Birth of "Intel Foundry"

Now a bit over two years later, and Intel is just starting to see the first fruits from that aggressive roadmap, both in terms of technologies and customers. Products based on Intel’s first EUV-based node, Intel 4, are available in the market today, and its high-volume counterpart, Intel 3, is ready as well. Meanwhile, Intel is putting the final touches on its first Gate-All-Around (GAAFET)/RibbonFET for 2024 and 2025. It’s a heady time for the company, but it’s also a critical one. Intel has reached the point where they need to deliver on those promises – and they need to do so in a very visible way.

To that end, today Intel’s Foundry group – the artist formally known as Intel Foundry Services – is holding its first conference, Direct Connect. And even more than being a showcase for customers and press, this is Intel’s coming-out party for the fab industry as a whole, where Intel’s foundry (and only Intel’s foundry) gets the spotlight, a rarity in the massive business that is Intel.

Along with outlining Intel’s progress on meeting their 5 nodes in 4 years goal, Direct Connect is also Intel’s first chance to talk about what will come after those first 5 nodes. As Intel Foundry expands in capacity, customers, and tooling, the group is looking at a slate of not only even more advanced nodes, but also a slew of increasingly necessary packaging technologies to back them up. And while today’s event won’t match the overall audacity of Gelsinger’s 2021 proclamation, it’s still an important look at what’s in store over the next several years for the once (and future?) foundry king.

Altogether, there are several announcements of note here, so let’s dive right in.

14A will be Intel’s first production use of High-Numerical Aperture (High-NA) EUV, the next generation of Extreme Ultraviolet lithography. High-NA EUV promises even finer features, allowing wafers to be processed without relying on multiple patterning, which is expected to become necessary with conventional EUV at smaller node sizes. Intel has just about bet their foundry business on High-NA, a sharp contrast from Intel being relatively late in the game to pick up EUV (Intel 4/Meteor Lake being their first product), to the point where Intel has secured the world’s sole prototype High-NA scanner.

The long and short of matters here is that Intel is reiterating once more that the 4 year plan remains on track. Intel’s 4 year plan ended with 18A being production ready in 2025, and in 2024, customers can already begin designing chips for Intel’s most ambitious node.

Notably here, Intel has recently finished – and is announcing today – the tape-out for their own 18A lead product, Clearwater Forest. Clearwater is Intel’s second-generation E-core based Xeon (the successor to Sierra Forest), and is a tour de force of Intel foundry technologies. Besides 18A for the compute elements, Clearwater uses Intel 3 for its base die, EMIB for further die connetions, and even Foveros Direct (hybrid bonding) for those die-to-die connections. Clearwater will eventually be joined by the consumer Panther Lake as Intel’s first two big 18A projects.

With its combination of feature size, RibbonFET transistors, and PowerVia backside power delivery, Intel has previously stated that they expect to regain process leadership with 18A. And as of today’s event, that remains Intel’s projection for when they’ll return to the top.

Meanwhile, a bit closer to production, Intel is reporting that Intel 3, their high-volume EUV process node, is ready for high-volume manufacturing. It’s predecessor, Intel 4, is already shipping today for Meteor Lake, and intel 3 is the refined version of it with a full range of cell libraries available (rather than the high-performance-only Intel 4).

Given that Intel is only shipping products using the second of their 5 nodes at this point, there is no getting around the fact that, at least as an outside observer, a lot of Intel’s “on track” announcement is taking the company’s word for it. But given that Intel’s timeline from the very start has been based on internal (risk production) milestones and not product shipment milestones, it was never going to be any other way.

Still, absent Clearwater Forest chips in our hands today, the fact that they have designs taped out and are ready for customer designs is about as promising a sign as one could hope for.


I speculated years ago that Intel would eventually go fabless (see AnandTech Forums). It hasn’t quite turned out that way. But if you, as I did then, consider Intel to be at the core a product designer competing against AMD, then they are pretty much there now — Intel Product division a separate legal entity operating as a fabless design house.

The question now is only whether this Samsung conglomerate model holds for the long-term, or ownership will split down the road. While Intel obviously understands the need for confidentiality and independence between the entities, potential Intel Foundry customers may look for more than reassurances.

"Under the dramatic reorganization and rebranding, we now have Intel Foundry and Intel Product—separate legal entities that remain parts of the same overall company but even have their own sales forces and back-end business systems. Intel Foundry will obviously make the processors that Intel Product develops but via arm’s-length transactions. Intel Foundry’s other customers get strict confidentiality, and everyone’s welcome. “If we’re going to be the Western foundry at scale, we can’t be discriminating in who’s participating in that,” Gelsinger said yesterday.”

