I'd say that as of now, ARM's only viable business model going forward will be as a patent troll. Allow me to expand:
There was a time when CPU manufacturers exerted a form of lock-in upon their customer base and software ecosystem via their ISA. This was especially the case 30 years ago when customers would buy a piece of software once and use it for a decade (or more). Backwards compatibility was essential and SAAS was not a thing.
But that changed over time with the growth of the internet, digital distribution, and most importantly, smart-phones with tightly controlled distribution models. Since then, the domination of ISAs like ARM and x86 has been driven solely by inertia. AAPL and QCOM paid ARM because they've always paid ARM. There was no reason to change.
Now ARM has given all the major players (including AAPL and NVDA, I would argue) a very important reason to change. And with RISC-V, there's really nothing standing in their way. There isn't a large legacy software ecosystem sitting out there that demands unending backwards compatibility. If today you have an app on the Apple App Store or Google Play Store, you are on the hook to keep it up to date and publish new versions at least once a year (if not more frequently), whether it's because of security updates, permissions changes, or straight-up platform-holder BS. If you have an app on the store, you are effectively committing to maintaining it forever (or at least as long as you want Apple and Google to continue to offer it for installation).
So there's no friction involved in having to re-compile to target a new architecture, because no matter what, Apple and Google were bound to do something that would cause you to need to publish a new build anyway eventually. And that's before we even get into things like LLVM and SIL and the fact that, at least Apple, can probably re-target all developer-submitted App Store binaries to a new microarchitecture without any 3P developer involvement at all!
So, why pay the ARM tax? Inertia.
But now that's gone, and I expect we'll see both Apple--who's always been careful to brand their chips as Apple Silicon, *not* ARM--and Google drop ARM within the next 5 years, probably for RISC V. China has already been on this path for geopolitical reasons. In the end, litigation is going to be the only revenue source left for ARM, as they devolve into patent trolls desperately trying to wring settlement money out of their former customer base.
That could be true what with all the hyperscalers beholden to ARM and ARM CPU cores supporting NVDA AI hardware. But then, Ian, being British, you would be biased 😛
Would any of this happen if Softbank had not bought ARM? If Masayoshi could not extract value by selling to NVDA then by Jove the new model is squeezing existing customers while newer upmarket ones come online. How much of this is payback for the NVDA deal being scuttled by opposition from QCOM amongst others?
Unless QCOM starts offering the Nuvia micro-architecture with a rapid change to the RISC-V ISA, the hyperscalers don't really have any quick options. Even then optimizing the lower-level software stack takes years. The biggest mistake QCOM made was abandoning their server efforts.
Which begs the question, why didn't Nuvia get bought for an actual server CPU effort? It would have been chump change for any of the hyperscalers. Heck even Oracle could have spent money to get that team into Ampere.
And if QC would have done that would have given ARM much less legal grounds to object to such a deal. One question not entirely unrelated: what status does Huawei's license from ARM have? It should be an ALA of some kind. However, Huawei did move its design (multi-threaded ARM core, no less) for server CPUs also into their Smartphone and Tablet SoCs. Might become interesting as the legal fireworks with Qualcomm go off.
So my take is that they will settle in the end and some money will be exchanged. QCOM survived the Chinese and Korean government anti-trust trials and AAPL. If anything they know litigation in technology licensing a bit too well instead of keeping around actual CPU design teams that they once had.
BTW I say this as a holder of QCOM having made the idiotic mistake of selling AVGO in 2015 thinking I owned too many RF chip names (also owned QRVO SWKS at the time). Ouch! Should have sold QCOM then instead. Coulda shoulda woulda …
But now I’m not selling QCOM. For once they have a great design team. This kind of brinkmanship has to end. Too much at stake for everyone. Unlike Chernobyl this one still has time to be unwound.
I don't know how you can possibly justify statements such as "RISC-V was in disarray, fragmented beyond recognition"
In the markets where ARMv8{.0, .1, .2, .3, .4, .5, .6, .7,.8,.9) and ARMv9{.0,.1,.2,.3,.4,.5,.6} are relevant there are currently exactly three RISC-V specifications:
- RVA20 (aka RV64GC) which Linux and others currently target as their baseline
- RVA22{+V}, for which the first low performance hardware (C908 & X60 cores, Canaan K230 and SpacemiT K1/M1 SoCs) has started to ship in the last 12 months with much faster (e.g. SG2380, SiFIve P670 cores) coming next year.
