Arm releases first in-house chip, with Meta as debut customer
cnbc.com108 points by goplayoutside 5 days ago
108 points by goplayoutside 5 days ago
Arm came from Acorn and Acorn did make the first ARM CPUs for their computers, so it's not really the first time they do this.
This is going to be a strategic challenge for ARM unless they are going to focus on chips that nobody else wants to make. And given the AI focus, that doesn’t seem to be the case. I would think that the RISC-V folks would be salivating at the prospect of flipping some existing ARM licensees to RISC-V.
Meta bought a RISC-V startup six months ago: https://hn.algolia.com/?q=meta+rivos
Guess at the end of the day, no-one ever got fired for building ARM.
Rivos is about making GPUs. It will be interesting to see how this all nets out.
It is going to be a huge 24 months for RISC-V. My biggest concern is that everybody will have already placed their bets before then.
focus on chips that nobody else wants to make
That's what happened here. Meta wants a Neoverse V3 CPU but no one will make it for them. So Arm has to make it.
ARM does not have their own fab, someone else is doing the actual making. ARM helped Meta design the thing.
That’s overly pedantic.
Then you’d say that Apple doesn’t make their laptops. Foxconn does.
The kind of work ARM would do to “make” a chip themselves goes beyond just design. It’s synthesis, P&R, test, packaging (generally a different company than the fab), yield management, inventory/logistics, etc.
The Acorn Archimedes came with Acorn branded CPUs (the "ARM250" IIRC) already in the late 80s. I can't recall what company made the chips for ARM at that time, but in the later Archimedes models it was VLSI.
VLSI from the very start according to https://en.wikipedia.org/wiki/ARM_architecture_family#ARM1
After Amazon, Google, and Apple all have had successes with in house ARM, I had naively assumed Meta would do the same. Given the speeds with which they have been developed, it must not be "that hard" to spin up a chip. You could have easily framed it as a long-term plan - custom chips for the Occlus.
That is not a dupe
I think it is. The launch announcement explicitly says the same thing, “Meta is our lead partner and customer”.
On HN, dupe means duplicate discussion for the same link
> duplicate discussion for the same link
I hate to be the bearer of bad news but this is incorrect per moderator dang at https://news.ycombinator.com/item?id=43765252:
> On HN, dupeness is more a question of whether the underlying story is substantively the same or not
I believe dang's most recent in-depth explanation can be found here: https://news.ycombinator.com/item?id=43738815 and you can search for more at https://hn.algolia.com?query=author%3Adang+dupe&sort=byDate&...
Submissions of the same exact URL are automatically merged into the previous discussion server side, and are discouraged for about a year.
But one is a press release from ARM and the other is a report from CNBC. How are the two the same?
By your logic, there shouldn't be a gazillion posts about Apple Events the day it happens.
Apple usually announces like 3-5 new products, each in a distinct market / audience fit. Arm announced one product for one customer.
But sometimes two long discussions ensue on separate days for one event/product/announcement, if it's big enough. Often the discussions are merged later on. No big deal.
And even for big news events (which, this might qualify as), people can miss the first discussion. Even if the discussions end up merged later on, the different discussions can still be fruitful.
Which is why, even if it is a duplicate conversation, the mods generally allow things to play out organically. There's either going to be more discussion above, or people have already said their peace and we move on.
When the two submissions aren't the exact same link, it becomes a subjective question as to whether they're similar enough to count as a dupe or not. They aren't automatically always a dupe just because the overall general topic is the same, but nor are they automatically considered not a dupe just because they're not identical.
In this case the consensus (that I agree with) certainly seems to be that they're similar enough to be considered a dupe. Though that doesn't force the moderators to have to treat it like a dupe and merge comments.
It's not my logic, it's the logic of the moderator(s) of HN. Here's more, cut+paste from the link previously provided (https://news.ycombinator.com/item?id=43738815):
> I agree—they're not all the same story. On the other hand: stories in an ongoing sequence usually lead to repetitive discussion, which is bad for HN
> stories in an ongoing sequence usually lead to repetitive discussion, which is bad for HN
You mean like the daily repetitive AI news?
How does this fit with Meta's decision to acquire Rivos?
I think they are still working on silicon. But the two solutions do not really compete.
Makes me wonder if Mark Zuckerberg had not had this weird vision of making Second Life VR for Meta and focused on AI as it was looming if they could have built a serious competitor to Anthropic and OpenAI. I know he tried, but it was already late to the party, but still, had he tried a lot sooner, would he have gotten more built? I think his obsession with making the VR stuff happen is holding him back.
Is being a serious competitor to OpenAI a good business proposition? OpenAI burns through insane amounts of cash and it seems pretty likely that it will ultimately just be replaced by cheaper Chinese models/inference.
The real product is the agent harnesses, which to be fair can be trained specifically to work with an in-house harness, but not sure it's necessary to own the models, especially if Chinese companies are licensing theirs for fine-tuning like we see with Cursor.
The key early signals from OpenAI FB missed were when OpenAI Five was featured DOTA 2's The International in 2018 and in April 2019 when it beat the pro team OG.
I think the people who care about that game and used the types media channels FB monitored was too small to show up on the company's radar.
But I'm not convinced Zuck was truly all in on VR. I thought the switch to Meta was a hasty attempt to rebrand under fire of a whistleblower / document leak cycle.
