent back a calculation. It was a straightforward—if limited—approach to computing, but it carried us through the dot-com era and helped usher in the internet-connected world of PCs we enjoy today. 

Intel certainly wasn’t the only semiconductor maker racing to implement multithreading technology on consumer PCs, but it had a lot to lose if another company went to market with the technology first. One competitor in particular was likely furrowing the brows of Intel’s C-suite that April: AMD. Although Intel enjoyed a commanding share of the PC processor market in 2001, it was giving up ground to AMD, which was Intel’s only serious competition. By the end of 2001, Intel owned 78.7% of the PC processor market, down from 82.2% in 2000. It had ceded 3.5 points of share to AMD, which grew from 16.7% in 2000 to 20.2% in 2001. 

So, in addition to the pressures that Intel engineers likely felt that spring to deliver the Hyper-Threading project on time, they were undoubtedly aware that their research could have an impact on Intel’s position in the semiconductor market for years to come. 

Trust and Antitrust

As the multithreading project raced toward completion, a legal storm was brewing that had just as much potential to impact Intel’s dominant position in the silicon industry. On April 7, the European Commission confirmed that it had opened an antitrust investigation of Intel. Europe’s competition regulator had received complaints from AMD a year earlier suggesting that Intel was unfairly abusing its dominance on two levels. The first was a minor technical issue related to Intel’s system bus (which connects the CPU to the motherboard) and would soon be dismissed. But the second was potentially more serious: It posited that Intel’s agreements with PC makers and retailers—in which the company would pay rebates based on the percentage of PCs shipped with Intel processors—violated antitrust rules.

Publicly, Intel shrugged off the news. A company spokesperson claimed that the practice was “fair and lawful.” The European Commission acknowledged that the rebate issue wasn’t itself a problem but was rather tied to Intel’s success in the market. ''Loyalty schemes are fine when the company using them is not dominant,'' a Commission spokesperson said. 

But the complaints must have reverberated throughout the company at a time when another tech giant—Microsoft—was nearing the end of its own epic antitrust lawsuit. (The US case against Microsoft would be settled that autumn.) Partly as a result of Microsoft’s troubles, Intel proclaimed the intricacies of its antitrust compliance program to anyone who would listen. Founder and former CEO Andy Grove, who had overseen the transformation of Intel into the world’s largest semiconductor company, spoke candidly about antitrust challenges: “We pay marketing, business development, and sales people to be aggressive,” he told the Harvard Business Review that spring. “How do you impose this new guard band [margin of safety] behavior on a group of people for whom antitrust is antithetical?”

Intel’s answer was to simulate antitrust investigations at regular intervals, according to Grove. The company’s legal department would seize the records of a random management employee and even stage a mock deposition, as though the manager were under a real investigation. Surely this would be enough to discourage any kind of illegal anticompetitive behavior.

Intel Pulls Ahead

For a time, it seemed so. A few months later, the events of September 11 diverted the world’s attention from the troubles of Silicon Valley. The European Commission dropped part of its investigation (a reprieve that would prove to be short-lived). And Intel’s silicon was winning plenty of praise. The launch of Hyper-Threading on Xeon processors in early 2002 was a success. Using a Dell PowerEdge 2650 server, one of the first computers to employ the new technology, PC Magazine in July 2002 recorded performance gains approaching 15% in the number of requests per second the server could handle, as well as total throughput.

“That’s not bad, considering that the technology is free: Intel did not raise Xeon prices when it added Hyper-Threading,” Contributing Editor John R. Delaney wrote at the time.

The Hyper-Threading rollout proceeded slowly but surely, appearing on the first consumer-focused Pentium 4 PCs the following year. In October 2003, PC Magazine’s tests of video-editing software on a 2.4GHz Pentium 4 system with 2GB of memory running Windows XP found that Hyper-Threading always yielded faster scores than PCs equipped with non-Hyper-Threading processors. For years, consumers had comparison-shopped PCs based on their clock speeds, but now the number of cores and threads was shaping up to be just as important. 