- RVA23, just ratified a few days ago, several cores (e.g. SiFive P870) are available for licensing now, chips in 2-3 years)
I don't see how three RISC-V standards, each a superset of the one before it, is somehow "fragmented beyond recognition" compared to seventeen ARMv8 and ARMv9 specs?
RISC-V has mix-and-match extensions in the EMBEDDED space, where a company chooses or develops a core with exactly the extensions they need, and builds their software with the appropriate compiler flags. This is not fragmentation but specialisation.
In applications processors there is no RISC-V fragmentation, outside of companies of course being able to implement custom extensions, just as Apple does in the Arm world. That affects no one but that company and its customers -- and the customers are free to ignore the custom instructions and use only standard ones. Indeed, often customers aren't even told about custom instructions, the vendor just quietly uses them in system software or standard libraries.
Even outside of embedded, there is a lot of ISA fragmentation. Look at Tenstorrent. Datacenter chip full of baby RISC-V cores. Each flavor has it's own ISA extensions.
Tenstorrent don't have any datacentre chip that I'm aware of. They have AI chips.
The Blackhole, for example, has 24 bog standard SiFive X280 cores implementing RVA20+V (512 bit VLEN) for running Linux, plus hundreds of "Tensix" cores which are effectively competing in a GPU / NPU kind of space with, yes, a customised ISA, but you're not running Linux on those -- they run what is effectively embedded code generated by Tenstorrent's AI toolchain.
The earlier Tenstorrent generations have only the Tensix cores and a host x86 PC runs the Linux software.
RSIC-V has been a free for all. Everyone adding their own ISA extensions while a toothless standards org refuses to make needed changes. The Tech Tech Potato video covers some of the pain points.
To be clear, I think competition is a good thing and want RISC-V to succeed. Clearly, someone at ARM agrees given the epic bone-headed recent decision.
RISC-V doesn't yet *have* marketshare outside of embedded for the very simple reason that it is too new. The original July 2019 ratified spec was fine for embedded things, and experimental Linux, but at least RVA22 and preferably the just ratified last week RVA23 is required to have market-relevance. The first couple of low performance RVA22 chips (Canaan K230, SpacemiT K1/M1) hit the market in the last 12 months, a number of much faster things that *will* be relevant to non-embedded are in the pipeline from a number of companies.
It would be appropriate to compare RVA22 to ARMv8.0-A published in 2011 -- or perhaps to ARMv8.2-A in 2017, which included optional SVE. RVA23 can be compared to ARMv9-A announced in March 2021.
Re SiFive, they are just one of many RISC-V Ip providers. Like many startups, especially US ones, they go through layoffs. As well as the one you refer to, SiFive laid off 80 people in February 2020 and a similar number in May 2020. I'm sure there have been others I'm not aware of. They tend to coincide with a change of direction. For example the 2020 ones (which I was a victim of) were as a result of SiFive finding that there were in fact not hordes of people who would love to build custom chips with customised microcontrollers (sets of standard and custom extensions) and that the main market for CPU IP (at elast for now) was established companies who already designed and got manufactured their own chips and simply wanted to switch ISAs for some of their cores. So SiFive divested OpenSilicon again and laid off people working on physical design, the SiFive Chip Designer automation, people working on automated toolchain support for customers custom instructions, and so forth. In retrospect, most of that market is probably well-enough served by "hard core in FPGA" products such as MicroChip's PolarFire SoC (which uses SiFive cores) -- or of course Zynq if you don't want RISC-V.
But none of this has anything to do with the comment I was responding to which alleged that "a toothless standards org refuses to make needed changes" which , in the absence of explaining what those changes would be is plain FUD.
Why do you say that ARM has a 95% chance at winning the case? Did you have previous posts on your substack highlighting the assignability of Nuvia IP to a different ALA that I missed?
I have some previous coverage of ARM but my opinion is based on the latest legal filings. All of the interesting bits are aggressively redacted by reading between the lines... it looks like ARM found some juicy evidence in the discovery process.
