British IP giant Arm recently made the big move of submitting an IPO application. However, the previously rumored high-profile anchor investors such as Apple, Intel, Nvidia, Amazon, and other tech giants, have yet to show themselves. It's estimated that Arm's market value can reach US$60–70 billion after the IPO, but what gives Arm so much confidence to uphold such an expensive IPO valuation?

It's worth noting that South Korean media has disclosed earlier that Samsung Electronics, a previously reported anchor investor, is rumored to back away from the deal. Is this also a sign that as the industry steps into the era of AI, companies are reevaluating the "indispensability" of Arm? This is in stark contrast to 2020 when Nvidia made a staggering US$40 billion bid to acquire Arm.

Arm's application document showed that SoftBank and Arm seem to be trying to ride the global AI trend to further boost its IPO valuation. They emphasized that Arm's technology is crucial to AI applications. They even explicitly stated that the CPU is indispensable in all AI systems, whether it's fully processing AI workloads or used in conjunction with processors like GPUs and NPUs.

In the eyes of Softbank CEO Masayoshi Son, riding the AI wave can boost Arm's valuation. Ever since Softbank acquired Arm in 2016, Son has always viewed Arm as the brightest star in Softbank's investment portfolio. He even stated in November 2022 that if weren't for Softbank's crisis, Arm would be the last asset they wanted to sell.

The issue is that Arm's focus is on CPU architecture foundations and not the GPU and AI chip architectures required to build AI LLMs. Thus, despite the ongoing AI global trend, Arm seems to still be on the outside looking in.

With the push from the AI trend, the growth prospects of data centers look promising. However, without the support of a successful Nvidia acquisition, Arm's architecture can only reach the outer edges and is unable to reach the core data center market.

The massive data processing demands of machine learning (ML) have led to a significant surge in sales of computing accelerators like GPUs, specialized AI chips, and network chips. While Nvidia's GPU needs to be combined with Arm's CPU architecture (e.g. Nvidia's latest super chip GH200 includes a CPU with Arm architecture), high-end AI GPUs from Nvidia or AMD can also be paired with x86 architecture CPUs to form computing modules. Arm CPUs are not the only option.

To a certain extent, Arm can certainly ride Nvidia's AI wave, but not all of Nvidia's GPUs need to be sold together with Arm CPUs. Nvidia CEO Jensen Huang previously stated that processors like traditional CPUs can no longer keep up with highly complex computing tasks.

With generative AI acting as the tipping point, the expansion of CPU computing capabilities has slowed down. Nvidia currently holds a dominant position in the AI GPU market. Therefore, the number of CPUs can be greatly reduced, as only a certain number of CPUs are needed to complement the millions of GPUs. This setup is enough to fulfill the high computing power demand for AI workloads.

Looking at Arm's main business models, the first is the CPU IP licensing model. The other is the royalty-based model that charges on a per-chip basis. In particular, with Arm's current mobile GPU technology, they haven't dedicated a lot of resources to high-end AI GPUs or AI accelerator chips. In its emphasized CPU sector, it still needs to compete with Intel, AMD, and other x86 architecture CPUs for data center AI opportunities. In this wave of AI server trends, Arm hasn't benefited a lot and can only look ahead to the next wave of edge AI opportunities.

By: DocMemory
Copyright © 2023 CST, Inc. All Rights Reserved