Intel on the Brink of Death

Intel’s board is incompetent and its horrible decisions over the decades are going to push it towards death. The decision to fire Pat Gelsinger, put in charge a CFO + career sales and marketing leader, and cut spending on fabs in favor of a renewed focus on x86 is an example of the incompetence that will end Intel. Fabricated Knowledge wrote The Death of Intel: When Boards Fail recently explaining how board issues around leadership and planning have failed the company. Simply put, the Intel board has escaped blame for over a decade of failures. This decade of failure culminates in the ultimate mistake: dismissing CEO Pat Gelsinger.

Upon closer inspection, these failures are no surprise. 7 of 11 members have no relevant semiconductor experience. Two more are accomplished in the field but as academics, not industry players. They have no experience making hard decisions, understanding critical business inflections, and are not qualified for what’s at stake. The only member with a strong and relevant CV, Stacy J. Smith, joined just this year as a replacement for Lip-Bu Tan.

The past decade is the worst in Intel’s history, with much of the damage done in the 2010s. Pat Gelsinger’s spent the past few years spent correcting the course, but turning the Titanic is not a quick task. There is one constant in Intel’s demise: seven board members. They are still at large on the board today.

Let’s compare the combined resumes of the board that fired Gelsinger against his: which would you choose to rescue Intel? We think there’s still a path forward from here, but it isn’t with the current board.

Short of changing his name to Grove, Noyce, or Moore, it’s hard to see how Gelsinger could be better suited to the job. Regrettably, the board lost its nerve before the dawn of Gelsinger’s turnaround. Semiconductors are one of the most complex industries, and it takes many years to course correct. In his final board meeting, Gelsinger presented an updated capital spending plan. The plan was to maintain substantial spending in Intel Foundry Services, but the board was so unhappy with the plan that they forced him out.

Gelsinger could have been better as a CEO without a doubt. He is somewhat irrationally optimistic, but that’s what Intel needs. He’s had his fair share of missteps, for example, their current AI strategy is still broken with Gaudi 3 and Falcon Shores meaning Intel will never get much of the GenAI inference or training market. The cultural rot that mainly contributed to the 10nm issues has mostly been fixed from a technical perspective, but external customers don’t get serviced appropriately. The Tower Semiconductor acquisition would have brought much-needed customer-facing expertise, but after the deal break due to regulators, there was no plan B. Intel has a reputation for being hard to work with and has no customer-centric view; meanwhile, TSMC is hesitant even to raise prices on its customers – it’s not a surprise who’s winning. Gelsinger’s famous pre-keynote pushups did not exactly bulk the customer-centric muscle up at all.

Still, Gelsinger was a qualified candidate who wanted the job. In that respect, he was probably a one-of-one. Today, the board has mostly worse options.

Intel’s Failures

The problems at Intel began with the 10nm node (arguably 14nm). In 2016, TSMC and Intel planned to introduce their 10nm processes into volume production. While TSMC executed on schedule with a lower performing node, Intel pushed an aggressive shrink requiring quadruple patterning, novel Cobalt interconnects, and contact over the active gate. The yield was bad, and the node took three years to fix. By the time Intel shipped 10nm products in volume, TSMC had sold more than half a million N7 wafers and was sampling N5.

Intel’s products then suffered because of stagnant process tech. Competitors like AMD had the advantage of TSMC’s fabrication and in many cases better chip designs/architectures. Datacenter market share began to slip, and Intel’s business issues only snowballed.

That’s a well-understood part of Intel’s history. However, for some reason, the board’s incompetence and leadership decisions are not usually part of the narrative. The actual fall of one of the greatest companies ever started more than a decade before this, and it started with Intel no longer being Intel culturally.

Culture – Rotten to the Core

The story of Intel’s cultural rot goes back to Paul Otellini. Paul and Pat Gelsinger were the front runners for the CEO position. This is the classic leadership choice of business bro versus technologist. The result was that Intel chose its first non-engineer CEO.

Paul was ultimately chosen due to his ruthless anti-competitive business decisions that locked AMD out of the CPU market and cemented Intel’s role as a monopoly for more than a decade. Paul instituted a policy that involved paying various OEMs and system integrators not to use AMD, which choked out AMD’s revenue, R&D, and fab investments. Dell alone was paid ~$4.3 billion, and this was the only reason Dell was profitable during this period. Intel and the EU are still fighting out this anti-competitive behavior in courts to this day.