Except that Qualcomm's Oryon SoC won't be the only non-x86 Alternative in the non-Apple space for long. Mediatek and Nvidia have been busy working on exactly such an alternative for a while now (ARM designed CPU cores, Nvidia-designed GPU and AI/NPU, Mediatek designed overall SoC). Persistent strong rumors indicate that this collaboration also has Microsoft's strong blessing*. And, very likely ARM's as well, as Mediatek uses stock-ARM designed cores under their TLA. The already shipping Dimensity 9400 which is close in performance to the new Snapdragon Elite SoC for smartphones shows that stock ARM cores can be quite competitive with Oryon and Oryon 2. The new Dimensity 9400 is (IMHO) also a bit of a dress rehearsal for the CPU part of the upcoming Mediatek/Nvidia SoC for notebooks. It's certainly beefy enough, and would have a lot more thermal headroom in a notebook or 2-in-1 than in a Smartphone.
* Might also be that Microsoft is hedging it's bets here over the legal uncertainty surrounding Oryon. MS put significant money (time and effort) into getting Windows on ARM somewhat shipshape. Plus, Redmond is ruthless.
I think ARM did this because they know that Qualcomm's new Oryon v2 CPU is much better than their Cortex series CPU. So, everyone will now buy ARM chips from Qualcomm only, destroying their own chip business and leaving ARM to just survive on royalists with ALA license. Going forward, I think ARM will stop to give ALA to any company which wants to make ARM chips for selling it to other companies and will only give TLA or sell chips directly.
In addition to some irrational grievances, ARM might well have considered this:
- the vast majority of its customers outside the server space have TLAs, not ALAs. I don't know if Samsung gave theirs back or gave it up after Mongoose didn't work out for them, but otherwise, who was left besides Qualcomm. Huawei, maybe?
- While ALAs cost customers more upfront, ARM doesn't (AFAIK) get as much of a royalty fee on a per volume basis as they get for a TLA (stock ARM designs, with handholding etc).
- ARM really wants to push its royalty income and fee-for-service business, which is basically closed off to them if an ALA holder uses their (ALA holder's) custom designs.
- If they wouldn't have challenged Qualcomm on first buying Nuvia and then moving Nuvia's Phoenix -> Oryon design into end-user SoCs, what would have prevented (for example) Mediatek from (again for example) acquiring Ampere for their custom core designs for server CPUs and then adapting those for SoCs for smartphones, tablets and notebooks? All under Ampere's ALA, of course.
I believe this last point is one that ARM saw as a key threat for both their existing business model and even more so for their plans to increase their revenue.
One factor that I believe is also at work here is (pun intended) an irrational one. I suspect that ARM - both its major shareholder, and its employees- are likely still seething over the failure of their hoped-for acquisition by Nvidia. The loudest voice speaking out against that deal when it was considered for regulatory approval was Qualcomm. Now, imagine you'd be an ARM employee, especially in leadership positions. If the acquisition would have happened, those ARM employees and, of course, especially their then 100% owner - SoftBank - would have gotten some, or a bunch, or - in the case of SoftBank- a huge pile of Nvidia shares. And those shares have since more than tripled in value. So, I suspect there's lots of bad blood because of this also.
I think it is also strange since ARM lacks a good range of high end processors that are general available (so, not custom for cloud providers or Apple only). With the Oryon core that was a first step in that direction.
If Oryon is indeed that good, my advise for ARM would be to make a settlement that includes the possibility that ARM can sell Oryon cores.
I'm an amateur at analysis, but will write from memory. I vividly recalling that during the 2008 financial crisis, there was a very similar parallel (although I realize this comparison isn't exactly the same). First, the analogy gains clarity when you remember the Dodd–Frank Wall Street Reform, and the Volcker Rule.
Now, for this analogy to make any kind of sense, imagine ARM is a savings and loans bank. It charges chipmakers a relatively small fee (at least for smaller chipmakers) to borrow a design, which it can use to sell as part of a product. ARM itself doesn't make chip themselves (although they will soon)- they just ensure that the instruction sets are reliable- that is like cash. S&L banks can be relatively stable, as long as there is not inflation or too many bad loans- that's what caused the S&L crisis in the 1980s:
Now, look at Intel. Intel designs chips but also licenses them. In other words, they're primarily like Goldman Sachs (investments & banking), but in theory, could fluctuate like AIG, or Merrill Lynch, if they made really bad bets. In 2008, Lehman's Brothers went bankrupt, because they were the first and only major investment firm domino to fall, before the Federal Reserve did an 11th hour Sunday night 700 billion non-negotiable cash infusion to the top 5 banks. Bank of America bought Merrill Lynch because, they had no leverage, and, got saved from bankruptcy (with the government's help).
