Track Hyper | Besides contraction, Intel has no other options

Author: Zhou Yuan / Wall Street News

The predicament faced by Intel has not ended with the departure of former CEO Pat Gelsinger.

Currently, Intel has not made any progress in new technology darlings such as AI accelerator cards, AI servers, and the smartphone sector; in the traditional CPU market, Intel's advantages have also diminished, facing strong pressure from competitors like AMD.

The interim co-CEOs of Intel, Michelle Johnston Holthaus (also serving as Product CEO) and David Zinsner (also serving as CFO), have been in office for two months, but aside from halting some products promoted by Gelsinger, such as stopping the market launch of Falcon Shores (the first AI accelerator chip), modifying the CPU technology roadmap, and delaying the release of the 18A technology node, they have not provided more effective solutions.

Two months into their tenure as co-CEOs, these two top leaders of Intel admitted during the Q4 2024 earnings call that they do not have a quick solution to the significant challenges Intel is currently facing.

This statement has been interpreted by the market as a "balanced and pragmatic" attitude.

In reality, the inability of these two CEOs to present effective solutions has led the market to believe that Intel's transformation journey may take several more years. Additionally, the work to spin off Intel Foundry (IF) into an independent business will also require a similarly long time.

The Intel 18A process technology, originally seen as capable of surpassing TSMC, has also seen a slowdown in its launch speed.

The flagship product representing this technology is Intel's Clearwater Forest, which has now been postponed to the first half of 2026.

This is about 12 months later than the original plan.

Clearwater Forest is Intel's new generation Xeon efficiency core (E-core) processor (the previous generation product is Sierra Forest) and is one of the first products to adopt Intel's 18A manufacturing process, as well as the industry's first chip to feature both back power delivery and GAA technology.

This processor is characterized by integrating three process technologies: Intel 7, Intel 3, and Intel 18A, targeting data center server CPUs.

The core consists of five blocks, with the left and right blocks based on Intel 7 process providing I/O module functions, while the middle part of each block integrates four computing modules based on Intel 18A process and a basic module using Intel 3 series process, which includes the main SRAM cache, voltage regulation circuit, and internal communication network.

Clearwater Forest adopts a more aggressive 3D Chiplets packaging strategy, featuring the new Darkmont microarchitecture, which is a slight upgrade from Skymont.

This chip utilizes Foveros Direct 3D stacking technology to directly stack the computing core and cache module, significantly enhancing performance and bandwidth efficiency, allowing data to transfer faster between the core and cache, reducing data wait times, and improving overall operational speed At the same time, due to the adoption of EMIB 3.5D technology, it can further optimize communication between different chips, reduce latency, and enable smoother collaboration between multiple cores, thereby enhancing overall system performance.

In terms of performance and process technology, Clearwater Forest is extremely powerful, but the cost is too high, market validation will also take time, and there are shortcomings in the ecosystem; more importantly, this server CPU has higher power consumption compared to similar chips using ARM architecture from AMD.

The delay caused by Intel 18A affects not only Clearwater Forest but also Panther Lake (including Granite Rapids: the sixth-generation Xeon server CPU designed for HPC), which is a mobile processor originally scheduled for release in the second half of this year. Panther Lake will mainly be manufactured by Intel itself, with a small portion outsourced to TSMC.

Regarding Intel's own manufacturing, it is different from the general sense of self-manufacturing.

Intel has recently spun off its IF department, making it a subsidiary of Intel with an independent board of directors, primarily engaged in chip foundry business. Holthaus stated that Intel views this subsidiary as a role similar to that of other suppliers, and it will not have preferential treatment; this subsidiary must compete for Intel's chip foundry business just like other suppliers.

The subsequent iteration of Panther Lake, the all-platform PC processor Nova Lake (replacing the current Arrow Lake), is scheduled for mass production in the second half of 2026, using Intel 14A, which is a level higher than Intel 18A (roughly equivalent to TSMC's 2nm). It is unlikely to be launched on time, and self-manufacturing may not be achieved, so it will likely be fully outsourced.

However, Holthaus stated that Nova Lake will still partially use its own processes, with another portion outsourced. Given the current situation, it is extremely difficult for Intel to manage Nova Lake's manufacturing on its own, and the realistic possibility of completion is very low.

In addition to the delay in mass production of processors using related processes due to the postponement of the Intel 18A technology node, existing products like the second-generation Core Ultra processor—Lunar Lake (launching on September 4, 2024)—will actually reduce Intel's overall product profit margin and gross margin due to cost issues.

Zinsner explained that the memory and packaging costs of Lunar Lake are too high, leading to pressure on Intel's performance in 2025.

A product that has already been launched is facing cost issues, which may be related to the price war that Intel is currently promoting. For example, Granite Rapids (P-core, 3nm) saw its price significantly reduced from $17,800 to $12,460, a reduction of up to 30%.

Zinsner stated that the aggressive pricing strategy Intel is adopting will inevitably affect product gross margins. "By 2025, our business will not show significant improvement; our cost structure will not improve at all before the launch of Panther Lake." The most practical approach that the two interim co-CEOs of Intel can come up with now is to comprehensively shrink the battlefield