Three months after Softbank announced its acquisition of Arm in 2016, Masa Son (founder and CEO of Softbank) invited Jensen Huang (founder and CEO of Nvidia) for dinner at his California mansion. Just the two of them.
Among the many things they talked about at this “dinner date”, chronicled in Aiming High (a biography of Masa originally published in Japanese), Masa eagerly shared with Jensen his desire to merge Arm and Nvidia someday. Masa had wanted to buy Arm long before 2016. But Masa had wanted to own a big chunk of Nvidia even more.
Fast forward to September 2020, when Nvidia announced its plan to acquire Arm, the deal was mostly in Nvidia shares, not cash, exactly how Masa would have liked it. In his own words, Masa described the deal as “a purchase kind of sale, a sale kind of purchase.” Had the Nvidia-Arm “marriage” been approved, Softbank would have become the largest shareholder of Nvidia. Masa Son would have completed his master plan that was at least four years in the making. Of course, that all fell apart last week.
The failed Nvidia-Arm acquisition has been heavily reported already, so I won’t repeat the main facts here. What was noteworthy was that the most fierce antitrust challenge of this deal came from US regulators, with heavy support from many big tech companies all fearing a monopoly of chip design IP under Nvidia. When the deal was first announced last year, I was anticipating Chinese regulators to torpedo the deal, using Arm’s embattled China division as leverage. As it turns out, the Chinese regulators didn’t even get their turn.
What was even more noteworthy, but buried under all the Nvidia-Arm headlines, was Intel making moves to get closer to RISC-V, the open source ISA (instruction-set architecture) that aims to become an alternative to Arm. With Nvidia and Arm breaking up, while Intel’s courtship of RISC-V begins, what’s ahead are new technologies, new alliances, new conflicts, and new battles – “Romance of the Three Kingdoms” style, except among semiconductor ISAs, not warring dynasties of ancient China.
New Romance: Intel loves RISC-V
On the same day that Nvidia officially called off its Arm acquisition last week, Intel made two announcements: a new $1 billion dollar fund for startups building in an open ecosystem of semiconductor IPs, and joining the RISC-V International foundation as a Premier Member.
This coincidence in timing was a bit unfortunate for Intel, because its pair of announcements should have received much more attention. Instead, most mainstream business media covered the Nvidia-Arm breakup, without mentioning Intel’s latest moves, even though what Intel is doing is highly relevant to the post-Nvidia-Arm world.
Intel, as the proprietor of the closed-source x86 ISA, supporting RISC-V is on the same industry-shifting scale as Microsoft’s eventual embrace of Linux. As all layers of technology get eaten by open source solutions, traditional tech giants who initially rose to prominence using a proprietary model simply must adapt. Intel is the latest example.
The Intel-RISC-V alliance likely would have happened whether the Nvidia-Arm acquisition succeeded or not, because of the age-old adage: “the enemy of my enemy is my friend.”
Arm is a huge threat to Intel – gaining market share in both laptop CPUs (see Apple’s M1 chip) and cloud computing servers (see AWS’s Graviton processor). Laptops and cloud servers, being less sensitive to power consumption than smartphones, have traditionally been dominated by x86-based CPUs. Not any more.
RISC-V, being open sourced, is a threat to every proprietary ISA, but less to Intel’s x86, more to Arm. That’s because both Arm and RISC-V are based on the RISC family of chip architectures, thus can address more overlapping use cases. (See more of our writings on RISC-V here.)
The enemy (RISC-V) of my enemy (Arm) is my (Intel) friend.
It’s hard to know whether a $1 billion fund is enough to jumpstart a RISC-V anchored ecosystem that can rival Arm. It may sound like a lot, but the break-up fee that Softbank will collect from Nvidia from the failed acquisition is $1.25 billion.
What is more valuable than the $1 billion is Intel’s public commitment to RISC-V – giving confidence to users of a still-young project that it has the backing of an industry stalwart. Intel’s membership also has two implications – one technical, one geopolitical.
The technical implication lies in a new concept Intel is touting with its foundry service: “open chiplet”. In plain terms, “open chiplet” is an alternative to “system on a chip” (SoC). SoC is a highly integrated way of packing all (or most) of a computer processor (CPU or GPU) onto a chip using one architecture. It is the industry standard right now. “Open chiplet” will be a modularized platform that allows semiconductor products to leverage multiple architectures at once, be it x86, Arm, or RISC-V. It is a classic open source way of doing things, embracing modularity and compatibility. Intel is no longer “x86 way or the highway”; it is trying to stay competitive and ahead of the curve by being an open chip design IP platform, as long as its foundry gets to manufacture the chips.