Meanwhile, AMD’s market share remained pegged toward the bottom of the scale. To some industry watchers (and perhaps to Intel’s delight), the company seemed mired in a technological funk. In 2004, the second generation of AMD's groundbreaking 64-bit Opteron processors for servers might have presented a serious challenge for Intel, which was relying on a separate Itanium architecture for its 64-bit server chips. Both companies were hoping for what we would refer to today as a “killer app”—the planned 64-bit version of Windows XP. But adoption of both Opteron and Itanium remained lukewarm at best, and they were eventually killed off.

AMD focused part of its attention on other types of processors that didn’t compete directly with Intel. In 2006, the company made a momentous acquisition: the graphics-processing company ATI, a major player in the market for consumer GPUs. AMD used ATI’s expertise to launch a new product line called Fusion. Combining CPU and GPU functions onto a single die, Fusion could be used not only in PCs but also in other types of consumer electronics, such as game consoles. Fusion chips eventually found their way into the Sony PlayStation and Microsoft Xbox consoles. 

AMD also touted its consumer-focused Athlon XP processors as superior to Intel’s Pentium counterparts. In advertisements, the company tried to use benchmark results to turn Intel’s own claims (that the best processor is the one that completes the task in the least amount of time) against it. A full-page magazine ad proclaimed, “Intel talks about performance. AMD Athlon XP delivers it.” A few benchmark performance graphs below claimed to show that the Athlon XP 1800+ outperformed a 1.8GHz Pentium 4 processor by up to 21%. 

AMD executives went on the offensive, too. "We started this challenge because Intel has been delivering inferior technology and scrambling to keep up, but because of their marketing muscle it's not always clear they are struggling," AMD’s then–chief marketing officer Henri Richard told CNN in September 2005. "The best way for us to expose their predicament is to challenge them in the court of public opinion to show that their technology just doesn't (match up) with ours."

But AMD’s salvos seemed to have little effect. The company’s market share gained a bit but remained dismal, and Intel continued to occupy a commanding lead in the PC market. AMD’s share of the market for desktop processors peaked at 26% in 2006, according to data from Mercury Research. Its notebook processor market share peak in 2006 was even lower, at 16%. Both companies announced their first dual-core consumer PCs in September 2004, promising another revolutionary performance leap to rival that of multithreading. But it was Intel’s first dual-core entrant, the Pentium Extreme Edition 840 (which also supported Hyper-Threading), that captivated the media and the public. This magazine tested the first Pentium Extreme Edition in May 2005 and found that it handily outclassed single-core chips from both companies on nearly  every computing task except gaming. 

Intel’s market share skyrocketed again. Measured by how many PC owners uploaded their performance benchmark results to the PassMark database, the number of Intel CPUs in use leapt nearly 8% in the first quarter of 2006 alone. AMD chips saw a commensurate decline. AMD was doing better in the server market, with market share roughly equal to Intel’s, but both companies spent (and still spend) a significant portion of their marketing budget on PC enthusiasts, who are most likely to upload their benchmark scores. So the benchmark measurement was a powerful albeit symbolic feather in Intel’s cap. 

The Intel Investigation Goes Worldwide

As Intel battled with AMD for technological superiority and market-share wins in the early 2000s, its legal troubles did not quietly disappear, as Andy Grove might have hoped. The European Commission dismissed the technical complaints from 2001, but it continued to investigate the rebate payments. And AMD launched new complaints with competition authorities around the globe—in Japan, South Korea, and the United States. 

The Japanese investigation began in 2004 and centered on issues that were much the same as the ones that AMD had originally brought before the European competition watchdog. After 11 months, the Japan Fair Trade Commission (JFTC) ruled that Intel’s Japanese subsidiary had offered rebates or other marketing incentives to the country’s five largest PC makers, with the condition that they use Intel processors for most models sold. Intel offered its OEM customers three rebate tiers, according to the findings: (1) Use Intel chips in every PC; (2) use Intel chips in 90% of PCs sold; or (3) reserve one series of PCs for non-Intel chips and use Intel chips in all the others. 

The result, according to the JFTC, was that the entire remainder of the market (those PCs with non-Intel chips sold in Japan, which were mostly supplied by AMD and the now-defunct Transmeta) cratered, dropping from 24% in 2002 to 11% in 2003. Japanese authorities offered Intel a proposal that seems laughably lenient by the standards of judgments yet to come: Accept the findings and end the rebate program in Japan, and no charges or fines would be levied. The company would not even have to admit to any wrongdoing. Intel took the deal. 

