Intel obviously did not stop at the announcement of its (bad) figures for the third quarter of 2023, many other subjects were addressed in the process by Patoche Gelsinger, the CEO. One of them was Lunar Lake, Intel’s latest generation for the mobile segment, a range of processors made up of Core Ultra 200V. Lunar Lake is special in several ways. First of all, the vast majority of the CPU components were manufactured (at a high price, it seems) by the Taiwanese founder and competitor TSMC. There is no doubt that this (largely constrained) choice of external rather than internal production also hurt Intel’s gross margin. With Lunar Lake, Intel has also experimented with integrating all components on a single package, including DRAM, with up to 32 GB of LPDDR5X-8533 on a 128-bit interface. On a technical level, there are certain advantages to doing this. Apple also adopted this model with its M series CPUs. But for Intel, although very efficient according to the CEO, this solution also turned out to be expensive and therefore affects the profitability of Lunar Lake. In addition, it offers less flexibility to manufacturers of PC. So, what future for it?
Well, it won’t have any, as the CEO explained during the conference with analysts and investors:
On-package memory is a unique case with Lunar Lake. This will not be the case with Panther Lake, Nova Lake and their successors. We’ll build them in a more traditional way with out-of-package memory in the CPU, GPU, NPU, and in-package I/O capabilities. But volume memory will be out of package in the upcoming roadmap.”
He went on to explain that Lunar Lake was initially intended to be a niche product, but changing market conditions changed that:
Lunar Lake was initially designed to be a niche product where we wanted to get the best performance and battery life, and then the AI PC came along. With the AI PC, we have moved from a niche product to a high volume product. In relative terms, we are not talking about 50 or 100 million units, but a significant part of our total mix from a relatively small share. So as this change took place, the impact on margins became more significant, both for Lunar Lake and the company as a whole.
There you have it, the message is quite clear. For various reasons, a priori mainly financial, Intel has no intention of continuing the experiment with DRAM integrated into the CPU package. We imagine that priority could then instead be given to the adoption of CAMM2.
Intel also spoke about Panther Lakethe next generation of CPUs from the founder. It is due to debut in 2025 and will be important for the company’s image, because along with Clearwater Forest, it will be one of the first large-scale commercial uses of its process Intel 18A. The ordeal by fire, what! But contrary to what has sometimes been speculated in recent months, Intel will not manufacture the entire Panther Lake CPU. The CEO confirmed that around 70% will be homemade, including the main compute tile (the compute tile). The rest will always come from TSMC.
In Panther Lake, some tiles will be external, but the majority of square millimeters in the package are again manufactured in-house. More than 70% of the silicon surface is homemade. Thus, the majority of Panther Lake’s wafer production capacity will return to Intel.
The return of production to Intel Foundry will therefore be done gradually. With Nova Lakethe reintegration of production will continue, but it will still not yet be complete. This is because the majority of the generation and more tiles will be manufactured in-house. This implies that certain references and components will continue in part to be sourced elsewhere, obviously, at TSMC. It is also possible that the Intel/TSMC share may vary from one model to another. Logically, the margin will be better with those manufactured mainly at Intel Foundry and lower with the others.
With Nova Lake, we definitely have a few SKUs that we plan to continue to leverage externally, but the vast majority of Nova Lake and more additional tiles will have come back internally as well. We still have some flexibility in the Nova Lake product, but the vast majority is engaged with Intel Product or Intel Foundry. So overall, we are executing perfectly on the strategy that we have set out, which is to bring the wafers home.
This gradual and cautious approach can be understood. Intel is never safe from a hiccup with the 18A, even though it would be very unenviable for the company at this stage. Fortunately, things seem to be going well. In any case, the CEO clearly thought about covering himself, since he stressed that Intel reserves the option of using an external process if it considers that it could prove beneficial to one or the other product. In other words, it is very likely that Intel chips will continue to come out of TSMC, even if most of them are manufactured at home. Nothing new under the sun, Intel has already used TSMC services on several occasions for less important components, such as Atom CPUs, FPGAs and ASICs. The current situation is unprecedented, because Intel has major chips in its portfolio manufactured by the Taiwanese. For example, Arrow Lake and Lunar Lake have almost nothing from Intel, which only does the assembly and packaging. (Source: Seeking Alpha via Tom’s, Tom’s)
TSMC has been a great partner. Lunar Lake has clearly demonstrated the strength of this partnership, which we will selectively use in our product lines in the future.
Pat is not ungrateful, he says thank you and doesn’t close any doors… Enough to patch things up with the Taiwanese founder and get his discount back?