The Flagship 18A Intel Xeon Clearwater Forest Just Moved

3
Intel Xeon Clearwater Forest
Intel Xeon Clearwater Forest

On today’s Intel earnings call, we heard a bit about future products and processes at the company. 18A is still on track to ramp production in the second half of 2025, but the flagship product codenamed “Clearwater Forest” seems to have moved. It might be the timing of milestones, but it feels like something has shifted.

Did the Flagship 18A Intel Xeon Clearwater Forest Just Move?

When we covered the first E-core Xeon launch, theĀ Intel Xeon 6 6700E Sierra Forest three quarters ago, the Intel Xeon 6900E was clearly on the roadmap. That is the Sierra Forest-AP part that we covered in our Substack that Intel is changing its positioning on and why that is a positive tailwind for AMD. It still has customers, it is just how it is being sold will change.

Intel Xeon 6 Roadmap
Intel Xeon 6 Roadmap

The other big E-core part that we covered several times on STH last year was set to be the 2025 Clearwater Forest.

Intel DCAI Update March Xeon Roadmap
Intel DCAI Update March Xeon Roadmap

We did not just talk about it in the abstract. In August 2024, Intel said Clearwater Forest and Panther Lake Booted on Intel 18A. By September 2024, we had seen the chips and the new packaging technology.

Intel Xeon 6 Clearwater Forest Close
Intel Xeon 6 Clearwater Forest Close

Intel was in a bit of a pickle. Sierra Forest-AP was set to launch in Q1 2025, and Clearwater Forest was targeted at Q3 2025. Then the Q4 2024 earnings call happened and we learned:

“We are also making good progress on Clearwater Forest, our first Intel 18A server product that we plan to launch in the first half of next year.” (Source: Michelle Johnston Holthaus Intel co-CEO on Intel Q4 2024 earnings call)

Later in the call, MJ said that the E-core market has not materialized as fast, and that the packaging side of 18A was the reason it was moving to 2026.

Now it is probably worth discussing what is a “launch.” Is it when there is a marketing event? Is it when parts start to ship for revenue even if they are not publicly available? Those are all valid questions and ones that apply not just to Intel but AMD and NVIDIA as well.

Some Important Context

I am going to give one of perhaps the most challenging takes on what is happening. There is a very real possibility I am wrong, but I wanted to layer some context here because a lot of folks covering this whiffed on the market dynamics part.

In the server market right now the macro trend is that AI servers are fairly easy to sell because there are still GPU constraints in the market. The demand is very strong. To fund that, companies are looking to web servers and relatively lower-performance machines and saying things like: “I understand that I can take a 4-year old machine and consolidate 8:1 using today’s new cloud-native processors, but I can also spend my budget this year on AI servers and let those servers run another year or two.”

That mindset is having huge implications in the server industry. We test many servers. I can tell you that whether you pick 4:1, 6:1, 8:1, or even more, a modern cloud native processor is great for web servers and other cloud native workloads. It is clearly one of the biggest revolutions in servers we have seen in more than a decade.

Intel Sierra Forest 6700E Pads 1
Intel Sierra Forest 6700E Pads 1

The problem is: The innovation is in a space that is not receiving a lot of attention.

It impacts more than just Intel. For Ampere Computing, as an example, it is trying to sell Arm servers that are tailored for cloud native workloads, but they are facing the same challenge that spend is being allocated to GPU servers. While we have shown platforms like the 16x NVIDIA GPU 128 Core Arm Server from Supermicro and the Gigabyte Ampere Altra Max and 4x NVIDIA A100 platforms, the GPU server trend is for higher-performance per core CPUs than what we would see in the cloud-native optimized chips.

I think Intel actually has a great product for the cloud-native market. Sierra Forest as a first gen design is very useful. At the same time, folks know that Clearwater Forest is going to be the big one and Sierra Forest is more of a part to satisfy some demand until the new part launches. That is why pushing a Clearwater Forest “Launch” to Q1 is disappointing but it might also make sense. On the server side it does not help, but if 18A is ramping in the second quarter of this year, and there are more packaging challenges on the Clearwater Forest side, then why allocate 18A capacity to a server part in a market where buyers are consistently allocating their spend to AI servers instead? A “launch” of a server CPU often means that large customers already have parts it is just the mass market is getting access, so we shall see what it means in this context.

Final Words

Personally, I was looking forward to Clearwater Forest in Q3 2025, so this is a bummer. I also think the next question is that Intel has been saying its packaging proficiency would be an asset beyond just 18A so saying Clearwater Forest is now set for Q1 2026 because of packaging challenges does not sit very well. To me, the bigger challenge facing the cloud native processor folks including Intel with its E-cores is that spend on the cloud native processor segment has been re-allocated to the AI segment at a surprising rate. Between this and the changes to the go-to-market for the Intel Xeon 6900E “Sierra Forest-AP” series, it feels like cloud native is not going to be the big 2025 push.

On the other hand, if you are building infrastructure for hosting, general VM, and so forth, Sierra Forest-SP (the Intel Xeon 6700E series) is probably doing better than many thought. AMD has Turin Dense. Ampere has its Altra Max and AmpereOne. All of thse options can offer fairly significant cloud-native consolidation today.

3 COMMENTS

  1. It’s refreshing to read this from someone who knows what’s actually happening. Your Substack article today on the Sierra Forest-AP change was very insightful as well.

  2. I was looking forward to a possible future of 512 Core in 2026 once they perfected Clearwater Forest. Now that seems to be further down the road. No wonder why AMD is also not in a hurry with Zen5c. The GPU budget issue makes a lot of sense. Now it will have to be Zen 6c before I could see a Dual Socket 1024 vCPU Server.

    I just wish Nvidia use this opportunity to make further inroad on Server CPU with ARM. That would put more competition in the Server Core / Price point.

  3. Another way of looking at the 8:1 consolidation is, if you can shrink 8 racks of CPU down to one, that’s 7 racks you now have that you can fill with GPU for AI. (The math isn’t that clean but you get the idea.) Considering how much money is sloshing around for AI and how tight the DC real estate market is, that might be more quickly and easily achievable than getting more infrastructure space.

LEAVE A REPLY

Please enter your comment!
Please enter your name here

This site uses Akismet to reduce spam. Learn how your comment data is processed.