For AMD, the Japan case was “a step in the right direction,” but the complaints were far from over. In 2005, AMD sued Intel in the United States and Korea, alleging the same anti- competitive rebate practices. Armed with the Japanese decision and Intel’s acceptance of it, AMD claimed in its US District Court filing that Intel’s practices were also illegal under American antitrust law. This private lawsuit was consolidated with several other class-action lawsuits. Naturally, Intel objected to AMD’s complaints—it had not admitted to any wrongdoing in any international cases thus far—and asked the court to dismiss the case. In 2007, Intel’s motion to dismiss was denied, and the case was scheduled for trial in 2009. 

Some of the most epic and expensive maneuvering in modern legal history ensued, ensuring that the battle between the two silicon giants would be remembered as much for its legal innovations as its technical ones. By one unofficial measure, in early 2009, Intel had already spent at least $116 million on legal representation for the AMD lawsuit in the US. And the European investigation was still ongoing. In two separate sets of charges in 2007 and 2008, the European Union competition authority corroborated Japan’s findings and added a few of its own. Intel had apparently abused its dominant position in the market for PC processors, not only through rebates but also by paying computer makers to delay or cancel product lines and selling server processors below their actual cost of production.

In the course of arguing its position in Europe, AMD won a significant ancillary US Supreme Court victory, in which the court ruled in 2004 that litigants in foreign legal proceedings can request evidence in a US federal court. This so-called “Section 1782 discovery” has been used several times since, most notably in recent developments in the decades-long, Kafka-esque pollution dispute between Chevron and the indigenous inhabitants of the Amazon rainforest. 

But for non-lawyers, the most memorable part of the legal battles between AMD and Intel, by far, are the aforementioned emails that Andrew Cuomo presented to the court of public opinion, as the US and European cases came to a preliminary conclusion in 2009. Intel agreed to pay $1.25 billion to settle all outstanding disputes with AMD in the US. Meanwhile, European officials levied a €1.06 billion ($1.2 billion) fine, with a trove of damning emails to lend credibility to the eyebrow-raising amount. Intel is still appealing that fine today, but the messages have been made public—thanks in part to Cuomo, who included many of them in yet another antitrust suit filed in late 2009. 

If you’re a fan of legal wrangling, it’s worth scanning the details of Cuomo’s complaint with a giant tub of popcorn handy. You’ll discover that executives at HP and Dell were allegedly beside themselves with worry, because they believed AMD’s Opteron chips were technologically superior to Intel’s, but they thought that without Intel’s financial support, they couldn’t make money. You’ll read about how Intel was allegedly “prepared for jihad if Dell joins the AMD exodus.” (The original European complaint replaces “jihad” with “all-out war” because commissioners believed that the original wording was too offensive. Cuomo, perhaps courting controversy, went with the original.)

Intel disputed the relevancy of the emails, claiming in a statement to The New York Times that the European investigation was heavy on “speculation found in emails from lower-level employees that did not participate in the negotiation of the relevant agreements,” and that the messages were cited “if they favored the commission’s case.” A federal court said Cuomo could not seek damages for many of the sensational allegations in the suit, and it was settled by his successor, Eric Schneiderman, in 2012 for just $6.5 million. 

AMD Finds Its Zen

Damning details, billion-dollar fines, technical one-upmanship, and scandalous emails aside, the AMD-Intel battle is far from over. But two surprising things happened in 2017 that will likely define the next phase of the battle. One favored Intel: The European Court of Justice agreed with Intel’s appeal of the €1.06 billion fine, remanding it back to the competition authority in a retrial that is still ongoing. The second surprise is technical: After years of trying to compete with Intel using Opteron and Athlon processors, AMD launched an entirely new processor architecture called “Zen,” to rave reviews and reassuring market-share gains. 

To date, there have been five generations of Zen-based processors sold under two main brands—Ryzen (for consumer PCs and workstations) and Epyc (for servers). Their performance has bested Intel competitors on many benchmark tests and real-world applications, thanks both to the efficiency of their multicore, multithreaded designs and the fact that AMD has sold many of them for less than competing Intel Core and Xeon products go for.