From then on, business decisions reigned supreme over technology. Intel formers believe Paul is the reason the organization became a political hellscape. Before Paul, Intel had the Andy Grove culture of “constructive confrontation”, data-driven decision-making, an intense focus on execution, and extreme accountability. There are many good books on it, but get an Intel old-timer reminiscing, and they will tell you people always had numbers to back up their view and often yelled at each other during decision-making processes. Still, the path was chosen, and the company unified and drove forward. Intel, then was an unstoppable freight train.

Intel had a harsh way of getting things done, but it was a productive culture. Paul Otellini changed this. Technical decisions were pushed aside for political power, and the path forward throughout the company was through power struggles among the various fiefdoms. This cultural rot started under Paul Otellini and continued to decay until Pat Gelsinger arrived.

Besides changing the culture, there were also numerous horrible decisions that the board greenlighted such as going on a silly acquisition spree buying completely unrelated firms such as McAfee and not even trying to win the iPhone business even though it was dropped in their lap. The failures getting into mobile and rise of Arm, which we detailed here, is a decision which will forever haunt Intel. Paul’s successor was only worse.

Brian Krzanich was a disaster as CEO. He presided over the 10nm debacle. This mismanagement of the fabs is the single greatest issue the company faced, because that is the core of Intel. Despite this, Krzanich was only fired when an illicit workplace relationship came to light. The board that nominated him included John Donahoe—the CEO who made Nike uncool—along with future chairman Frank Yeary. Books could be written about Brian’s failures as a leader, bad acquisitions, failures in AI, and rotting technical leadership, but the reality was that he was a byproduct of the toxic culture and he accelerated it hugely. He probably goes down as Intel’s worst CEO ever.

Financial Engineering Paves the Path to Hell

Not to be outdone, the 2018 board iteration replaced Krzanich with the first truly non-technical CEO in Intel’s history: Bob Swan. Technically, Paul Otellini was the first non-engineer to lead Intel, but he spent more than 30 years with the company, including his time as a technical advisor to the legendary Andy Grove and leading the microprocessor division.

Swan was a professional CFO – Intel was his 10^th CFO role – and so process engineering took a backseat to financial engineering. Swan’s Intel spent as much on stock buybacks as it did capital expenditures on fabs over his tenure: more than $36 billion towards buybacks versus $38 billion in Capex. This was malpractice in a capital-intensive industry when the company was bleeding market share and more than two nodes behind its chief rival.

Brian Krzanich, Bob Swan, and Intel’s board cut not only Capex, but also technical talent, in droves. From 2013 to 2020, 4 out of 7 years had shrinking headcount all while the business lost its technical leadership and had fantastic profitability. Their idle, non-technical eye led to technical incompetence and politicians such as Navin Shenoy, Murthy Renduchintala, Aicha Evans, Remi El-Ouazzane, and many others to drive horrible decision-making throughout the organization. This is just a shortlist. Pat Gelsinger was the cure, and he cleared out many of the fiefdoms and incompetency of these managers. Our only criticism of culture is that he is too nice, gives them chances, and doesn’t do it faster.

What of the directors that promoted this behavior? Five of those ten members on the board just fired Pat Gelsinger. Post-firing, chairman Frank Yeary stated in a press release that “as a board, we know first and foremost that we must put our product group at the center of all we do.”

This is a backward strategy that will not save Intel. The board is making yet another short-sighted mistake it seems to specialize in. The board is blind, but so are the new co-CEOs, the CFO David Zinsner, and Michelle Johnston Holthaus. Michelle is the former sales marketing and communications lead and chief revenue officer, and Zinsner joined in 2022. Intel needs leadership at this critical time, and this is not the co-CEOs for the job. Read some of Michelle’s work with a disconnection from reality, business, and marketing strategy below:

There is No Moat in x86 and the Product Group

Intel’s glory days were when they had superior process technology combined with their x86 moat. The x86 moat was also twofold: Intel had a moat in x86, and x86 had a moat in computing. Today, neither of those moats hold.

Before the smartphone era, x86 was the dominant instruction set within general-purpose CPUs. Almost every PC and server was guaranteed to have an x86-based CPU as the software was written to be compatible with the x86 instruction set. This was propagated by the “Wintel” (Windows and Intel) alliance, where Windows was the dominant Operating System that ran exclusively on x86. Software developers would rationally focus their efforts on developing software for the largest user base: Windows, and that meant making software for x86. This was a classic ecosystem: customers would want Windows PCs because of the larger suite of software options, and to use Windows meant buying an x86-based CPU.

