How Intel Got So Close To Failure

Intel was once the biggest CPU maker in the world, and certainly the biggest desktop CPU maker, but now? Now they’re struggling, and are at a point where without significant changes, they’re heading straight for a repeat of AMD in 2012 through 2018. In other words, Intel is now a sinking ship, and it needs to turn course quickly, but with a boat this large, it often struggles to make such rapid maneuvers. But to best understand how they got here, we need to take at least a brief look back at Intel’s history, and the path that led them here. This is the story of Intel’s (potential) imminent collapse. 

Intel has been around a long time – founded in 1968 by two of the traitorous eight that left Shockley to found Fairchild, who then left to found Intel, those being Gordon E Moore (the Moore’s law guy), Robert Noyce and Arthur Rock. I’ve already covered a lot of Intel’s early history in another video you can check out here (in the cards above), so suffice to say they started small and worked up, going from a couple thousand transistor CPUs, to millions. Let’s pick up after the introduction of their X86 CPUs, like the 8086, 80286 and 80386. This is where Intel first got some competition – but not from where you might think. Back in those days, the companies who were buying Intel’s CPUs – like IBM for the IBM PC – required Intel to offer a second-source manufacturer for their products, so if Intel couldn’t provide enough units, their second-source could. AMD was one of those second-sourcers, making their own version of basically all Intel CPUs, at the request of Intel! By the 286 and 386 days though, Intel was getting tired of essentially splitting the profits of their own designs with people like AMD, so started cutting second-sourcers out. AMD, among others, spent a lot of time reverse engineering Intel’s chips, and designing their own, often slightly better, versions. 

Intel’s first big anti-competitive move was using, copywriting and enforcing legal action on microcode inside their CPUs – what we’d probably call ‘firmware’ today. While even the original 8086 had some microcode, the 486 and later P6 Pentium Pro’s expanded the use of microcode, and because software is copywritable (but hardware isn’t), Intel could sue competitors like AMD and NEC for using Intel’s microcode. The NEC case is what decided microcode was software and therefore copywriteable, despite NEC being able to prove they wholly reverse engineered the chips and did not use Intel’s own microcode, instead using ‘clean room principles’ to recreate them. The result of all of that is that Intel walked away with $40 million from AMD in 1995, although AMD was allowed to keep making reverse engineered versions of Intel’s chips. Ironically enough, AMD’s own foundries were ahead on the process node front, allowing them to offer die-shrunk versions of the 486 – the DX4 – that was faster than even Intel’s much newer early Pentium chips, but Intel still held the lion’s share of the market. Other competitors like Cyrix came and went, with only AMD sticking it out. AMD offered a few sell-out hits with the K6, Duron and truly peaked with the first dual core, the Athlon X2, and later quad core X4. They also brought 64 bit chips to the market, with their AMD64 instruction set extension for x86, which is when, at least according to PassMark, AMD hit their highest ever desktop market share – actually surpassing Intel for a short while. 

Then, of course, things went downhill. Their Phenom line of chips, including the infamous Phenom II X3 which, at least the early units, could be unlocked to run as the full-fat X6 instead, and worse, the FX line really hit AMD hard. Their core design, especially for the FX chips, was pretty limiting, splitting one floating point unit across two integer units and still calling that “two cores” meant they didn’t perform overly well, and combined with their foundries drowning money, they opted to do the unthinkable – sell off their foundries. AMD spun out GlobalFoundries in 2009, although AMD almost signed their death certificate with it by signing an exclusivity contract that locked them in to use GF for a number of years, requiring AMD to order their wafers from GlobalFoundries. That was wholly necessary to get any amount of value for the foundry business, but brought them to their knees. I have a feeling this might end up being prophetic for Intel…

During these years Intel first launched their Core line of chips, which especially with Sandy Bridge in 2011 absolutely took off. Both enthusiast and mainstream SKUs saw massive success, and started Intel’s true hay-day. The next generation, Ivy Bridge, was basically a die shrink and gate design change, but the architecture remained the same. Haswell (and Devil’s Canyon) came next, offering a generally similar core design, although new things like a wider core design, more instructions, and a new fully integrated voltage regulator onboard the CPU offering lower power C6 and C7 sleep states were added in. Skylake – that’s 6th gen – was generally a die shrink to the infamous 14nm process node, with tweaks to power consumption, and if we are being honest, until the 12th gen Alder Lake chips, we basically had the same core design with minor tweaks since Skylake. Hell, until, what, Core Ultra, even, at least on the P cores. That sort of stagnation came from two sources, superiority and market position, and being stuck on the same process node for almost a decade. At this point AMD was in the dumps, with AMD having no more than 20% market share and the disparity was only growing, and in the server space Intel’s Xeon chips were basically the only choice – PassMark reckons AMD had just 2% of the server market share in late 2014 – so Intel felt, and frankly looked, invincible. That invincibility meant they weren’t pushed to improve, hence functionally the same core design for more than a decade, and it wasn’t until AMD came back swinging with their Ryzen chips did Intel respond by finally giving us more than four cores, 6 at first (to AMD’s eight), then ten, then eight when they realised ten 14nm++++ cores were a fire hazard. 