To understand how completely AMD has turned the tables with Zen, just glance at reviews of processors such as the Ryzen 7 5800X, a favorite of PC-building enthusiasts searching for the ultimate blend of price and performance for their next custom, tricked-out gaming rig or content-creation workstation. Based on our own performance testing late last year, PCMag didn’t see any areas where this chip could be considered objectively worse than Intel’s comparative offerings. And as we noted in the review, “Once you factor in the price-to-performance ratio, Intel finishes further behind in the race than ever.” The Zen architecture is now on its third revision—5800X and other 5000 Series desktop CPUs are Zen 3-based—and Zen 4 is in the works. 

But AMD hasn’t succeeded in only the high-performance desktop PC market. Even its laptop processors outclass their Intel competitors in some measures. One of the best inexpensive laptops of 2020 won the approval of our editors based to a great extent on the fact that its AMD Ryzen 5 processor offers better computing performance than competing laptops with Intel processors in some laptops that cost twice as much.

The Zen architecture took years to develop, and during that time, AMD benefited from major strategic shifts. The first is new executive leadership, including the 2014 appointment of Dr. Lisa Su as AMD’s president and CEO. Widely admired inside and outside the company, Dr. Su’s background is in semiconductor research and development. Before joining AMD, she had spent four years at Freescale Semiconductor, where she was chief technology officer, following that company’s spinoff from Motorola. (Freescale was perhaps best known for making the PowerPC processors for Mac laptops and desktops until Apple’s transition to Intel processors in 2006.) Before Freescale, Dr. Su spent 13 years at IBM, including a stint as technical assistant to then-CEO Louis Gerstner. 

Dr. Su’s move to AMD in 2012 came amidst a multi-year management overhaul. In just two years, she went from senior vice president to chief operating officer to CEO. One of her first significant strategic decisions was to double down on the company’s business for PC processors, instead of abandoning it to focus on refining Fusion for consoles and other electronics. To do so, Dr. Su had to free AMD from the setbacks it had faced in developing its Athlon chips, including finding more research and development capacity. She negotiated a breakup in 2013 with AMD’s former supplier, GlobalFoundries, which cost a cool $320 million in termination fees. 

The about-face was risky. “You would think that the decisions get easier because you have more resources, but they actually don’t, because you have a much broader set of opportunities,” Dr. Su told The Wall Street Journal. Broadening AMD’s manufacturing horizons ended up paying off big time for the development of the new Zen architecture. AMD can now rely upon an ecosystem of partners to produce current chips and refine the technology for future generations. 

One of the company’s most crucial partners is Taiwan Semiconductor Manufacturing Company (TSMC), whose fabrication facilities (fabs) assemble the Zen chips. Unlike Intel, which owns its own fabs, AMD outsources almost all chip production. Still, once everything came together, the Zen launch couldn’t have been better timed to coincide with the climax of negative attention that Intel had garnered with its billions of dollars in proposed fines for alleged anti-competitive concerns. 

Recommended by Our Editors

The Best Budget CPUs for 2024

AMD CEO: Chip Supplies to Remain 'Quite Tight' for the Rest of 2021

Swapping Nanometers for Angstroms: 4 Ways Intel's 2025 Roadmap Promises Smarter PCs

Add in AMD’s remarkable technical advances, so different from its flagging Athlon and FX products of a decade earlier, and you get a remarkable result: The company’s market share is coming back from the brink at an impressive clip. Turning to the PassMark benchmark database once more, it’s clear that the share of PCs with Intel CPUs peaked in late 2016 at 83% and then began a fitful decline. Following a precipitous rise in the third quarter of 2019 (which mirrors 2006’s precipitous drop), AMD’s share of the installed CPU base now stands at 43% to Intel’s 57%. That figure includes all CPU categories—laptops, consumer desktops, and servers. When you limit the PassMark data to just desktop chips, AMD’s reversal of fortunes is even more significant: AMD slightly surpassed Intel in this category in the first quarter of 2021, and the two companies are now essentially tied for desktop CPUs. 

Measured by sales of new products, the market-share trends are slightly less dramatic but still represent a win for AMD. The company’s share of notebook processors reached 19% last year, up from just 5% in 2015, according to Mercury Research. And during the same five-year period, Intel’s notebook-processor market share declined from 95% to 81%. 