Now Intel, despite having lots of cash, has not only its x86 architecture to compete with ARM and RISC-V, but it also makes wildly speculative investments in deciding whether to buy EUV for 18NA and 14NA. Thus, if they spend $340 million on an High NA machine but TSMC underprices their chips, their investment essentially becomes like a highly speculative derivative hedge fund trade without liquidity. And so they could be short billions of dollars, which can't pay their suppliers and employees.
Now, as I indicated above, ARM is getting into the chipmaking business. There is and was no rule against investment banking on certain types of trades, but for large banks that are considered "systemically important financial institutions," or "SIFIs", the Volcker rule:
"was originally proposed by American economist and former United States Federal Reserve Chairman Paul Volcker in 2010 to restrict United States banks from making certain kinds of speculative investments that do not benefit their customers.[2]"
and was set in place in 2010 to limit important banks from having too many assets in subprime derivatives (CDOS which are still sold as exotic tranche products in recent years for certain high net worth investors). Because if a bank that affects too many people fails, then it becomes a national crisis:
ARM Ltd is owned by Softbank. Softbank's CEO was known for making wild investments such as WeWork. It's not impossible to think that the AI-chip design investment that ARM is making could be of a 2008-level bubble:
Unlike ARM, Intel at least has the government to bail itself out- I do not know enough about Japanese banking and company regulations as to know how easily they'd be allowed to fail if they didn't have Softbank backing them. The pretext for Intel is that it is the only DoD-authorized chipmaker for certain technology. I also remember the 1997 Asian Financial Crisis from the 2018 movie Default, and I am sure Softbank wants to avoid South Korea’s IMF-style bailout.
As for your latter point on Government intervention should SoftBank get in trouble, just one name: Rapidus. Massive push by the Japanese government and heavyweights from Japan Inc to jumpstart a EUV Fab in Japan. Basically by a company with no underlying physical assets that started out two years ago.
Page 77 of Qualcomm's counter claim after discovering ARM failed to destroy Nuvia property and shared it with competitors.
251. There are no provisions in either the NUVIA-ARM agreements or the QualcommARM agreements that: a. b. c. d. e. f. g. prohibited Qualcomm from purchasing NUVIA or acquiring NUVIA’s technology; required Qualcomm to obtain ARM’s consent to purchase NUVIA or access NUVIA’s technology; mandate that Qualcomm stop using any NUVIA technology it acquired; mandate that Qualcomm destroy NUVIA’s technology; prohibit the transfer or disclosure of NUVIA’s technology or confidential information to Qualcomm; limit the use of NUVIA technology only to NUVIA; or require Qualcomm to obtain ARM’s consent to further develop any in-process designs or technology that Qualcomm acquired from NUVIA
It shows that Arm clearly sees revenue beyond a Qualcomm-dominated balance sheet in the future.
I'd say that as of now, ARM's only viable business model going forward will be as a patent troll. Allow me to expand:
There was a time when CPU manufacturers exerted a form of lock-in upon their customer base and software ecosystem via their ISA. This was especially the case 30 years ago when customers would buy a piece of software once and use it for a decade (or more). Backwards compatibility was essential and SAAS was not a thing.
But that changed over time with the growth of the internet, digital distribution, and most importantly, smart-phones with tightly controlled distribution models. Since then, the domination of ISAs like ARM and x86 has been driven solely by inertia. AAPL and QCOM paid ARM because they've always paid ARM. There was no reason to change.
Now ARM has given all the major players (including AAPL and NVDA, I would argue) a very important reason to change. And with RISC-V, there's really nothing standing in their way. There isn't a large legacy software ecosystem sitting out there that demands unending backwards compatibility. If today you have an app on the Apple App Store or Google Play Store, you are on the hook to keep it up to date and publish new versions at least once a year (if not more frequently), whether it's because of security updates, permissions changes, or straight-up platform-holder BS. If you have an app on the store, you are effectively committing to maintaining it forever (or at least as long as you want Apple and Google to continue to offer it for installation).