That level of stagnation was only emphasised by the finance bros at the helm squeezing every drop of ‘value’ out of us, their customers. Everything from locking basic system settings (like RAM speed) to high end unlocked CPU SKUs AND motherboards to changing socket every, or every other, CPU generation to force you to buy a new motherboard, and so, so much more, including steadily increasing CPU prices without much to show for it. The worst stuff was all behind the scenes though. In 2009 the EU fined Intel a whopping 1.06 billion euros for anti-competitive practices. Just to emphasise how evil this was, they were effectively paying distributors to stop them buying AMD CPUs – in other words paying for exclusivity. That included system integrators – people like Dell or HP for example – to only use Intel CPUs, and yeah paying major retailers to only buy Intel CPUs or sell Intel systems. AMD literally tried to give one million free CPUs to a large computer manufacturer, but thanks to Intel’s ‘rebates’, accepting those FREE CPUS would have cost the maker millions, so they could only accept 160,000 of those, again, FREE CPUS. They even made a major PC maker delay their first AMD based business desktop by six months so Intel could get out ahead. Yikes. Oh, and that isn’t all. The FTC fined Intel $1.25 billion (just a year after the EU’s billion euro fine) for the same tactics in the USA, but there’s more too. Intel changed compiler settings to favour Intel CPUs, without telling developers, to hurt AMD’s benchmark performance, and paid and coerced software and hardware vendors to either not, or at least limit their support for AMD CPUs. Jesus Intel, really? You were already winning bro, chill… 

Then there’s the process node problem. From 2015 to 2021, that’s Skylake to Rocket Lake, or 6th to 11th generation, Intel was stuck on their 14nm process node. For those counting, that’s Skylake, Kaby Lake, Coffee Lake, Comet Lake, and Rocket Lake, or five generations of CPUs, all on essentially the same process node. Why? Well there’s a whole bunch of reasons. The 14nm process node was delayed at first, and because Intel didn’t have leapfrogging development teams – as in one team working on the current node, and one working on the next, simultaneously – that caused 10nm to be delayed. They shot for a frankly ridiculous transistor density target at first, which meant yields were reportedly sub 1% – as in for every 100 chips made, maybe one of them would actually work – which meant the only 10nm chip they produced at first was a tiny dual core which drew more power than a 22nm chip and the iGPU was broken. That’s how bad it was. Intel iterated the design for Ice Lake, but the process node was still not good enough for desktop so all they could manage was their almost unchallenged laptop SKUs. Tiger Lake brought another iteration with 10nm Super Fin, but again it wasn’t good enough for desktop. Finally, in 2021, they launched 10nm Enhanced Super Fin (later renamed to Intel 7 to align with TSMC’s N7 node, which has similar densities) with Alder Lake, although it’s worth noting that only that, and Raptor Lake (13th and 14th generation) used that node. For context, TSMC’s N7 node was first producing customer chips in 2018 (Apple’s A12 Bionic), with AMD using it in 2019, two years earlier than Intel’s first mainstream 10nm/7nm node. Intel’s next node, now called Intel 4, clearly struggled to hit yields too, as only the mobile Meteor Lake chips used it, as the following desktop and mobile chips, Arrow and Lunar Lake opted to use TSMC’s N3B node for the cores at least. The next node (skipping over the never-used Intel 3 node), 18A is what the last CEO, Pat Gelsinger, bet the company on, and yet it’s currently unclear whether or not Intel’s next generation of Panther Lake chips will even use it, or if they’ll outsource to TSMC instead. 

Intel is, at least currently, spending between $14bn and $18bn a year on R&D alone, and it sure looks like they’re struggling to find any results from it. While they claim 18A is 25% faster, or 36% more efficient than Intel 3, that’s a bit of a poor comparison since Intel has never released a single Intel 3 based chip. Intel has invested heavily in improving their foundries, including buying the entire year’s worth of production of ASML’s High NA EUV machines – video about those in the cards above if you’re interested – which are a cool $400 million each, although it’s very unlikely these will be used in 18A, more likely for the upcoming 14A node instead, although part of that spend is their new IDM business. Until recently, Intel’s fabs were used exclusively by Intel. They did actually open their foundries between 2013 and 2018, but this “IDM 2.0” approach was a bid to attract customers to get their chips made by Intel, rather than TSMC. Unfortunately for Intel, they don’t seem to have landed many, or any customers, and importantly for their flagship nodes, they really require “hero customers” to justify the R&D spend, and if they themselves aren’t even using it, why would anyone else?

