When ARM was bought by SoftBank the main argument for it being acceptable was that SoftBank were not chip manufacturers in their own right. And that because of this, ARM would have every incentive to carry on treating all of ARM's existing customers reasonably fairly under SoftBank's ownership.
That's obviously not the case with Nvidia.
So while, it's possible to imagine that Nvidia wouldn't (at least initially) abuse any ownership of ARM. From a societal pov, why risk allowing a situation that relies on humans behaving well on an ongoing basis not to fail? And two, given the way ARM operates i.e. all customer $$$'s are equal, why would Nvidia bother if they didn't think there was an advantage?
Probably a mixture of both, I think NVIDIA had a chance to push ARM to new heights it would also have forced them to be far more open.
Longer term the market might be in a better position with a public ARM.
I think a lot of the concerns around NVIDIA were driven by hyperbolic statements and memes.
ARM is a good example of a monumental success and relative failure as in their cores and their architecture is in everything but they never have been able to capitalize on their success at the same scale at least as far as it goes for converting it into revenue which sits only at around $2B per year.
NVIDIA on the other hand has mastered capitalizing on every success no matter how small it is.
Overall as acquisitions go on paper it was a good match. It would’ve given ARM access to a lot of capital, engineering resources and a driven management with a strategic vision.
NVIDIA would’ve gotten the ability to set the path for one of the most commonly used CPU architecture and being able to offer a completely vertically integrated solution to both enterprise and end consumers.
Arguably NVIDIA could achieve the latter on its own too at least by licensing ARM I do think it’s a shame that they don’t have an x86 license too. ARM on the other hand would need to undergo a massive change to get into the position they arguably deserve to be in.
> ARM is a good example of a monumental success and relative failure as in their cores and their architecture is in everything but they never have been able to capitalize on their success at the same scale at least as far as it goes for converting it into revenue which sits only at around $2B per year.
ARM got to where it is by not being too greedy. If others perceived them to be making power grab then a lot fewer people would have been willing to stake their own futures on the architecture.
To mix metaphors, ARM went with a 'rising tides floats all boats' approach to grow the pie in general instead of just their own slice of it.
ARM has serious issues with investment their operating margins under SoftBank dropped form 50% to below 10% they can barely afford their current R&D investment.
And the market now is more dependent on their reference designs and core IP than ever.
ARM as a company isn’t in a good position, they need a parent that can actual drive them forward or go public and get the investment they need and honestly deserve.
What exactly has happened to them under SoftBank that caused their margins to drop so much? I have no idea why they can't turn a decent profit, given that, as you said yourself, the market is more dependent on their reference designs and core IP than ever.
My guess is that it was because their revenue relatively remained flat whilst R&D expenses inflated as they always do with more and more advanced SoCs and processors.
If ARM was only about the instruction set and high level IP it wouldn’t be an issue but they develop full designs and those are the ones that get implemented by most users.
Considering the exponential growth in ARM processors and SoCs in mobile devices and IOT/IOE devices since SoftBank acquired ARM it’s really a mystery tbh on how the hell they mismanaged them so badly that they didn’t managed to capitalize on an exponentially growing TAM.
They kinda flatlined before that too but at least they had a steady growth rate in the years prior to the acquisition.
I think it's not a huge mystery, it's exactly what people are talking about higher in this thread or elsewhere in the discussion: the business world has gotten used to treating ARM as an at-cost utility, and there's significant business risk to ARM attempting to shift that model to a more revenue-focused one. Few quotes from around this thread:
> ARM got to where it is by not being too greedy. If others perceived them to be making power grab then a lot fewer people would have been willing to stake their own futures on the architecture.
> No, it shouldn’t. ARM is ARM precisely because it doesn’t charge a huge amount - if it did, it would not be as widely used.
ARM has managed to be successful by becoming a "public utility" that does their work essentially at-cost, which is then monetized by their customers. Everyone agrees that ARM generates a ton of value, but almost all of it is captured by their clients. And the problem is - the customers that were slippery enough to escape other ISAs to ARM are generally slippery enough to do it again if it becomes worthwhile.
You said elsewhere that there's other avenues to raising revenue without customers paying it and I'm curious what you mean, because that really seems to be the problem. I don't really see what you mean - for ARM to make more money, it has to be coming somewhere, and even if it wasn't up-front architectural/SIP licensing fees directly (even if it was, let's say, per-device licensing fees at time of manufacture/sale - like the HDMI Consortium does) those costs are still coming from somewhere and ultimately borne by consumers. It doesn't really matter to me whether I pay $1 a phone for incremental licensing or pay $1 a phone to amortize Qualcomm's license for ARM IPs, it's still the same money either way.
