On Intel's Q1 Earnings: A Triple Test of CPU, 18A Yield Rates, and Foundry Orders

[Fred Chen]

This article was written shortly before Intel's Q1 Earnings Call, but it still gives helpful perspectives on what Intel's future holds. Here are some excerpts:

On April 8th, SemiAnalysis' chief analyst pointed out that CPUs are facing a very severe production shortage. Currently, the ratio of CPUs to GPUs in AI data centers is roughly four to eight, but driven by the widespread application of AI agents, this ratio will soon become one to one or one to two.

The company also pointed out that current production capacity cannot fully meet strong demand, and supply bottlenecks are expected to peak in the first quarter and gradually ease starting in the second quarter. However, the shortage of memory chips has constrained the progress of PC assembly, and full-year ODM shipments are expected to decline by double digits. The revenue of the CCG division may decline by more than 13% quarter-on-quarter, which will partially offset the growth momentum brought by the server business.

If the supply and demand dynamics of server CPUs determine how many can be sold at the moment, then the mass production curve of the 18A process points to a more fundamental question: how much does Intel actually earn for every chip sold?

The 18A process has reached the high-volume production stage, with a yield rate maintained above 60%, and commercial profitability mainly depends on this condition. Chen Liwu believes that the yield rate was not ideal when he took over, but after introducing professional partners such as PDF Solutions and KLA, the yield rate has been rising at a rate of seven to eight percentage points per month.

KeyBanc points out that while the current yield is lower than TSMC's initial 70-80% yield in the early stages of 2nm mass production, it is better than Samsung's SF2 yield of less than 40%. The 18A uses RibbonFET all-around gate transistors and PowerVia back-end power supply as two important technologies, which improves performance per watt by 15%, increases chip density by 30%, and reduces overall power consumption by about 25%.

Panther Lake is the first consumer product to use the 18A platform, which will be officially unveiled at CES 2026. The platform has a total computing power of 180 TOPS.

Chief Financial Officer David Zinsner has previously stated that current yield rates can support shipment volume, but not normal profit margins. He expects to reach the desired cost level by the end of 2026 and to be in line with the industry average by 2027.

Whether the gross margin returns to normal levels after the mass production of the 18A can be used to determine whether Intel has truly returned to a profitable track.

Intel's foundry business is the most imaginative chapter in its transformation story. A UBS research report indicates that Intel is currently at a critical stage in securing several important contracts, and the release of version 1.0 of the 14A process design kit is seen as an important milestone.

https://www.tradingkey.com/zh-hans/analysis/stocks/us-stock/261794766-intc-intel-q1-cpu-ai-tradingkey

 

[Fred Chen]

During the call, DZ said,

"Intel Foundry operating loss in Q1 was $2.4 billion, improved $72 million quarter-over-quarter as better yields across Intel 4, Intel 3, and 18A drove higher gross margins. This was mostly offset by increased operating expenses associated with an intentional step-up in Intel 14A investments to support both internal and external customer evaluations."
He also noted that "Intel 18A is still early in its ramp, and rising input costs, especially in memory, present growing headwinds in the second half that we need to overcome."
"18A is going to be a pretty decent headwind to our gross margins. If you look at Panther Lake volume increases, it is going to be up six or seven times in the second quarter relative to the first quarter."
Perhaps not surprisingly, using outside foundries, i.e., TSMC, is a key part of enabling Intel to increase wafer starts. You need tool availability besides yield and throughput.

Here's the relevant exchange between C. J. Muse and DZ and LBT:

C.J. Muse: Good afternoon. Thank you for taking the question. Could you walk through how you are planning to drive increased output through the second half of the year? How much of it is yields? How much of it is cycle times? How much is incremental wafer fab equipment, as well as outsourcing to TSMC?

David Zinsner: First and foremost, we are increasing wafer starts in all three of our nodes—Intel 10/7, Intel 3, and 18A. More meaningfully on the EUV nodes, of course, but even Intel 10/7 will be increasing wafer starts this year. That is a key component of our ability to meet demand. That said, Lip Bu has pushed the team really hard to provide more supply the old-fashioned way with better yields and better throughput, and that is largely how we got it in the first quarter. We can expect him to do that through the year, and I think that will be a meaningful contributor to our output. Of course, we use outside foundries as well, and we flex them as needed.