Most of those x86 CPUs were Intel CPUs. While AMD also had the IP rights to design x86-based CPUs, AMD was for a long time tied to its own fabs (now spun off as Global Foundries) with inferior process technology to Intel’s, making it uncompetitive. Ironically, this is Intel’s position today. Intel’s failures in fabs and TSMC’s relentless march enabled AMD to utilize its superior architecture and process technology to finally fight back.

A CPU based on a different architecture stood little chance without OS and software support, absent a platform shift. This platform shift is precisely what happened with the smartphone era. The smartphone paradigm meant new software ecosystems emerged in the form of iOS and Android, with new sets of applications, meaning x86’s legacy moat in personal computing didn’t carry over to smartphones.

While Intel did make an attempt in smartphone AP’s, it was late, half-hearted, and with the wrong strategy of the PC mindset. They didn’t understand how to innovate beyond their golden goose. Crucially, power efficiency was far more critical for smartphones due to battery constraints (much less of an issue for PCs) and this is where Intel lost out to competing ARM-based smartphone SoCs.

It could be argued that this wasn’t disastrous, as Intel only lost out on a new market opportunity rather than losing its stronghold in PCs, but the rise of smartphones accelerated the maturity of the client CPU market. Smartphones’ convenience, ubiquity, and functionality meant consumers spent more of their time on their smartphones, which met or exceeded many PC capabilities. Of course, this cannibalized the client PC market, with many consumers’ primary method of personal computing being smartphones, and not dedicated PCs.

Competition is Coming Even for x86 Client CPUs

This began the fading relevance of Windows and Intel, replaced by the Apple and Arm era. This partnership has encroached on Intel Product Group’s core: Apple took the knowledge and experience from designing APs with their A-series iPhone SoCs and parlayed that into the hugely successful Arm-based M-series SoCs for their client notebooks and desktops in 2020. Fifteen years after succumbing to the dominance of x86 over IBM PowerPC, Apple ended its Intel partnership.

This transition was only made possible with a tremendous effort to port software written for x86 to Arm. The key piece was the Rosetta 2 emulator, which recompiled apps at install to work with Apple silicon, enabling a seamless transition. The Apple M1 unlocked substantial performance gains with various accelerator engines not offered by Intel along with considerable boosts in battery life. It was a hit. For Apple, this meant higher performance and better margins. While Intel’s new Lunar Lake catches up on accelerators and battery life competition, it’s too late; the ship has sailed. Furthermore, Intel has an uncompetitive cost structure due to the use of TSMC’s leading-edge process technology.

The x86 software lock is being broken. This inspired other chipmakers to attack the mature but lucrative client CPU market using the open (for a price) Arm-based ISA. Even Microsoft of the former Wintel alliance responded by launching Windows for Arm, completing the non-x86 ecosystem.

Qualcomm released Snapdragon X for Windows PC in 2024, and there will be more entrants. Nvidia and MediaTek are both independently working on Arm client PC chips; more details on these chips later. AMD, despite being a beneficiary of the x86 ecosystem, sees the writing on the wall and is also developing an Arm-based CPU for Microsoft as a semi-custom chip.

Yes, Arm for PC still has many kinks to iron out, so Qualcomm’s Snapdragon X hasn’t taken much market share. What’s important is that the dam has broken and a flood will start soon. Arm for PC will happen because there is now a quorum of important players in the ecosystem (Microsoft, Arm, Qualcomm, Nvidia, Mediatek) who want to and are set on making Arm for PC happen.

To summarize, Intel’s loss of moat in Client CPU: what once was a single-player market now has five heavyweight competitors (Intel, AMD, Nvidia, Qualcomm, Apple). x86 has a very tenuous moat in mobile PC.

Datacenter Permanent TAM Loss to Hyperscalers, AI, and Accelerated Computing

While the client PC market slowed due to smartphones, x86 CPU still grew in server CPU and datacenter workloads. Intel relied on its datacenter product segment for growth, especially from hyperscaler and cloud services customers. But the story with server x86 mirrors that of client x86, with all anchor customers turning to Arm. The front line was already broken, with AWS’s Graviton line of Arm-based CPU servers as the pioneer.

A stunning datapoint from AWS Re:Invent captures the success of Graviton:

Over the last two years, more than 50 percent of all the CPU capacity landed in our datacenters was on AWS Graviton. Think about that. That’s more Graviton processors than all the other processor types combined.

Dave Brown, Leading AWS EC2 Compute and Networking

Other hyperscalers took note and are following suit: Google has Axion, Microsoft has Cobalt, Alibaba has Yitian, and Meta is also developing an Arm based CPU. Arm is making it incredibly easy for customers to design custom chips with their CSS offering.

