Three Growth Drivers of Arm | AI Dehydration

Author: Zhang Yifan

Editor: Shen Siqi

Source: Hard AI

With the launch of Qualcomm's first AI PC, edge AI has once again reached a climax.

Recently, Bank of America released a report stating that ARM and Micron will become the two biggest beneficiaries in the AI PC and AI phone markets.

Benefits for ARM ——

1. Transition from X86 to ARM, increasing ARM's market share (Hard AI has previously analyzed how ARM is gaining new market share in the AI era in the article "ARM makes bold claims again, aiming to capture 50% of the Windows market");

2. Upgrading ARM architecture from ARM v8 to ARM v9. This new architecture can achieve higher royalty rates, increasing from 1-2% for ARM v8 to 2-4.5% for ARM v9, and supporting an overall computing solution with higher royalty rates, CSS (8-10% royalty rate) (details analyzed later in the article).

Benefits for Micron (MU) ——

1. Increasing demand for high-bandwidth memory in AI servers in cloud computing;

2. Annual growth rate of memory demand for AI PCs and AI phones will reach 12-15%;

I. ARM

Bank of America believes that in the future, ARM will benefit from the AI wave from three perspectives: ARM v9 architecture, overall computing solution CSS, and increased licensing revenue.

1) ARM v9 Architecture

AI is the catalyst for ARM's transition from v8 to v9 architecture. Bank of America predicts that by 2026, ARM v9 architecture will contribute 60%-70% of revenue (currently only 22%). The biggest highlight is that the fees for v9 are higher than v8, increasing from 2% to 4.5%. An increase in the proportion of v9 means revenue growth.

Compared to the v8 architecture, ARM v9 architecture improves overall chip performance and power efficiency. Additionally, this architecture enhances crucial security features for large models.

In terms of performance, ARM v9 architecture introduces X-level processor cores, enabling PC-level performance on smartphones.

Currently, all mainstream high-end Android phones (priced at $600 or above) use processors based on v9 architecture, such as Qualcomm's Snapdragon 8/7 Gen 3, MediaTek's Dimensity, and Samsung's Exynos, which have been using ARM v9 architecture for 1-2 years.

As of now, Apple has been the only major company using the older v8 architecture to design its processors. However, with the introduction of Apple Intelligence, this situation is about to change The new M4 processor for iPad and Mac (expected to be launched in 2024) will be based on the v9 architecture. Similarly, the A18 Pro processor planned for the iPhone 16 Pro series in 2024 is also expected to use the v9 architecture.

According to the company's financial report, in 2024 Q1, ARM's royalty income from the v9 architecture accounted for approximately 22% of the total royalty income. Bank of America predicts that by Q4 2026, this number will rise to 60%-70%.

2) Overall Solution CSS Drives Blended Royalty Rate Increase

Through ARM's overall solution CSS, ARM may double its royalty rate based on the v9 architecture, reaching 8-10%.

CSS is a set of pre-packaged processor building components introduced by ARM, pre-validated, integrated, and verified by ARM. This saves customers, especially chip designers (AP suppliers), a lot of time and resources as they do not have to build everything from scratch.

This is particularly useful for AP suppliers trying to compete with established manufacturers like Qualcomm, as they can quickly develop competitive chips using CSS.

Currently, chip manufacturer MediaTek has successfully taken market share from Qualcomm using CSS.

Bank of America states that the royalty rate for CSS is high, reaching twice that of the v9 architecture, driving the overall increase in ARM's royalty rate.

3) ARM - Increase in Licensing Revenue

ARM has adopted a dual-mode of licensing revenue (Licensing) + device usage royalty (Royalty) similar to Qualcomm's royalty income model. This means that as terminal devices increase in volume, the royalty income based on the number of devices will see growth.

In the past 1-2 years, ARM has signed new device quantity royalty agreements with new customers, and existing customers are also expanding agreements. The licensing revenue has already been reflected in patent licensing income, but has not yet been reflected in device usage royalty income.

This will become one of the important sources of growth for ARM.

2. Micron (MU) - AI Drives Memory Demand

Bank of America points out that with the proliferation of artificial intelligence, the demand for DRAM in edge AI devices is growing significantly, and this trend is expected to bring huge benefits to memory suppliers in the coming years.

• AI PC Memory Demand: In 2023, the average memory for personal computers is 13.5GB, while the minimum memory requirement for AI PCs is 16GB. It is expected that this number will further increase, exceeding 19GB by 2026 (compound annual growth rate of 12%); • AI smartphone memory requirements: By 2023, the average DRAM capacity of AI smartphones is expected to be around 7GB, and it is projected to reach about 11GB by 2026 (with a compound annual growth rate of 15%);

Specifically, in the iOS ecosystem, Apple's AI functions are only available on models with 8GB or more memory.

Currently, only about 61 million iPhones (7.6% market share) have sufficient memory to support this function, while the majority of iPhones (92.4%) are equipped with 2GB-6GB of memory, unable to utilize Apple Intelligence.

This means that the upgrade wave driven by AI will push the average memory of Apple phones to 8GB, which is twice the current average level.

In the Android ecosystem, Qualcomm's Snapdragon 7/8 Gen 3 processors support AI functions, requiring at least 8GB of memory. And 12GB is also becoming more common, such as in the Samsung Galaxy S24+/Ultra.

This trend, coupled with the wider adoption of AI processors, is expected to drive the average memory capacity of smartphones to exceed 7GB