Hacker News Reader

Comment by ckastner | Hacker News Reader

Softbank paid $32B for ARM in 2016.

A 25% gain over a horizon of four years is not bad for your average investment -- but this isn't an average investment.

First, compared to the SP500, this underperforms over the same horizon (even compared to end of 2019 rather than the inflated prices right now).

Second, ARM's sector (semiconductors) has performed far, far better in that time. The PHOX (Philadelphia Semiconductor Index) doubled in the same time period.

And looking at AMD and NVIDIA, it feels as if ARM would have been in a position to benefit from the surrounding euphoria.

On the other hand, unless I'm misremembering, ARM back then was already considered massively overvalued precisely because it was such a prime takeover target, so perhaps its the $32B that are throwing me off here.

View on HN

It vastly overperforms the Softbank ventures we usually hear about (excluding BABA).

To be honest, cash deposited in a boring current account outperforms Softbank's Vision Fund

There's also a fundamental threat to ARM in the raise of RISC-V.

RISC-V is only a tiiny player in the low end embedded space (there are only a few parts actually available with that instruction set) and no competition at all at the high end.

Maybe in a decade or so it may be more relevant but not today. Calling this a "fundamental threat" is a tad exaggerated.

Where are people getting the "decade" number from? Looks demand for RISC-V just went up significantly and it's not like it's broken or not-existing, RISC-V "just worked" a year ago already: https://news.ycombinator.com/item?id=19118642

The "decade" comes from the fact that there is currently no RISC-V silicon with performance anywhere near the current ARM Cortex A series.

And a realistic estimate how long it will take to develop something that would be on par with today's Snapdragon, Exynos or Apple's chips is at least those 10 years. You need quite a bit more to have a high performance processor than just the instruction set.

The "just worked" chips are microcontrollers, something you may want to put in your toaster or fridge but not a SoC at the level of e.g. Snapdragon 835 (which is an old design, at that).

Also the Sipeed chips are mostly just unoptimized reference designs, they have a fairly poor performance.

Most people who talk and hype RISC-V don't realize this.

A ground-up new architecture takes 4-5 years.

Alibaba recently said their XT910 was slightly faster than the A73. Since the first actual A73 launched in Q4 2016, that would imply they are at most 4 years behind.

SiFive's U8 design from last year claimed to have the same performance as A72 with 50% greater performance per watt and using half the die area. Consider how popular the Raspberry PI is with it's A72 cores. With those RISCV cores, they could drastically increase the cache size and even potentially add more PCIe lanes within the same die size and power limits.

Finding out new things takes much more time than re-implementing what is known to already work. As with other things, the 80/20 rule applies. ARM has caught up a few orders of magnitude in the past decade. RISCV can easily do the same and give the lack of royalties. Meanwhile, the collaboration helps to share costs and discoveries which might mean progress will be even faster.

Hold on, the reason why RISV-V cores slower is that companies who make it doesn't have an existing backend or just got into CPU game.

I'm not saying Apple can drop-in RISC-V front-end to their silicon and call it a day, but you get the idea.

Sifive has a pretty decent chance at making performant chips within next few years.

But you don't buy a company for tens of billions ignoring anything that's not relevant today. You pay basing on what you predict the company will earn between today and liquidation. When Softbank originally bought ARM, no competition was on the radar. Now there is some. Hence, valuation drops.

Softbank bought ARM for $32 billion, sold for $40 billion, plus they were collecting profits during all that time. Not quite sure how that meshes with "valuation drops" ...

RISC-V in high performance computing is years out even if big players like Samsung or Qualcomm decided to switch today. New silicon takes time to develop.

And Nvidia really couldn't care less about the $1-$10 RISC-V microcontrollers that companies like SiPeed or Gigadevice are churning out today (and even there ARM micros are outselling these by several orders of magnitude).

Let me just quote what I originally responded to:

> A 25% gain over a horizon of four years is not bad for your average investment -- but this isn't an average investment.

> First, compared to the SP500, this underperforms over the same horizon (even compared to end of 2019 rather than the inflated prices right now).

> Second, ARM's sector (semiconductors) has performed far, far better in that time. The PHOX (Philadelphia Semiconductor Index) doubled in the same time period.

> And looking at AMD and NVIDIA, it feels as if ARM would have been in a position to benefit from the surrounding euphoria.

That's the "valuation drops" - relatively to the market, ARM has significantly underperformed, despite the business being actually healthy and on the rise.

Reply on news.ycombinator.com