At the same time, Arm is taking CPU market share, and accelerated workloads are taking share from general purpose CPU. Gen AI is necessitating the biggest compute infrastructure buildout ever, centered around GPUs and other AI Accelerators. Even x86’s minor participation in this buildout is dwindling. Nvidia’s GB200, the most in-demand next-generation Blackwell SKU, uses Nvidia’s Grace CPU (Arm-based) to feed the GPUs. This is instead of the x86 Xeon CPUs used for this generation’s most common AI server configuration: the Hopper HGX. Intel’s Gaudi 3, Falcon Shores, and more are not competitive with AMD’s roadmap, let alone Nvidia’s.

And it’s not just AI. Other workloads are being accelerated too. Chips like the AWS Nitro, Google Argos VPU, and Meta MSVP were all designed with the intention to deploy cost-effective silicon in large, specialized workloads that general purpose CPUs are not optimized for. They reduce the TAM of x86 CPUs, just like Nvidia’s Accelerated computing strategy does.

Like client PCs, x86 in the datacenter is losing share from multiple sides: Arm and accelerated computing. Nothing can be done to stop this, and even Intel’s new products won’t solve it. We’ll talk more about how Intel’s datacenter roadmap is completely uncompetitive even in 2026 unless the fabs drive a huge cost advantage.

No, x86 will not disappear overnight. It is still a large market and potentially a cash cow business. But cash cow status only happens if large swaths of employees are fired, choking innovation long term. Even then, AMD and the various Arm players likely grab market share faster than the Intel board is thinking. The board’s ”focus on product” strategy sounds like a dead end.

Intel Product Cannot Be Competitive Without Fabs

However, the problem is that without Intel’s old manufacturing prowess, Intel’s x86 is no longer competitive with AMD, let alone the Arm-based options. Intel can bite the bullet and take the gross margin hit by outsourcing manufacturing to TSMC. This levels the playing field with AMD but doesn’t solve the issue that Intel cannot out-design AMD.

This is why products like Lunar Lake, which are primarily outsourced to TSMC, cannot be ramped. They have a gross margin in the teens. The board doesn’t understand this because they don’t understand semiconductor manufacturing. The client CPU organization still ships the majority of Raptor Lake monolithic dies made by Intel’s fabs for a reason. If they didn’t, Intel would be losing money even faster.

The Intel Product group has been spoiled with exclusive access to a superior process for decades, which covered up any flaws in their microarchitecture. The consequence is that Intel uses 2x as much silicon area for their product today compared to best-in-class peers: AMD, Nvidia, and Qualcomm. That does not sound like a leading design firm, and Intel’s product group should not be the focus. It simply is a legacy of Intel’s technology leadership in logic fabrication and the dominance of the x86 ISA in general purpose CPU. That is no longer relevant today.

Intel Foundry is the most important part of the company, and it must be saved.

Intel Foundry is the Most Important Part Of Intel

Intel Foundry is the future of Intel. It has massive strategic value to the United States and the Western Hemisphere. Leading-edge semiconductors are critical to consumer, industrial, and military applications, yet the West has no ability to produce them at scale.

TSMC is the only manufacturer in high volumes, and the Taiwanese government has been explicit that it will not allow offshore production of the latest node. The Arizona project has a capacity of less than a fifth of that in Taiwan at 5nm and 3nm. Furthermore, TSMC Arizona will soon be two process nodes behind the leading edge. More supply is needed to achieve any level of national security, and this is no guarantee in a crisis.

The only entity even remotely able to fill this gap is Intel Foundry. Intel has brought many more manufacturing technologies to market first, such as high-K metal gates, FinFET, and much more. They lost EUV to TSMC, but their current roadmap has them bringing gate all around, backside power delivery, high NA EUV, and DSA before TSMC.

18A will likely be the best of the rest outside TSMC when (if) it ramps into high volume next year, and 14A has a legitimate chance at beating TSMC’s latest around 2027. To be clear, Intel has had some challenges including a PDK 1.0 delay for 18A and yield issues on pre-1.0 PDKs leaked by Broadcom, but they are coming to market before TSMC with both gate all around transistors and backside power delivery. Unlike Intel’s floundering product group, IFS is a competitively advantaged business. Of course, this is contingent on Intel Foundry surviving that long.