Intel splits itself in two to aid CEO Pat Gelsinger’s turnaround plans—and links up with archrival Arm (

"These organizations will be legally distinct and have their own staff and processes with minimal overlap to ensure confidentiality, explained IFS head Stu Pann. “We’re very disciplined about this. We have two separate sales forces. We’re building two separate ERP systems. We’re outlining two separate legal entities,” he said. “Intel Foundry will do arm’s length transactions with the Product groups.”

Gelsinger splits Intel in two to advance foundry vision (

"Well, if you go back to the picture I showed today, Paul, there are Intel products and Intel foundry, There’s a clean line between those, and as I said on the last earnings call, we’ll have a setup separate legal entity for Intel foundry this year, " Gelsinger responded.“We’ll start posting separate financials associated with that going forward. And the foundry team’s objective is simple: Fill. The. Fabs. Deliver to the broadest set of customers on the planet. […] So I want my foundry to be used by everybody. Period. We want to help build Nvidia chips, and AMD chips, and TPU chips for Google, and inference chips for Amazon. Period. We want to help them and give them the most powerful, performant, and efficient technologies for them to build their systems. Period. Full stop.”

Intel CEO Pat Gelsinger: I hope to build chips for Lisa Su and AMD | Tom’s Hardware (

The Intel CEO almost stamping his foot proclaiming he wants to build AMD chips to fill fabs — who would have thought that back in the day?


I have always viewed Intel as a manufacturing company. It has been said their real business is construction, but I think that was a little in jest. In reality, X86 really isn’t all that, but because of the “Other Things” Intel is, it won. Clearly ARM is now the more dominant architecture though.

Even though Intel is screaming “open, open, open”, keep in mind Intel product groups get the leading edge technologies about a year ahead of foundry customers. While this is irrelevant when TSMC has the better technology it could create some interesting dynamics in the future. From my view, we see Intel_20A clients in 2024, Intel_18A servers in 2025, and foundry in 2026.

That said, there is HUGE risk in the Intel20A/18A schedule. B of A/ML continue to call Intel as neutral based on the assumption 18A slips significantly. Others have said that if Intel hits their 18A targets, everything changes. All we can do now is wait and see.


Does that continue to hold? Makes sense, I suppose, but may not be the way to make the most money – we just have to watch this unfold over the next few years.

I haven’t dismissed Intel, even Intel Product, yet. That said… Nvidia today… dang!

Yeah, I’ve come to realise that myself. And manufacturing appears to be the calling of CEO Pat Gelsinger. While he is the chief architect of the i486 and has done a lot of other great product related things at Intel, I think his heart lies in Intel’s manufacturing history. His passion seems to be to restore Intel’s capability in this area, in the footsteps and memory of Andy Grove and Gordon Moore, his mentors and role models.

And he apparently doesn’t mind producing chips for bitter rivals of the past to get there. The objective is a noble one: to rebalance the semiconductor supply chain and restore manufacturing capability and capacity in USA and Europe. I hope he succeeds — and that we one day see him holding up an Intel-produced AMD chip with Lisa Su on stage.

So, if Intel should eventually split up, it may be the product division that spins off — although with the brand recognition Intel has in the PC space, it is hard to imagine today.

1 Like

Yeah, I’ve come to realise that myself. And manufacturing appears to be the calling of CEO Pat Gelsinger. While he is the chief architect of the i486 and has done a lot of other great product related things at Intel, I think his heart lies in Intel’s manufacturing history. His passion seems to be to restore Intel’s capability in this area, in the footsteps and memory of Andy Grove and Gordon Moore, his mentors and role models.

And he apparently doesn’t mind producing chips for bitter rivals of the past to get there. The objective is a noble one: to rebalance the semiconductor supply chain and restore manufacturing capability and capacity in USA and Europe. I hope he succeeds — and that we one day see him holding up an Intel-produced AMD chip with Lisa Su on stage.

I’d disagree, somewhat. The opportunity isn’t pursued because it’s noble; it’s pursued because a) there’s an enormous gap in the market for chipmaking once you’re outside Taiwan and Korea, and no other company besides Intel could even consider going there-- it ain’t GloFo that’s going to be the next great foundry b) the x86 CPU opportunity looks less and less promising every day. Even AMD will probably return to ARM cores for mobile again at some point. Apple has proven beyond doubt that you can build world-leading performance around an Arm chip, given the M3 – which is a hellacious piece of equipment. (Seen a colleague running on his MacBook Pro with an M3 Max, I think, an LLM at ~7x the speeds I could get on an AMD-based 6000 series laptop with AMD discrete GPU, and probably close to that even with a 7000-series CPU and Nvidia GPU of last year’s vintage.) So… Either let Intel die replaying the last battle, or reinvent Intel, let the CPU business find its own way.