So there's no friction involved in having to re-compile to target a new architecture, because no matter what, Apple and Google were bound to do something that would cause you to need to publish a new build anyway eventually. And that's before we even get into things like LLVM and SIL and the fact that, at least Apple, can probably re-target all developer-submitted App Store binaries to a new microarchitecture without any 3P developer involvement at all!
So, why pay the ARM tax? Inertia.
But now that's gone, and I expect we'll see both Apple--who's always been careful to brand their chips as Apple Silicon, *not* ARM--and Google drop ARM within the next 5 years, probably for RISC V. China has already been on this path for geopolitical reasons. In the end, litigation is going to be the only revenue source left for ARM, as they devolve into patent trolls desperately trying to wring settlement money out of their former customer base.
That could be true what with all the hyperscalers beholden to ARM and ARM CPU cores supporting NVDA AI hardware. But then, Ian, being British, you would be biased 😛
Would any of this happen if Softbank had not bought ARM? If Masayoshi could not extract value by selling to NVDA then by Jove the new model is squeezing existing customers while newer upmarket ones come online. How much of this is payback for the NVDA deal being scuttled by opposition from QCOM amongst others?
Unless QCOM starts offering the Nuvia micro-architecture with a rapid change to the RISC-V ISA, the hyperscalers don't really have any quick options. Even then optimizing the lower-level software stack takes years. The biggest mistake QCOM made was abandoning their server efforts.
Which begs the question, why didn't Nuvia get bought for an actual server CPU effort? It would have been chump change for any of the hyperscalers. Heck even Oracle could have spent money to get that team into Ampere.
Qualcomm could license the Nuvia cores as IP. Sell to Chinese hyperscalers.
And if QC would have done that would have given ARM much less legal grounds to object to such a deal. One question not entirely unrelated: what status does Huawei's license from ARM have? It should be an ALA of some kind. However, Huawei did move its design (multi-threaded ARM core, no less) for server CPUs also into their Smartphone and Tablet SoCs. Might become interesting as the legal fireworks with Qualcomm go off.
Great question! Huawei has an ALA though ARM China, a separate entity that SoftBank/ARM has a 49% minority stake. Huawei designs their own cores for server and smartphone. Old Reuters article that you may enjoy: https://www.reuters.com/markets/deals/arms-china-relationship-complicates-ipo-2023-08-22/
So my take is that they will settle in the end and some money will be exchanged. QCOM survived the Chinese and Korean government anti-trust trials and AAPL. If anything they know litigation in technology licensing a bit too well instead of keeping around actual CPU design teams that they once had.
BTW I say this as a holder of QCOM having made the idiotic mistake of selling AVGO in 2015 thinking I owned too many RF chip names (also owned QRVO SWKS at the time). Ouch! Should have sold QCOM then instead. Coulda shoulda woulda …
But now I’m not selling QCOM. For once they have a great design team. This kind of brinkmanship has to end. Too much at stake for everyone. Unlike Chernobyl this one still has time to be unwound.
Hey hey... Chernobyl did have time to be wound down. Liquidators and all those helicopters.
RISC-V will be bigger than ARM in 10 years. If ARM had a case they would just win it, make QCOM pay and move on.
RISC-V reminds me of Frankenstein's monster. Arm's 60 day notice was the lightning bolt that will bring it to life.
I sold my SOXX and SOXL when the ARM news broke. Thankfully I was near my computer. I also bought some $INTC.
I have never owned ARM because it traded at too high of a premium. I think ARM will be <70 with in a year.
ARM will likely keep AAPL as a customer for a long time but most everyone else will eventually move away from ARM.
I don't know how you can possibly justify statements such as "RISC-V was in disarray, fragmented beyond recognition"
In the markets where ARMv8{.0, .1, .2, .3, .4, .5, .6, .7,.8,.9) and ARMv9{.0,.1,.2,.3,.4,.5,.6} are relevant there are currently exactly three RISC-V specifications:
- RVA20 (aka RV64GC) which Linux and others currently target as their baseline
- RVA22{+V}, for which the first low performance hardware (C908 & X60 cores, Canaan K230 and SpacemiT K1/M1 SoCs) has started to ship in the last 12 months with much faster (e.g. SG2380, SiFIve P670 cores) coming next year.
- RVA23, just ratified a few days ago, several cores (e.g. SiFive P870) are available for licensing now, chips in 2-3 years)
I don't see how three RISC-V standards, each a superset of the one before it, is somehow "fragmented beyond recognition" compared to seventeen ARMv8 and ARMv9 specs?