Confidence in their foundries was most definitely shaken when the rustgate scandal set the internet ablaze. Throughout 2024 Intel’s 13th and 14th generation of CPUs started experiencing massive failure rates – orders of magnitude higher than usual – especially on the higher end parts. While initially there was a lot of speculation – and Intel only fueled those fires in an effort to not point the finger at themselves – it turns out that there was two very much Intel problems that were the cause. A microcode issue making the CPUs request the motherboard to burn them alive, and oxidization of the vias inside the CPU. The former was eventually fixed, but the latter was only fixed with a replacement CPU, and Intel’s handling of the event was quite the sight to behold. See when AMD’s Ryzen 7000 CPUs started melting in their sockets, regardless of the blame AMD immediately offered brand new CPUs and motherboards, and started investigating the issue, and addressed the route cause quickly. Intel, on the other hand, blamed motherboard vendors and users, only to then admit they knew about the oxidization issue in “late 2022”, but still shipped the damn CPUs anyway! They said they’d extend owner’s warranties, but then denied a whole bunch of RMA requests for dead chips. It wasn’t a good look for the brand, or their manufacturing capabilities.

Ok, just because this is funny I’ll include this little bit of fraudulent misrepresentation they did at Computex in 2018. At that time AMD was starting to cook with their Ryzen, and importantly for this story, Threadripper, line of chips. They’d already launched the first generation Threadripper with up to 16 cores, but the second gen chips were coming, and AMD was promising a 250W 32 core chip, all for $1800. So Intel, clearly in a panic, decided to haul out the highest core count chip they had, a Xeon chip they claimed would be “coming later this year”. They ran benchmarks on stage showing the chip running at 5GHz and wolfing down Cinebench multi-thread. Pretty cool, right? Well that’s the thing, what they didn’t say – in fact what they hid under the desk – was that to run that chip at 5GHz you need something close to 1,000 watts, AND a 1,700 watt water chiller. That’s like doing a “naturally aspirated” drag race, but hiding the nitrous bottle under your boot floor. 

In the present then, Intel looks to be in trouble. What’s somewhat worrying is their new CEO, Lip-Bu Tan, seems to be trying to make aggressive changes to, and I quote, “pull off a comeback that will be studied in business schools for generations to come”, and yet maybe is throwing the baby out with the bath water. He’s sacking ANOTHER 25,000 employees by the end of the year (after Intel has already cut up to 15,000 staff, and a good couple thousand of those in this year alone), and importantly he’s pulling back on the foundry business, stating, “We will build what our customers need, when they need it”, and making it clear that “Going forward, our investment in Intel 14A will be based on confirmed customer commitments. There are no more blank checks.” That seems a bit circular, as customers will not invest in designing for a node that doesn’t exist, and that can’t exist without their commitment. Chicken, meet egg. 

To me it seems Intel has two main challenge areas. The products, and the foundry. The foundry is floundering, with its success hinging on 18A and 18A-P, but the product front is another kettle of fish. Intel is losing market share in the server space, it seems to be the biggest loser in the AI race, and in the consumer space the Core Ultra 200 series is a disaster. According to Mindfactory’s sales data, in week 28 of 2025, they sold just 140 Intel CPUs, compared to 1725 AMD chips – and the average selling price was nearly 60 euros higher for AMD too. Even worse, of those 140 CPUs sold, only 20 were Core Ultra chips, with the vast majority being the older LGA 1700 based chips. Yikes. With AMD’s strong success from their X3D chips offering ridiculous performance benefits, Intel is going to really struggle to keep up at least in the gaming space. AMD has been remarkably flexible, even in the server space, but by contrast Intel has been rigid. That’ll likely have to change, and pretty quickly, for them to regain their footing. 

To be clear, while Intel has done some genuinely heinous stuff over the years, we as consumers need them to survive. We need competition – just look at AMD. Their underdog state is what led them here, and the lack of competition is why Intel is here too. They both need to be making competitive products for us consumers to see the benefit – which is better products at lower prices. Competition keeps them all in check, and that’s something we sorely need. So, I hope that Intel can turn it around, be that by selling off their foundry arm (although I fear that’d be a dangerous act), or actually pulling through on 18A and 14A. Only time will tell on that one though.