The geopolitical implication is evident if you do a before-and-after comparison of RISC-V International Foundation’s membership page. Prior to Intel’s logo being displayed in the “Premier Members” section, the page was dominated by big Chinese tech companies and Taiwanese companies like Andes (the only publicly listed RISC-V company, now an Intel partner). American semiconductor companies were nowhere to be seen. Intel’s entrance finally changes that “balance of power”, and may signal American industry’s growing embrace of open source technology – a direction I advocated for last year in Wired magazine.
Three (ISA) Kingdoms: x86, Arm, RISC-V
Intel’s support of RISC-V does not guarantee its success. As I’ve written before, RISC-V has a long way to go to build a vibrant software developer ecosystem, especially in the compiler layer (the gold standard of such a developer ecosystem is Nvidia’s CUDA).
But with Arm decidedly weaker, now that it has to survive and thrive on its own without Nvidia’s ownership, the stage is set for a “three kingdoms” deadlock of ISAs: x86, Arm, RISC-V. Each architecture's ecosystem has intertwining interests with the other two, with both reasons to cooperate and reasons to compete.
In the ancient China version, the “three kingdoms” deadlock lasted about 35 years (229 AD - 263 AD) before one conquered the other two, eventually resulting in a “winner take all” ending. Of course, that was a war over power, territory, and dynastic legitimacy. There are no antitrust regulators in war.
I don’t see a “winner take all” ending happening among the three ISAs, whether it’s 35 years or 350 years. Romance of the Three Kingdoms: Semiconductor Edition is here to stay.
快进到2020年9月，当Nvidia宣布其收购Arm的计划时，交易主要成分是Nvidia的股票，不是现金，这正是孙正义所希望的。用他自己的话说，孙正义把这一收购描述为 "一种购买的销售，一种销售的购买"。如果Nvidia-Arm的 "婚姻" 被批准了，软银将成为Nvidia的最大股东，孙正义策划至少四年的算盘就会如愿实现。当然，这一切在上周都黄了。
同样更值得注意的是，被埋没在所有Nvidia-Arm头条下的英特尔靠近RISC-V的举动。RISC-V是一款开源的半导体ISA（指令集架构），旨在有朝一日替代Arm。随着Nvidia和Arm的“分手”，以及英特尔“追”RISC-V的开始，未来会有很多新的技术、新的联盟、新的冲突和新的战斗 —— 好比《三国演义》，只不过是在三种不同的半导体ISA之间，而不是曹操、孙权和刘备。
Arm对英特尔是一个巨大的威胁 —— 在笔记本CPU（见苹果的M1芯片）和云计算服务器（见AWS的Graviton处理器）方面的市场份额都在增长。笔记本CPU和云计算服务器，由于对功耗没有智能手机那么敏感，传统上一直由基于X86的CPU主导。现在不是了。
比10亿美元更有价值的是英特尔对RISC-V的公开承诺 —— 给一个仍然年轻的项目的用户足够信心，作为RISC-V背后坚实的后盾。英特尔成为RISC-V国际基金会最高级成员资格也有两个含义：一个是技术的，一个是地缘关系的。
技术上的影响来自于英特尔通过其芯片代工服务兜售的一个新概念，"开放芯片"。通俗地说，"开放芯片，open chiplet" 是 "system on a chip"（SoC）的替代版。SoC是一种高度集成的方式，用一种架构将所有（或大部分）计算机处理器（CPU或GPU）打包到一个芯片上。它是目前的工业标准。"开放芯片" 想达到的是模块化，允许半导体产品同时可以利用多种架构，无论是x86、Arm还是RISC-V。这是一种经典的“开源”玩法，拥抱模块化和兼容性。英特尔不再厮守x86了，而是通过成为一个开放的芯片设计IP平台来保持竞争力和领先性，只要其代工厂能拿到成产这些芯片的单子。
地缘关系的影响只要在英特尔加入的前后比较一下RISC-V国际基金会的会员网页就清楚了。在英特尔的标志显示在 "高级会员" 部分之前，页面主要是被中国科技大厂和台湾公司如Andes（唯一上市的RISC-V公司，现在是英特尔的合作伙伴）所占据。美国的半导体公司则不见踪影。英特尔的加入最终改变了这种 "实力平衡"，并可能标志着美国工业界对开源技术的日益拥护 – 这也是我去年在《Wired》杂志上倡导的方向。
但是，由于Arm的明显消弱 —— 毕竟失去了Nvidia这个大靠山，必须独立生存和发展下去 —— “三足鼎立”的大局已定：x86、Arm、RISC-V。每种架构的生态都与其他两种利益交织，既有合作的理由，也有竞争的理由。
在《三国演义》里，"三足鼎立" 的僵局持续了大约35年（公元229年至公元263年），最终是西晋统一， "赢家通吃"。当然，那是争夺权力、领土和皇位的战争。战争中没有反垄断监管机构。
我不认为 "赢家通吃" 的结局会发生在 “ISA三国” 之间，无论是35年还是350年。《三国演义》的半导体版将会长期持续下去。