RISC-V has mix-and-match extensions in the EMBEDDED space, where a company chooses or develops a core with exactly the extensions they need, and builds their software with the appropriate compiler flags. This is not fragmentation but specialisation.
In applications processors there is no RISC-V fragmentation, outside of companies of course being able to implement custom extensions, just as Apple does in the Arm world. That affects no one but that company and its customers -- and the customers are free to ignore the custom instructions and use only standard ones. Indeed, often customers aren't even told about custom instructions, the vendor just quietly uses them in system software or standard libraries.
Even outside of embedded, there is a lot of ISA fragmentation. Look at Tenstorrent. Datacenter chip full of baby RISC-V cores. Each flavor has it's own ISA extensions.
Tenstorrent don't have any datacentre chip that I'm aware of. They have AI chips.
The Blackhole, for example, has 24 bog standard SiFive X280 cores implementing RVA20+V (512 bit VLEN) for running Linux, plus hundreds of "Tensix" cores which are effectively competing in a GPU / NPU kind of space with, yes, a customised ISA, but you're not running Linux on those -- they run what is effectively embedded code generated by Tenstorrent's AI toolchain.
The earlier Tenstorrent generations have only the Tensix cores and a host x86 PC runs the Linux software.
RSIC-V has been a free for all. Everyone adding their own ISA extensions while a toothless standards org refuses to make needed changes. The Tech Tech Potato video covers some of the pain points.
I'm sorry but this is complete rubbish.
Could say the same about RISC-V market share outside of embedded. SiFive had massive layoff because the things are going well? https://www.youtube.com/watch?v=l0DUHZ1e48U
To be clear, I think competition is a good thing and want RISC-V to succeed. Clearly, someone at ARM agrees given the epic bone-headed recent decision.
RISC-V doesn't yet *have* marketshare outside of embedded for the very simple reason that it is too new. The original July 2019 ratified spec was fine for embedded things, and experimental Linux, but at least RVA22 and preferably the just ratified last week RVA23 is required to have market-relevance. The first couple of low performance RVA22 chips (Canaan K230, SpacemiT K1/M1) hit the market in the last 12 months, a number of much faster things that *will* be relevant to non-embedded are in the pipeline from a number of companies.
It would be appropriate to compare RVA22 to ARMv8.0-A published in 2011 -- or perhaps to ARMv8.2-A in 2017, which included optional SVE. RVA23 can be compared to ARMv9-A announced in March 2021.
Re SiFive, they are just one of many RISC-V Ip providers. Like many startups, especially US ones, they go through layoffs. As well as the one you refer to, SiFive laid off 80 people in February 2020 and a similar number in May 2020. I'm sure there have been others I'm not aware of. They tend to coincide with a change of direction. For example the 2020 ones (which I was a victim of) were as a result of SiFive finding that there were in fact not hordes of people who would love to build custom chips with customised microcontrollers (sets of standard and custom extensions) and that the main market for CPU IP (at elast for now) was established companies who already designed and got manufactured their own chips and simply wanted to switch ISAs for some of their cores. So SiFive divested OpenSilicon again and laid off people working on physical design, the SiFive Chip Designer automation, people working on automated toolchain support for customers custom instructions, and so forth. In retrospect, most of that market is probably well-enough served by "hard core in FPGA" products such as MicroChip's PolarFire SoC (which uses SiFive cores) -- or of course Zynq if you don't want RISC-V.
But none of this has anything to do with the comment I was responding to which alleged that "a toothless standards org refuses to make needed changes" which , in the absence of explaining what those changes would be is plain FUD.
Why do you say that ARM has a 95% chance at winning the case? Did you have previous posts on your substack highlighting the assignability of Nuvia IP to a different ALA that I missed?
Thanks!
I have some previous coverage of ARM but my opinion is based on the latest legal filings. All of the interesting bits are aggressively redacted by reading between the lines... it looks like ARM found some juicy evidence in the discovery process.
Care to elaborate?
Oddly ARM went down more than QCOM. Maybe the market is sharing your sentiment where it’s worse for ARM in the long term.
Absolutely wild that they’d do this. Crazy even. Qualcomm delivered THE ONLY viable alternative to Intel on the not-Apple laptop market.
