So, Intel has bet the farm on its upcoming 18A node. Meanwhile, the company just doubled down on what was supposedly an ultra-aggressive existing roadmap by revealing plans for a new 10A or 1nm process in 2027. But here's the thing. In terms of products you can actually buy, Intel has achieved scarily little since Pat Gelsinger arrived and subsequently rolled out the so-called "IDM 2.0" master plan.
Just to recap, IDM 2.0 was the new strategy Intel unveiled just under three years ago. The IDM bit spells out Integrated Device Manager and the top-level ideas involved reasserting its leadership position in chip design and manufacture and then carving out an additional role as a major provider of chip foundry services to third parties, including rivals such as AMD and Nvidia.
A major part of that master plan, something which superficially seemed like an incredibly ambitious self-imposed roadmap, comprised what Intel claimed were no fewer than five new chip production nodes to be rolled out within four years, culminating in that all-important 18A process. Intel predicted this would return it to the top spot in terms of advanced chip technology, overtaking TSMC in the process.
Now, upon closer inspection, you could argue that Intel's plan has in reality added up to far fewer than five nodes. Officially, the new nodes in question are Intel 7, Intel 4, Intel 3, Intel 20A, and Intel 18A.
A cynic might argue that Intel 7 is really just a rebrand and tweak of its existing and troubled 10nm node, Intel 3 is derived from Intel 4, and the same applies to Intel 18A with respect to Intel 20A. In which case, at the time of announcing the new roadmap, Intel was really only committing to two fully new nodes, Intel 4 and Intel 20A.
But it was still very ambitious given Intel's frankly terrible recent track record with new silicon nodes. Back in 2012, Intel was planning on unleashing 10nm CPUs as soon as 2015. In reality, it didn't manage to launch a truly commercial 10nm product until September 2019 with Ice Lake. It was then over four years before it managed, just, to release a CPU on the next fully new node, Intel 4, when its latest Meteor Lake mobile chips stumbled onto the market with just a few weeks of 2023 remaining.
And, remember, Meteor Lake only contains a tiny slither of Intel 4 silicon. Most of the chiplets that make up a Meteor Lake CPU are produced by TSMC, not Intel. Only the actual compute tile is an Intel 4 chip. So, you could say that all Intel has achieved since Gelsinger has taken over, at least in terms of chips you can buy, is a tiny volume of Meteor Lake CPU dies on that Intel 4 node.
For sure, there have been other CPU launches, including the Alder Lake and Raptor Lake families. But those didn't get Intel any closer to delivering on that mission-critical roadmap. And here we are in 2024 and Intel has until the end of next year to deliver on the whole thing.
All of which raises the question of what, exactly, is going on at Intel. Some answers were to be found at the recent Intel Foundry Direct Connect event. Intel showed a graph mapping out its planned chip production capacity up to 2029.
What it revealed was actually a little shocking. According to Intel's plans, its overall capacity to produce chips is shrinking slightly over the 2023 and 2024 period and won't again exceed 2023 levels until 2027.
What's more, that same graph shows that even in 2025, chips produced on the Intel 4 and 3 nodes (which it sensibly brackets together for this measure of its performance despite characterising them as distinct nodes for its PR-friendly roadmap) will be a very small minority and, oddly, will have already been slightly overtaken by the 20A/18A node (again, bracketed together because, really, they are two versions of essentially the same node).
The real measure of success won't come until the latter half of the decade. Meaning it's still an awfully long way back for Intel.
Capacity for both the Intel 4/3 and 20A/18A nodes will grow slightly in 2026, but it's not until 2027 that Intel itself expects to be banging out a proper ton of chips on those new nodes. That isn't quite what you might have expected when Intel announced that bold roadmap remap back in 2021, is it? Five new nodes in four years, except it's only two nodes and we won't be making anything with them in large volumes for six years, might have been more accurate.
Anyway, that wafer capacity data Intel revealed helps to make sense of the fact that Intel is planning to use TSMC to help with its CPU manufacturing for several years to come. It's not until at least 2026—and really not until 2027—that it will have enough internal capacity to make large volumes of CPUs on its newest nodes.
The graph also makes sense of rumours that even the CPU die in its upcoming Arrow Lake desktop CPUs will mostly be made at TSMC. Arrow Lake, famously, will be Intel's first consumer product to use its new 20A node for the CPU die, with the other chiplets in the CPU package such as the IO die and the graphics die made by TSMC.
Best CPU for gaming: The top chips from Intel and AMD.
Best gaming motherboard: The right boards.
Best graphics card: Your perfect pixel-pusher awaits.
Best SSD for gaming: Get into the game ahead of the rest.
However, stories are circulating that most models within the Arrow Lake CPU stack will have TSMC silicon for all chiplets including the CPU die. Only a few models with certain core counts will get Intel 20A silicon, or so the story goes. When you look at how long it's going to be before Intel ramps up its 20A capacity, according to its roadmap, those rumours suddenly make a lot of sense.
In that context, the idea that Intel is plotting "10A" silicon for 2027 does seem a bit farcical. It'll only just be getting going properly with Intel 4/3 and 20A/18A by then. And we haven't even mentioned the new 14A node.
When you add it all up, then, Intel's bold roadmap wasn't so brave after all. Maybe Intel will get a few Arrow Lake SKUs out the door this year with a tiny bit of 20A silicon inside. That will allow Gelsinger to claim everything is on track. But it will be a pretty cosmetic achievement. Frustratingly, the real measure of success won't come until the latter half of the decade. Meaning it's still an awfully long way back for Intel.