A Worldwide Pandemic Changes the Game

At this point, anyone who declares AMD ascendant versus a flagging Intel is credible—the scrappy Texas underdog has plenty of reasons to celebrate. But such a declaration would be premature, not only because the jury is still very much out on Intel’s alleged anti-competitive practices in Europe. In fact, the processor situation today is just as complicated as it was when AMD claimed technical superiority over the far-better-selling Intel chips in its advertisements more than 15 years ago. This complexity has many contributing factors, but two of the largest are the result of exogenous shocks that neither AMD nor Intel could have reasonably predicted, including the effects of COVID-19. 

Against many odds, the humble multithreaded desktop processor that originally pitted AMD against Intel is seeing a resurgence of interest. Industry experts and consumers alike once believed desktop PCs were a dying breed, taking a backseat to the allure of phones, tablets, and watches. In 2015, when AMD and Intel were still in the heat of battle, declining PC sales had vendors wringing their hands and trying to find ways to make the PC more vital to a person's computing experience. 

But the COVID-19 pandemic changed all that. PC sales skyrocketed early in the pandemic, as many white-collar workers and parents of school-age kids worldwide realized that they might be working and learning from home for the foreseeable future. Last year, PC vendors shipped 302 million units across the globe, an annual increase of 13%, according to the research firm IDC. “To put things into perspective, the last time the PC market saw annual growth of this magnitude was 2010, when the market grew 13%,” the company said in a research note. According to data collected by Canalys, global shipments of desktops, notebooks, and workstations in the first quarter of 2021 increased by an even greater 55% year over year. The raw numbers—82.7 million units—also make it the highest Q1 shipment order since 2012, though some of that is tempered by the fact that much of the world was under lockdown in the first quarter of 2020, which hampered shipments of everything. 

Rising demand and COVID-19-induced shortages in materials and production capacity have conspired to create a perfect storm for AMD. TSMC simply has no fab capacity to spare, and even if it did, AMD might not necessarily get to use it right away; TSMC fabs also make chips for many other companies, including Apple and Qualcomm. These clients typically book chip orders months or even years in advance, and TSMC CEO C.C. Wei said during his company’s last earnings call that it is witnessing “strong demand” for its manufacturing capabilities, which will keep chip supplies tight for at least the next 18 months. 

The trickledown result is that it can be difficult to find Ryzen- or Epyc-based computers. While the Passmark data paints a rosy picture for AMD in terms of the CPU preferences of PC performance enthusiasts, many consumers care more about whether they can buy a basic new laptop for homework or Zoom calls. In this respect, AMD’s picture is not so rosy. According to Mercury Research, the company’s chips were in just 19.5% of PCs sold in the second quarter of 2020 during the height of the pandemic. That’s up significantly from 7.3% of new PCs sold in 2015, but it still represents a small share of the market for new PCs. 

Intel has faced production shortages as well, but the output of its dedicated fabs has mitigated the situation. Now, some industry watchers are predicting that the differing manufacturing capabilities of the two combatants will open new battlefronts in the future. As Intel doubles down on its manufacturing strategy, including committing $20 billion for two new fabs in Arizona, it could actually be playing right into AMD’s hands. Loath to cede manufacturing prowess to Intel, TSMC could award AMD more manufacturing capacity, “cementing the company's x86 process technology leadership for the foreseeable future,” according to one analyst. 

In other words, it’s too soon to declare that because everyone wants PCs and AMD can’t produce chips fast enough, Intel will benefit. Indeed, AMD announced a new “3D” chip-stacking technology at Computex in June that takes advantage of TSMC’s latest manufacturing capabilities. By adding a second layer of cache memory on top of the standard processor, AMD can effectively triple the amount of cache the CPU can access, potentially reducing latency and speeding up many applications. Intel is working on chip stacking too, but TSMC says it is theoretically capable of producing up to 12 chip layers, so AMD has a lot of headroom for future technical innovations. 

The second unforeseen aspect that adds to the complexity of the AMD-Intel rivalry is that Intel corporate leadership, once a stable succession of engineers-turned-executives, has experienced unprecedented turnover in recent years. Following the 2013 retirement of CEO Paul Otellini (whose emails litter Cuomo’s trove), the company has gone through three other top bosses in eight years. Bob Swan, who occupied the job for little more than a year before he stepped down in February, inherited massive production problems and many suspicious employees, not to mention the competitive threat from AMD. Swan attempted to improve internal communication and camaraderie, unveiling a “One Intel” slogan and extolling employees to honor truth and transparency in their dealings with customers. He made it his mission to transform the company culture, pivoting it away from Andy Grove’s “only the paranoid survive” approach. 