Intel Foundry will need help. Intel will likely be sold off for parts; the board has stated Products are the priority, and Foundry’s greatest champion was just shown the door. The business is a capex black hole: we estimate that, even with highly reduced capacity buildouts, Intel Foundry will need $36.5B just for wafer fab equipment in the next 3 years. Fab shells and other expenses would add another $15-20B+. Intel doesn’t have the cash flow to support this due to the product group’s failings, even with CHIPS Act subsidies.

Yet the cost of losing access to cutting edge chips would be an order of magnitude more. According to the director of the National Economic Council, the 2021 chip shortage cost 1% of U.S. GDP that year – about $240B. That was a months-long disruption, mostly in mature chips. Restoring advanced logic capabilities from zero would take decades. If the 2021 shortage was a 10-foot swell, zeroing out advanced logic supply would be a 100-foot tsunami.

Besides the cost risk, there is a national security imperative. If you believe National Security Advisor Jake Sullivan that “there is probably no other technology that will be more critical to our national security in the years ahead” than advanced AI systems, then you must believe that Intel Foundry, as America’s best option to guarantee advanced logic supply, is critical as well. Imagine not having the means to produce the most crucial technology of the next century. Intel Foundry must be saved.

How to Save Intel Foundry

How then can Intel Foundry survive? The minimum survivable product is as a second-source for TSMC leading edge with significant product volumes from multiple top hyperscalers and fabless design companies. Customers want to de-risk their TSMC/Taiwan exposure the same way the national security community does.

At present, second sourcing with Intel Foundry would mean trading geopolitical risk for performance and cost. The 18A ecosystem is not mature – the PDK is not as good as TSMC’s, the library of silicon-proven IP is small, and EDA support is not strong (EDA companies will get much better returns for their efforts supporting TSMC than Intel Foundry currently).

Overall, the cost of porting a design to Intel 18A would be close if not more than that of a new design – at N2. The market has spoken here: it is too expensive and risky. Otherwise, we would see fabless companies second sourcing their flagship products.

Intel Foundry should be laser-focused on 1) a competitive process technology and 2) making design switch over from TSMC as cheap and easy as possible. The former is on track, but it is not clear what the latter is. A split from the Intel parent would reduce distractions and increase focus. Government support on national security grounds is necessary. Intel Foundry is the single best hedge America has against a Chinese-sponsored coup or invasion of Taiwan.

But note that Intel selling Intel Foundry will not work without a significant capitalization to the tune of ~$50B injected into Intel Foundry. AMD tried to spin off the fabs, and it was disastrous. Mubadala purchased the fabs from AMD and created GlobalFoundries. They then proceeded to lose $22.4 billion over the next decade.

That’s not all, IBM paid $1.5 billion to GlobalFoundries for the privilege of having GlobalFoundries acquire IBM’s fabs. AMD was also burdened with long term wafer supply agreements that prevented them from using other fabs. Even with all this, GlobalFoundries had to stop process development at 14nm and run a licensed Samsung’s process + a special purpose 14nm from IBM. GlobalFoundries skipped 10nm and quit entirely at 7nm. A repeat of this requires someone willing to spend $50B+ keeping Intel Foundry afloat.

No, instead, Intel has to sell the product groups like Client x86, Mobileye, and Altera to to private equity firms and other vultures like Broadcom and Qualcomm bundles alongside long-term agreements for fabrication.

While the Trump administration is probably allergic to anything that looks like “corporate welfare,” many key officials are national security hawks who recognize the importance of having advanced logic manufacturing capability on-shore. A standalone Intel Foundry that is capitalized and with long term manufacturing agreements from 2 of the largest semiconductor companies in the US is much easier for the government to support, both in dollar amounts and politically.

Intel Foundry won’t be laden with Intel’s lagging product team, Mobileye, or Altera. Intel Foundry will have one clear function, and it’s vital to national security and the future of America and the West.

Who should lead the charge here? Maybe a “Chip Czar” charged with restoring American logic prowess. We know someone with a great CV who has just become freed up for new opportunities…

Below we will discuss what happens next to Intel’s product group with David Zinsner and Michelle Johnston Holthaus at the head with the current board, Intel’s client CPU issues, their desperation play with Bartlett Lake. We will also dive into Nvidia’s custom PC chips and their aggressive 3 chip Tegra strategy that likely takes significant share from Intel. We will then discuss Intel’s current datacenter CPU roadmap from Sierra Forest and Granite Rapids all the way out to Clearwater Forest, Diamond Rapids, and Rogue River Forest. Then we will discuss how the Systems Foundry approach is the one to save Intel, alongside the need to sell the PC business.