This will hurt Qualcomm, but ultimately hurt ARM more. Qualcomm moving to RISC V, could start a landslide away from ARM and their licensing fees.
Except that Qualcomm's Oryon SoC won't be the only non-x86 Alternative in the non-Apple space for long. Mediatek and Nvidia have been busy working on exactly such an alternative for a while now (ARM designed CPU cores, Nvidia-designed GPU and AI/NPU, Mediatek designed overall SoC). Persistent strong rumors indicate that this collaboration also has Microsoft's strong blessing*. And, very likely ARM's as well, as Mediatek uses stock-ARM designed cores under their TLA. The already shipping Dimensity 9400 which is close in performance to the new Snapdragon Elite SoC for smartphones shows that stock ARM cores can be quite competitive with Oryon and Oryon 2. The new Dimensity 9400 is (IMHO) also a bit of a dress rehearsal for the CPU part of the upcoming Mediatek/Nvidia SoC for notebooks. It's certainly beefy enough, and would have a lot more thermal headroom in a notebook or 2-in-1 than in a Smartphone.
* Might also be that Microsoft is hedging it's bets here over the legal uncertainty surrounding Oryon. MS put significant money (time and effort) into getting Windows on ARM somewhat shipshape. Plus, Redmond is ruthless.
I think ARM did this because they know that Qualcomm's new Oryon v2 CPU is much better than their Cortex series CPU. So, everyone will now buy ARM chips from Qualcomm only, destroying their own chip business and leaving ARM to just survive on royalists with ALA license. Going forward, I think ARM will stop to give ALA to any company which wants to make ARM chips for selling it to other companies and will only give TLA or sell chips directly.
Of course the QCOM stock did not go down 60%, and I wonder how you get that Arm is going to win in court...
In addition to some irrational grievances, ARM might well have considered this:
- the vast majority of its customers outside the server space have TLAs, not ALAs. I don't know if Samsung gave theirs back or gave it up after Mongoose didn't work out for them, but otherwise, who was left besides Qualcomm. Huawei, maybe?
- While ALAs cost customers more upfront, ARM doesn't (AFAIK) get as much of a royalty fee on a per volume basis as they get for a TLA (stock ARM designs, with handholding etc).
- ARM really wants to push its royalty income and fee-for-service business, which is basically closed off to them if an ALA holder uses their (ALA holder's) custom designs.
- If they wouldn't have challenged Qualcomm on first buying Nuvia and then moving Nuvia's Phoenix -> Oryon design into end-user SoCs, what would have prevented (for example) Mediatek from (again for example) acquiring Ampere for their custom core designs for server CPUs and then adapting those for SoCs for smartphones, tablets and notebooks? All under Ampere's ALA, of course.
I believe this last point is one that ARM saw as a key threat for both their existing business model and even more so for their plans to increase their revenue.
One factor that I believe is also at work here is (pun intended) an irrational one. I suspect that ARM - both its major shareholder, and its employees- are likely still seething over the failure of their hoped-for acquisition by Nvidia. The loudest voice speaking out against that deal when it was considered for regulatory approval was Qualcomm. Now, imagine you'd be an ARM employee, especially in leadership positions. If the acquisition would have happened, those ARM employees and, of course, especially their then 100% owner - SoftBank - would have gotten some, or a bunch, or - in the case of SoftBank- a huge pile of Nvidia shares. And those shares have since more than tripled in value. So, I suspect there's lots of bad blood because of this also.
I think it is also strange since ARM lacks a good range of high end processors that are general available (so, not custom for cloud providers or Apple only). With the Oryon core that was a first step in that direction.
If Oryon is indeed that good, my advise for ARM would be to make a settlement that includes the possibility that ARM can sell Oryon cores.
I'm an amateur at analysis, but will write from memory. I vividly recalling that during the 2008 financial crisis, there was a very similar parallel (although I realize this comparison isn't exactly the same). First, the analogy gains clarity when you remember the Dodd–Frank Wall Street Reform, and the Volcker Rule.