Despite all the money spent on VR labs, I always thought the pivot was much more of a Philip-Morris -> Altria thing than Dunkin' Donuts -> Dunkin'.
Zuck is all-in on building their own platform and becoming a gatekeeper like Apple, Google and Microsoft, rather than operating within their ecosystems.
VR was one dream to achieve that, it’s more difficult to envision that with AI which is why I suspect he was more reluctant to embrace it. AI models are more likely to become a commodity and running inside one of the platforms of the gatekeepers, which is not what Zuck wants.
That’s fine but this dream was previously pursued with mobile hardware.
FWIW, oculus represented a true competitive advantage. And I don’t think it was a coincidence that the Supernatural acquisition was temporarily blocked by the feds.
I think Apple was genuinely worried that fitness was (or still is) a killer use case and threat to Vision Pro.
The delay was enough to make FB lose focus / pivot again into AI.
Meta will always need the next platform. Instagram, Facebook phone, Whatsapp, Reels, Marketplace, Portal, VR, AI... Some succed, some fail. When you are an Ads company, the surface where ads are delivered to is important. I don't think it's a fundamentally different business model from Google's. VR addresses an interesting and increasing niche of people that refuse human contact and prefer the online world - for those people clothes, going out, buying a car, spending money on whatever the current society spends money on might not be so interesting, instead a parallel, virtual world, might be.
As for Claude and OpenAI, no one has a revenue net positive business model yet. They are much better than Meta's Llamas, but the model quality doesn't equal cash in the bank. Things might still change in the end. More players are better for me as a consumer.
Facebook's strength has never been innovation, but adapting to the changes; mostly through acquisition.
With the 20/20 hindsight - I'd say the VR bet was too early for Facebook. Instead of trying to build a future tech, they should have acquired it another few years later, only after the tech has reached a more mature stage.
Meta still has a chance to catch up in the AI race given they are not trying to build afresh, but once again adapt by throwing cash at it (which has been the biggest strength of Facebook and Zuckerberg. see: instagram, whatsapp, reels, and many more...)
I think they were just early with VR in general (they did buy the best VR at the time). And then severely miscalculated what VR would actually be great for.
Eventually we'll get super cheap and light headsets or glasses and gaming will be pretty cool. They should have focused all in on that. It's already a huge industry
Meta was on a roll with their early research and work with Llama and other models. Then it all just seemed to trail off. I’m sure they’re doing some interesting things internally but it does feel like they were in the right place to have capitalized on it a lot more.
meta had the gpus to train llama because of their capital spend on horizons, so at least they were in the game, but maybe he didn't see "chatbot" as a trillion dollar product category
This may not seem like an appropriate forum to say this, but this is a relevant and serious issue to neglect. With the current political climate and this relentless push for hundreds of these massive datacenters, everyone seems to have completely forgotten about the carbon emissions and climate change. These datacenters are such massive resource hogs that living near them is unviable due to their economic impact and overconsumption. Their impact on global climate, economy and even technology (talking about the RAM crunch) is much worse. But nobody seems to be keeping tabs anymore.
One thing to remember is that the climate catastrophe is not a single cataclysmic event like falling off a cliff. It's more like a lanslide that starts small and then gradually accretes into a massive disaster that's barreling towards you. And we're in it already. We're already paying a price in terms of human lives and the planet's biomass as a whole due to natural disasters that are becoming more frequent. We don't notice it because the increase is gradual.
And all that for what? Writing reports, reading emails, generating endless slop and waging wars? I'm not against AI or any other technology. But this cost doesn't seem justified considering their contributions to serious endeavors like medical research and habitat loss. This is ironic because we were talking two decades ago about ditching interpreted languages in favor of compiled languages for servers/services, in order to improve their carbon footprint. It looks like a joke today considering what these AI datacenters and crypto farms do. But we really can't really afford to forget it now. Remember that when you pay for AI with your money, someone else pays for it with their blood.
We “forgot” about it because climate-as-energy-policy was a lever for control because energy touches everything. Now that AI will touch everything, AI is the focus of those who seek control over others.
There has never been an energy transition that wasn’t driven by economics. From wood to whale oil to coal to petroleum. If you care about climate and not control over others, that is the place to focus.
Everyone else on the planet is doing just fine our policymakers are just pretending like climate change isn't real. We could be investing in and subsidizing proven methods like mass transit, renewable energy, &c. Nobody has to be coerced for this to happen. Unless like getting less subsidies because fossil fuels need to get their externalities priced in is coercion lmao
"in-house" is misleading
> Like nearly all fabless AI chipmakers, Arm currently manufactures its CPU at Taiwan Semiconductor Manufacturing Company ’s fabrication plants.
It's not misleading for people in the industry. ARM so far was selling IP (Verilog source code) that other chip makers would include in a full chip design.
Now ARM for the first time (this century) is making its own chip [design], which like most of its customers, is manufactured by a fab like TSMC.
The title is clear.
Not really verilog source code. TSMC probably provides libraries to Apple
Apple is just a systems integrator
Meta only recently announced a “long term” partnership with Nvidia:
https://about.fb.com/news/2026/02/meta-nvidia-announce-long-...
So how does this fit in? Is it a replacement for Nvidia’s portfolio of chips? Or just an alternative option to avoid dependency on one vendor? Something else?
CPUs and GPUs sit in pretty distinct niches; they don't substitute like you're implying.