Lip Bu has great relationships with the external foundries, and he is able to leverage that to help us in that area as well. There will certainly be a component of that as we move through the year.

Lip Bu Tan: Just to add, TSMC is a very important partner for us. Morris and C.C. and I have decades of friendship. Our product groups will decide which is the best foundry. We are going to use a multi-foundry approach—our own internal and external—so we can benefit customers.

 

[Fred Chen]

Regarding new foundry customers:

Timothy Arcuri: Thanks a lot. Lip Bu, I wanted to ask about the evolution of your foundry model. You are of course pursuing typical foundry customers, but it seems like TeraFab is a little bit of a different deal and maybe even like a process licensing agreement. I would not normally ask about one particular customer, but he did talk about it yesterday. Is that going to be a typical foundry arrangement, or are you possibly going to turn the keys over on an entire fab to them?

Lip Bu Tan: Yeah, Timothy, thanks for the question. On 14A, we are making great progress in terms of yield and cycle time, and we are engaging heavily with multiple customers. My style is under-promise, over-deliver, so we have no plan to announce the customer unless the customer wants to announce it, and we support that. Back to TeraFab, clearly Elon and I believe that global supply is not keeping pace with the rapid acceleration in demand. We both share the vision that we are going to learn a lot together, exploring innovative ways in process and manufacturing. It is a very broad relationship, and we will update you as we go.

Clearly, this is a very exciting customer to work with, and we have multiple other customers we are engaging. Stay tuned.

 

[MKWVentures]

Some notes on foundry:
revenue went up 800M YOY based on higher wafer prices for 18A, Intel 3 wafers.
Costs went up 900M YOY based on higher wafer costs and spending for 18A
Total loss went up YOY from 2.3 to 2.4B (details in the 10Q)

According to Intel, Wafers are sold to the product group at market prices (TSMC price). So die yields should not factor into Intel foundry P&L, Correct? it impact product group GM.

Next quarter we will start to see the new finances for Intel Foundry: Impact of Ireland buyout and loans. Impact of 18A ramp (I think DZ said 6x more volume). Impact of increased Spending to support 18a/14A internal and external customers. Maybe also see what exactly is the "partnership" with Tesla.

Intel doesnt need to announce the customer names. Intel has and TSMC does report out tapeouts. Intel also has to announce prepayments in 10Q unless they are trivial.

On packaging and wafer sales, the issue is not die yield (Intel will figure that out). It is Wafer cost and development cost and capex. On packaging it is gross margins.

Lots of fun and announcements coming next quarter

 

[MKWVentures]

One other item:
In past comments, I said intel had this problem of people wanting 7nm product and not Intel 3 product. And Intel cannot force them to the new node like in the past as they could just jump to AMD
With a DC CPU shortage, Intel now can force encourage them to Intel 3 products and eventually Intel 18A products. This allows Intel to start filling fabs on new nodes (last quarter neither 34 nor 52 were running a planned capacity). Intel is playing that card now and I am hearing that Granite rapids is finally picking up 2 years after announcement. Ironicly, Intel added capacity to Intel 7 in Q1/Q2. Still need to compete with AMD and try to regain share but at least Intel has SOME leverage now. This and Wildcat lake should hopefully allow Intel to PROFITABLY reduce loadings on a 7/10 technologies and get more than 20% of its chips on EUV technologies. As the memory guys know, shortages are ALWAYS good for suppliers

 

[siliconbruh999]

In past comments, I said intel had this problem of people wanting 7nm product and not Intel 3 product. And Intel cannot force them to the new node like in the past as they could just jump to AMD

Cause Intel doesn't have a good product on Intel 3 for client purpose.

With a DC CPU shortage, Intel now can force encourage them to Intel 3 products and eventually Intel 18A products. This allows Intel to start filling fabs on new nodes (last quarter neither 34 nor 52 were running a planned capacity). Intel is playing that card now and I am hearing that Granite rapids is finally picking up 2 years after announcement. Ironicly, Intel added capacity to Intel 7 in Q1/Q2. Still need to compete with AMD and try to regain share but at least Intel has SOME leverage now. This and Wildcat lake should hopefully allow Intel to PROFITABLY reduce loadings on a 7/10 technologies and get more than 20% of its chips on EUV technologies. As the memory guys know, shortages are ALWAYS good for suppliers

Intel needs to have products DC products on 18A/AP there is none yet and none will be in like sooner time frame cause DMR is more or less delayed