https://en.wikipedia.org/wiki/Volcker_Rule
Now, for this analogy to make any kind of sense, imagine ARM is a savings and loans bank. It charges chipmakers a relatively small fee (at least for smaller chipmakers) to borrow a design, which it can use to sell as part of a product. ARM itself doesn't make chip themselves (although they will soon)- they just ensure that the instruction sets are reliable- that is like cash. S&L banks can be relatively stable, as long as there is not inflation or too many bad loans- that's what caused the S&L crisis in the 1980s:
https://en.wikipedia.org/wiki/Savings_and_loan_crisis
Now, look at Intel. Intel designs chips but also licenses them. In other words, they're primarily like Goldman Sachs (investments & banking), but in theory, could fluctuate like AIG, or Merrill Lynch, if they made really bad bets. In 2008, Lehman's Brothers went bankrupt, because they were the first and only major investment firm domino to fall, before the Federal Reserve did an 11th hour Sunday night 700 billion non-negotiable cash infusion to the top 5 banks. Bank of America bought Merrill Lynch because, they had no leverage, and, got saved from bankruptcy (with the government's help).
https://en.wikipedia.org/wiki/Lehman_Brothers
https://en.wikipedia.org/wiki/American_International_Group#2008_liquidity_crisis_and_government_bailout
Now Intel, despite having lots of cash, has not only its x86 architecture to compete with ARM and RISC-V, but it also makes wildly speculative investments in deciding whether to buy EUV for 18NA and 14NA. Thus, if they spend $340 million on an High NA machine but TSMC underprices their chips, their investment essentially becomes like a highly speculative derivative hedge fund trade without liquidity. And so they could be short billions of dollars, which can't pay their suppliers and employees.
They've become too big to fail. So the gov'ts CHIP act is like the 2008 Bailout of the 5 top banks- TARP(1). https://en.wikipedia.org/wiki/Troubled_Asset_Relief_Program
Now, as I indicated above, ARM is getting into the chipmaking business. There is and was no rule against investment banking on certain types of trades, but for large banks that are considered "systemically important financial institutions," or "SIFIs", the Volcker rule:
"was originally proposed by American economist and former United States Federal Reserve Chairman Paul Volcker in 2010 to restrict United States banks from making certain kinds of speculative investments that do not benefit their customers.[2]"
and was set in place in 2010 to limit important banks from having too many assets in subprime derivatives (CDOS which are still sold as exotic tranche products in recent years for certain high net worth investors). Because if a bank that affects too many people fails, then it becomes a national crisis:
https://www.wsj.com/business/crises-at-boeing-and-intel-are-a-national-emergency-093b6ee5
ARM Ltd is owned by Softbank. Softbank's CEO was known for making wild investments such as WeWork. It's not impossible to think that the AI-chip design investment that ARM is making could be of a 2008-level bubble:
https://www.reuters.com/technology/arm-holdings-plans-launch-ai-chips-2025-nikkei-reports-2024-05-11/
However, it's very possible that with enough AI chip manufacturing and chip license income, Softbank/ARM sees it can do without Qualcomm as a major client, as Ian Cutress believes https://www.cnbc.com/2024/07/04/softbank-shares-hit-record-high-after-24-years-on-arm-and-ai-boost.html
Unlike ARM, Intel at least has the government to bail itself out- I do not know enough about Japanese banking and company regulations as to know how easily they'd be allowed to fail if they didn't have Softbank backing them. The pretext for Intel is that it is the only DoD-authorized chipmaker for certain technology. I also remember the 1997 Asian Financial Crisis from the 2018 movie Default, and I am sure Softbank wants to avoid South Korea’s IMF-style bailout.
As for your latter point on Government intervention should SoftBank get in trouble, just one name: Rapidus. Massive push by the Japanese government and heavyweights from Japan Inc to jumpstart a EUV Fab in Japan. Basically by a company with no underlying physical assets that started out two years ago.
Page 77 of Qualcomm's counter claim after discovering ARM failed to destroy Nuvia property and shared it with competitors.
251. There are no provisions in either the NUVIA-ARM agreements or the QualcommARM agreements that: a. b. c. d. e. f. g. prohibited Qualcomm from purchasing NUVIA or acquiring NUVIA’s technology; required Qualcomm to obtain ARM’s consent to purchase NUVIA or access NUVIA’s technology; mandate that Qualcomm stop using any NUVIA technology it acquired; mandate that Qualcomm destroy NUVIA’s technology; prohibit the transfer or disclosure of NUVIA’s technology or confidential information to Qualcomm; limit the use of NUVIA technology only to NUVIA; or require Qualcomm to obtain ARM’s consent to further develop any in-process designs or technology that Qualcomm acquired from NUVIA