The M3 Max is a behemoth. Here’s an interesting piece of info I just looked up:

Nvidia H100 - up to 80B transistors
Apple M3 Max - 92B transistors

They’re both in the same ballpark size-wise and probably have similar processing power, certainly processing power / Watt. And Nvidia is charging something like $40k for an H100, while Apple charges substantially less (though they don’t sell raw chip systems, they sell products built around the chip systems).


And of course I have yet to see an H100 in a laptop form factor.

Starting to see info/rumors about the M4 surfacing. Apple has my respect, finally.

But imagine – a CPU and GPU in a package together, sharing an equally privileged interface to memory, and offering great performance on all workloads… Who would have thought?

M3 has basically convinced me that x86 has no real future. Thus the importance of Intel Foundry. AMD will shift to ARM-based designs if it needs to. Think how much of the cross-licensing agreement with Intel they can get out from under…

What have you heard? Googling now I can only find vague prognosticating about A17 and M4 possibly including beefier AI neural engine cores still produced on TSMC’s FinFET 3 nm node, with TSMC’s new GAAFET 2 nm “N2” node not being used until 2025 or 2026.

Basically, not much other than what you linked to. The M3 Pro is impressive, the M4 has to be another huge step up.

I won’t be happy with anything less than a 50% improvement in AI performance (not just performance/watt) from M3 to M4, even if M4 is still only on the 3nm node. :smile: But I suspect we’ll only see a 15% improvement at best.

Are you a current M3 user?

I saw this

A leaked set of performance benchmarks for Qualcomm’s Snapdragon X Elite Arm chip reveals a narrowing of the performance gap with Apple’s top-of-the-line M3 silicon.

The benchmarks published Monday on the Windows Latest website reported the Snapdragon X Elite 1800 had a single-core score of 2574 and a multi-core score of 12,562. By comparison, the M3 has a single-core score of 3181 and a multi-core score of 15,620.

The single-core score represents the performance of a single CPU core when running tasks that don’t benefit from multiple cores. The multi-core score represents the combined performance of all CPU cores when running multi-threaded tasks.

Windows Latest explained that the scores were generated using Geekbench, a cross-platform benchmarking tool that measures the performance of a system’s CPU and GPU, on a prototype device running early preview builds of Windows 11 version 24H2.

It also noted that the 2574/12,562 scores were the highest of five benchmarks generated by Geekbench. The lower scores could be attributable to optimizing the chip’s performance for conditions such as longer battery life or thermal throttling.

While the Apple M3 outperformed the X Elite on the Geekbench tests, Windows Latest added, it is important to understand that the Snapdragon X Elite was tested on a device running Windows 11 unoptimized preview builds, so when running on finalized hardware, it’s likely the performance numbers will be better.

is also interesting… Windows may prove something of an obstacle to parity-- overhead of various sorts?

As you may know, Apple Mac devices with the M3 chipsets run on macOS. The Snapdragon X Elite platform, on the other hand, is targeting Windows. According to Sascha Sagan, macOS is a much better operating system than Windows. That’s why the Qualcomm chipset won’t shine on laptops or notebooks even when it has good hardware.

To be exact, Sascha Sagan says, “The experience is not going to be the same because they’re running macOS and we’re running Windows.” She added, “But in terms of hardware, which is the only thing we can control, it’s good hardware.” Segan has also recommended that consumers should not purchase a new laptop right now.

No, just M2 in an iPad Pro, but, truthfully, I don’t even need the performance that the M1 would have provided. However, since I own a lot of Apple shares, I am always hoping Apple will continue to provide iterative massive hardware performance improvements now that they’re totally invested in “Apple Silicon” and ARM.

And yes, I had seen the performance claims and benchmarks pitting the Qualcomm X Elite against both the M3 and the latest Intel chips, but since Qualcomm is targeting Windows, and Apple is a closed hardware ecosystem, it’s Intel (and AMD) that have much more to be concerned about - perhaps - than Apple.

Apple’s “Walled Garden” is formidable, almost regardless of whatever hardware advantage or disadvantage Apple’s platforms might have.

That said, since I also own quite a few shares of AMD, and the X Elite seems to threaten the AI space, I remain a bit worried about AMD’s future in that arena.

And full disclosure: I own a few tens of shares of NVDA as well. Still kicking myself for not buying more a few years back! Ditto for TSM stock. I got in far too late on both of them I’m afraid.

  • John
1 Like

If I had bought Nvidia at the same time that I was so focused on AMD, I would have been retired years ago. Taiwan semiconductor I keep losing patience in. I need to be more persistent on that one. At this point I have a lot of Nvidia, but I’m trying to avoid massive concentration in a single stock.

I’m also trying to stay a bit more diversified outside of the chip business. Minting money on weight loss drugs, and some of the magnificent seven, but also positions in a number of other things.

I’m trying to resist the urge to just pile into Nvidia, cross my fingers and wait for the next wave.

1 Like

Ditto. Surely NVDA will have a pullback at some point.