Current CEO Pat Gelsinger has so far echoed Grove in taking a more combative approach. Like Dr. Lisa Su, Gelsinger is an engineer with deep roots in the semiconductor industry. He spent decades working at Grove’s Intel, helping to design groundbreaking processors in the 1980s, before decamping for EMC in 2009. But unlike Dr. Su and her shakeup at AMD, he has given no indication that Intel will undertake a major shift away from its traditional research, development, and manufacturing methods. Instead, Gelsinger is doubling down on in-house fabs as a means of catching up with AMD and TSMC. Gelsinger also reportedly told employees in January that Intel must “deliver better products to the PC ecosystem than any possible thing [made by] a lifestyle company in Cupertino,” a reference to Apple’s 2020 decision to begin ditching Intel processors in its Mac desktops and laptops in favor of proprietary silicon like the M1 chip. 

Could This Fierce Rivalry Fizzle?

Indeed, it is Gelisinger’s swipe at Apple that represents perhaps the biggest change to the terms of the Intel-AMD rivalry over the next decade or so. As impressive as Apple’s M1 chip is, it’s trivial in a bigger-market-picture sense at the moment—Macs are a drop in the bucket of overall PC sales. Instead, it’s what the M1 represents that could have a lasting impact on the way Intel and AMD do business. Longtime Intel and AMD clients may no longer rely on the two semiconductor veterans for all of their chip needs. 

The Apple shift away from Intel is just beginning, and so are other efforts, such as Qualcomm’s Snapdragon chips for PCs. But on the server side of the industry, the transformation has been underway for years. Amazon, Facebook, and Google are spending small fortunes to design their own server chips that can power everything from simple data cold storage to machine-learning algorithms. Their performance capabilities aren’t anything to write home about—Amazon Web Services’ first custom 16-core processor, launched in 2018, is reportedly less capable than even five cores of Intel’s Xeon E5-2697v4 CPU, from 2013.

But if AMD and Intel are tempted to gloat, it would be to their own detriment. This is no longer purely a game of Xeon versus Ewpyc and Operton in the server market. Even if the AWS chip can’t match a Xeon’s power, it requires no sales agreement and no middlemen. And plenty of servers don’t require powerful silicon at all, instead running only specialized artificial intelligence-boosted applications. Google’s TensorFlow server chips are “an order of magnitude higher performance per watt than commercial FPGAs and GPUs,” CEO Sundar Pichai claimed in 2016, referring to the field-programmable gate arrays and graphics processors that Intel, AMD, and other silicon veterans offer to server customers. 

The fact that Amazon, Apple, Google, and other tech giants can now design their own processors if Intel and AMD don’t offer suitable solutions could profoundly alter the semiconductor industry at a time when its veterans are facing protracted silicon supply problems. If that happens, a rebate scheme that runs afoul of anti-monopoly rules would be a quaint anachronism rather than a billion-dollar blunder. There would be no position to monopolize. Rather than locking horns, AMD and Intel could quickly find themselves circling the wagons to ward off competition from dozens of private-label processors.

Under Gelsinger, Intel has started planning for this possibility. The company’s new fabrication facilities in Arizona, together with existing ones in Oregon, will form a new division of the company dedicated to making chips from other companies. Qualcomm and Amazon have already signed up to be among the first clients of the new Intel Foundry Services. Some of the chips will be based on Intel technology, with Qualcomm planning to use Intel’s 20A production process, announced in July. If the contract chip-making business takes off, Intel could conceivably reinvent itself as a silicon middleman.

That would be a bittersweet eventuality from the perspective of PC enthusiasts and old-school IT execs. To many, the decades-long AMD-Intel rivalry is something of a pastime. To lawyers and OEMs, it was occasionally a very profitable one. But if budding silicon from the likes of Apple and Amazon already rival that of companies with decades of experience in semiconductor engineering, there might not be an opportunity left for either AMD or Intel to don a crown in the decade to come.

[Editors' note: An earlier version of this story used the term "Itanium" instead of "Opteron." The error was corrected on 9/8/21.]