I think you have an oversized understanding of what ARM does, and what value it brings to companies using an ARM chip
It is not intel, it is making the chip or even really designing the chip, though they do have reference designs...
ARM is ARM exactly because companies can take the instruction set and design their own chips around it for their own needs, they are not holding out for the next AMD or intel design..
That however means the individual companies take on more of the design costs, and risk if the design is a failure.
No Arm should not be one of the largest companies in the world right now, not even close.
You would need to provide a source for both those claims
First off, Qualcomm just spent 1.4 Billion on a design firm, for the purposes to bolstering the internal design team [1]
Then there is a 99% claim, is that by sales, volume, etc? Has I know many companies that design their own ARM processor that make up more than 1% of the market, Apple, and Samsung alone would refute that statement
Qualcomm is using reference ARM cores these days so does Amazon in Graviton and Samsung in Exynos for example their latest and greatest SoC uses Cortex-X2, Cortex-A710 and Cortex-A510 cores with an AMD GPU, Apple is probably the only big player today with custom cores.
The small players also all tend to use Cortex cores.
You still need a design team to integrate ARM IP blocks with your own IP or within the constraints of a given manufacturing process but almost no one is making their own cores.
Qualcomm used to have custom core, and maybe they bought Nuvia for that after seeing that the X2 cores from ARM won’t be enough to compete with Apple. My own personal bet is that they’ll attempt to dabble with custom again and will eventually give up as it’s too expensive.
Apple managed to make it work because they are designing cores for their own use and they essentially poached an entire development team from Intel.
> You still need a design team to integrate ARM IP blocks with your own IP or within the constraints of a given manufacturing process but almost no one is making their own cores.
To emphasize this a little, I think the headcount needed for the this is quite a bit larger than the additional headcount need for making your own core.
There is a lot of working in making a high performance SoC using Cortex cores, and a lot of work in making a custom core, but I think some commenters here think that so many more custom cores are being made than in reality because they think that the rest is the easy part and the hard part is all cpu design, so if these fabless semiconductor companies are spending (overlapping) years per chip with thousands of employees it must be because everything is custom (regardless of what you can learn just looking up SoCs on wikipedia)
Yeah I would agree, there is ton of work that has to go into getting as much performance from the SoC given all the design constraints and designing your own cores would probably not help to remove enough of the problems you would need to solve to be worth while especially if you need to hit a very wide range of products and use cases.
Apple is in a unique position they both have a world class leading design team and they have complete control over the entire product so they have far more levers to tweak and they don’t need to compete with anyone but themselves.
And you can see that with how the went about with their SoC design. For the most part they had a single design with a few power envelopes for cheaper / less powerful products their solution was always to use SoCs from previous years.
Even for special cases like the Apple Watch etc they tended to repurpose cores from their existing designs. The S series SoC is essentially one or two efficiency cores from their A series further clocked down and sometimes on a more power efficient node to squeeze a bit more battery life out of them.
But beyond that until the M series it was pretty much always you get a new A series SoC for the new iPhone/iPad with the only major difference being the power envelope and everything else would use an SoC form 1-3 years earlier.
If Qualcomm could’ve play this game they might have still be using custom cores too.
more than 1% of what market? I think it is possible apple + Samsung produce less than 1% of all arm cores by volume. But sources for all of this would be nice, there is a lot of speculation in this thread, and most of it seems wildly uninformed (not a dig at you, thinking of elsewhere in the thread where a good portion of comments don't have a great grasp of what arm does).
Another factor to consider: I believe in-house designed core, on an SoC, ends up producing more ARM IP cores than apple/samsung/nvidia/whoever IP cores. 8 or 12 in house designed main cores may be supported by up to 20 cortexes as various controllers, boot processors, audio whatsits, and security widgets. I don't know if Apple made their own small space and/or low power designs for supporting processors, but that's not how other in-house arm core based SoCs I am familiar with worked.
> Probably a mixture of both, I think NVIDIA had a chance to push ARM to new heights it would also have forced them to be far more open.
No way. Even losing the most popular electronics brand (Apple) didn't sway NVIDIA from its course.
> ARM on the other hand would need to undergo a massive change to get into the position they arguably deserve to be in.
Intel has completely botched the last six-ish years, and Apple proved that ARM-based processors cannot just compete with Intel's offerings but outright destroy them, while still keeping backwards compatibility. All that ARM (as an architecture) needs now is Microsoft also offering ARM support and runtime emulation in Windows and a CPU vendor willing to sell decently powerful chips to vendors... and then, snap, Intel is gone.
