SoftBank andArm

The explanations offered at the time

SoftBank gave a growth explanation. Its chief executive, Masayoshi Son, said that “ARM will be an excellent strategic fit within the SoftBank group as we invest to capture the very significant opportunities provided by the ‘Internet of Things’” (SoftBank announcement).

Investors did not object to the strategy so much as to the buyer's position. SoftBank's shares fell about 10 percent in Tokyo on the first trading day, and fund manager Mitsushige Akino said that “a fresh acquisition is not what the market wants from SoftBank,” which was still carrying a large debt load (Fierce Network, citing Reuters). Both views treat the price as a bet on how fast Arm's markets would grow.

How Arm earns money

Arm designs the instruction set a processor is built around, which is the basic vocabulary the chip understands, together with reference designs for the processor cores. It licenses both. A licensee such as Apple or Qualcomm pays an upfront fee for access, then a royalty on each chip it ships. The royalty is a few cents a chip, and the industry was shipping more than fifteen billion Arm-based chips a year at the time of the acquisition.

Two features make that income durable. Software compiled for one instruction set does not run on another until it is rebuilt and retested, so every licensee and every customer of a licensee carries a cost to moving away. And Arm does not manufacture chips, so it does not compete with the companies licensing from it. That mattered to licensees who competed fiercely with one another and were willing to share a common foundation only because its owner had no product of its own to favor.

Both features describe the licensing position and neither depends on smartphones. Consider a hypothetical licensee building a processor for a server. If it starts from an Arm design, its operating system, compilers and customer software all already exist. Choosing something else means paying for all of that again, so Arm's royalty follows the design into the new market.

What happened afterward

The record is consistent with that reading. Amazon built server processors on Arm designs for its own cloud, and Nvidia used an Arm-based processor alongside its accelerators. By 2026 industry figures put Arm-based machines at a substantial share of data centre server revenue, though a much smaller share of unit shipments, and the two measures differ enough that a reader should check which one is being quoted. SoftBank listed Arm on Nasdaq in September 2023 at $51 a share, a valuation of about $54.5 billion (eeNews Europe), and kept most of the company.

What this shows is limited. Growth outside mobile occurred, and the royalty structure was in place when it did. It does not show that SoftBank foresaw data centre demand. The public case in 2016 concerned connected devices, and nothing in the announcement mentions servers. What I infer is that the licensing position was a better asset than the end market that then supplied most of its revenue, and that a buyer who understood the position did not need the end market to grow quickly.

The strongest objection

The objection is that Arm's position is less secure than a royalty stream suggests, because an open instruction set called RISC-V gives licensees a way to build chips without paying Arm. If licensees can leave at an acceptable cost, the royalty is a price and not a position.

Two facts would decide the question. One is the share of new chip designs that choose RISC-V over Arm in each market. The other is how licensees behave when Arm's interests diverge from theirs, which became a live question when Arm announced a server processor of its own in March 2026. Rising RISC-V share in designs for phones or servers would weaken this note's argument.

What a reader should take from it

When a licensing or royalty business is sold, the useful questions are how costly it is for licensees to leave and whether the licensor competes with them. The current end market matters less than both. A buyer paying a high premium for a business whose end market is slowing is not necessarily mistaken, and a reader who wants to check the price should look at licensee behavior and not at unit growth.