> Apple proved that ARM-based processors cannot just compete with Intel's offerings but outright destroy them
Your average ARM processor is a Qualcomm Snapdragon or Amazon Graviton, it's not going to win any performance awards. Even the M1 loses out to most Desktop processors once you start talking about multi-threaded performance. It's a great laptop part, proof that a BIG.little architecture is a good idea, and it's massively energy efficient, but it's not 'destroying' Intel parts on raw performance.
ETA: We hear this same rhetoric every time AMD would come out with a part that was better than Intel (Athlon, Ryzen, etc). Intel isn't going anywhere, give them 4-5 years and they'll optimize and sell a part the eliminates the advantages. They've been doing exactly that for 30+ years.
> It's a great laptop part, proof that a BIG.little architecture is a good idea, and it's massively energy efficient, but it's not 'destroying' Intel parts on raw performance.
Raw performance does not matter for 99% of the market (which is PCs for corporate drones shifting data around in Word, Excel and a data warehouse application). Your average Snapdragon is performance-constrained on mobile anyway because of cooling and power usage concerns - put that flagship CPU in a laptop or a NUC-sized case, and you will get more than enough to satisfy said corporate drones. Especially those who have some KPI target for "corporate sustainability" - claiming to have halved your IT fleet's energy consumption will net your average VP/C-level exec quite the bonus.
All the market needs to do is provide the environment for that.
>All the market needs to do is provide the environment for that.
It's already been here and Dell/HP is still an X86 shop. Intel will survive, they've been doing this for 50+ years. With far, far fiercer competition in the past.
> It's already been here and Dell/HP is still an X86 shop
Yeah, because Dell and HP are enterprise vendors - and as long as there is no ARM Windows version that offers x86 backward compatibility, no enterprises (and frankly, most private customers) will shift to ARM.
"Windows on ARM can also run Win32 desktop appps[sic] compiled natively for ARM64 as well as your existing x86 Win32 apps unmodified, with good performance and a seamless user experience, just like any PC. These x86 Win32 apps don’t have to be recompiled for ARM and don’t even realize they are running on an ARM processor."
It doesn't work well for multithreaded applications because of the difference in memory model. It uses some heuristics to try and not issue memory barriers after each memory access, and sometimes gets it wrong at the expense of correctness.
sure, but the market wouldn't value them at $40b, is the thing. At least not without a plan to turn that $2b into a lot more sometime into the future.
if you're uber, or google, deliberately running in the red to grow the business, sure, maybe you deserve a higher valuation than your revenue alone justifies. That's not where ARM is, there's huge headwinds to them attempting to monetize more heavily, even if they wanted to crank revenue it's hard to see how they get there without chasing away their customers.
an underlying problem here is that SoftBank really overpaid for ARM (and then proceeded to mismanage it into the ground) and they want their money back. And sure, that's not the market's problem, but as you can see from NVIDIA or Qualcomm making plays here, there's companies for whom the asset is really worth $40b.
If you artificially restrict what ARM is allowed to do - that mindset in this thread of "the purpose of ARM is to be a public utility providing IP at-cost for their customers to monetize" combined with "they shouldn't be allowed to sell to any customer that might see any synergies from a merger" - then no, it's definitely not worth $40b. Because you're not going to be able to grow that $2b and you don't have any synergies that enhance your own business. But that's the market valuation that the asset could generate in the right hards.
It's kind of unfortunate for SoftBank that they've sort of sleepwalked into a de-facto nationalization of their asset. The synthesis of all of this is that they aren't going to be allowed to sell it at a market price, they are going to be forced to either continue operating it with minimal margins, or sell it to a consortium at a massive discount to the true value of the asset, so that someone else can continue operating it with minimal margins. Even an IPO would result in a significantly constrained valuation as a result of the legal strictures that are being applied - the asset very obviously cannot be sold freely at a market price, and assets are only worth what someone is willing to pay (or in this case, can get the government to approve paying/who the government will approve paying). So even if IPO'd, it will have to operate in the same fashion. Effectively, there is no scenario in which ARM isn't going to be operated as a de-facto public utility going forward.
De-facto I don't see a difference between "the government takes your asset and pays you 1/4th of market rate" and "you need to sell your asset, but the government won't allow you to sell it to anyone except one specific buyer who is only willing to pay 1/4th of market rate". The fact that the asset ends up in the hands of (a trust owned by) Google/MS/Amazon rather than operated directly in the government books isn't relevant to the owner, in fact it actually kind of makes the whole thing more odious. Google/MS/Amazon are now leveraging government power to reduce their supplier costs.
advancement for the sake of advancement is not something to celebrate. I can't imagine a situation where more consolidation in this space is something beneficial for anyone except the company doing the acquisition.
or
actually state protecting it?