I’m not so sure it’s negligible. My anecdotal experience is that since Apple Silicon chips were found to be “ok” enough to run inference with MLX, more non-technical people in my circle have asked me how they can run LLMs on their macs.
Surely a smaller market than gamers or datacenters for sure.
It's annoying I do LLMs for work and have a bit of an interest in them and doing stuff with GANS etc.
I have a bit of an interest in games too.
If I could get one platform for both, I could justify 2k maybe a bit more.
I can't justify that for just one half: running games on Mac, right now via Linux: no thanks.
And on the PC side, nvidia consumer cards only go to 24gb which is a bit limiting for LLMs, while being very expensive - I only play games every few months.
The new $2k card from Nvidia will be 32GB but your point stands. AMD is planning a unified chiplet based GPU architecture (AI/data center/workstation/gaming) called UDNA, which might alleviate some of these issues. It's been delayed and delayed though - hence the lackluster GPU offerings from team Red this cycle - so I haven't been getting my hopes up.
Maybe (LP)CAMM2 memory will make model usage just cheap enough that I can have a hosting server for it and do my usual midrange gaming GPU thing before then.
I mean negligible to their bottom line. There may be tons of units bought or not, but the margin on a single datacenter system would buy tens of these.
It’s purely an ecosystem play imho. It benefits the kind of people who will go on to make potentially cool things and will stay loyal.
> It’s purely an ecosystem play imho. It benefits the kind of people who will go on to make potentially cool things and will stay loyal.
It will be massive for research labs. Most academics have to jump through a lot of hoops to get to play with not just CUDA, but also GPUDirect/RDMA/Infiniband etc. If you get older/donated hardware, you may have a large cluster but not newer features.
They have, because until now Apple Silicon was the only practical way for many to work with larger models at home because they can be configured with 64-192GB of unified memory. Even the laptops can be configured with up to 128GB of unified memory.
Performance is not amazing (roughly 4060 level, I think?) but in many ways it was the only game in town unless you were willing and able to build a multi-3090/4090 rig.
I'm currently wondering how likely it is I'll get into deeper LLM usage, and therefore how much Apple Silicon I need (because I'm addicted to macOS). So I'm some way closer to your steel man than you'd expect. But I'm probably a niche within a niche.
Doubt it, a year ago useful local LLMs on a Mac (via something like ollama) was barely taking off.
If what you say it's true you were among the first 100 people on the planet who were doing this; which btw, further supports my argument on how extremely rare is that use case for Mac users.
People were running llama.cpp on Mac laptops in March 2023 and Llama2 was released in July 2023. People were buying Macs to run LLMs months before M3 machines became available in November 2023.
I keep thinking about stocks that have 100xd, and most seemed like obscure names to me as a layman. But man, Nvidia was a household name to anyone that ever played any game. And still so many of us never bothered buying the stock
Incredible fumble for me personally as an investor
Unless you predicted AI and Crypto then it was just really good, not 100x. It 20x from 2005-2020 but ~500x from 2005-2025
And if you truly did predict that Nvidia would own those markets and those markets would be massive, you could have also bought Amazon, Google or heck even Bitcoin. Anything you touched in tech really would have made you a millionaire really.
Survivors bias though. It's hard to name all the companies that failed in the dot com bust, but even among the ones that made it through, because they're not around any more, they're harder to remember than the winners. But MCI, Palm, RIM, Nortel, Compaq, Pets.com, Webvan all failed and went to zero. There's an uncountable number of ICOs and NFTs that ended up nowhere. SVB isn't exactly an tech stock but they were strongly connected to it and they failed.
It is interesting to think about crypto as a stairstep that Nvidia used to get to its current position in AI. It wasn't games > ai, but games > crypto > ai.
Nvidia joined S&P500 in 2001 so if you've been doing passive index fund investing, you probably got a little bit of it in your funds. So there was some upside to it.
There's a titanic market with people wanting some uncensored local LLM/image/video generation model. This market extremely overlaps with gamers today, but will grow exponentially every year.
How big is that market you claim? Local LLM image generation already exists out off the box on latest Samsung flagship phones and it's mostly a Gimmick that gets old pretty quickly. Hardly comparable to gaming in terms of market size and profitablity.
Plus, YouTube and the Google images is already full of AI generated slop and people are already tired of it. "AI fatigue" amongst majority of general consumers is a documented thing. Gaming fatigues is not.
It is. You may know it as the "I prefer to play board games (and feel smugly superior about it) because they're ${more social, require imagination, $whatever}" crowd.
"The global gaming market size was valued at approximately USD 221.24 billion in 2024. It is forecasted to reach USD 424.23 billion by 2033, growing at a CAGR of around 6.50% during the forecast period (2025-2033)"
Farmville style games underwent similar explosive estimates of growth, up until they collapsed.
Much of the growth in gaming of late has come from exploitive dark patterns, and those dark patterns eventually stop working because users become immune to them.
>Farmville style games underwent similar explosive estimates of growth, up until they collapsed.
They did not collapse, they moved to smartphones. The "free"-to-play gacha portion of the gaming market is so successful it is most of the market. "Live service" games are literally traditional game makers trying to grab a tiny slice of that market, because it's infinitely more profitable than making actual games.
>those dark patterns eventually stop working because users become immune to them.
Really? Slot machines have been around for generations and have not become any less effective. Gambling of all forms has relied on the exact same physiological response for millennia. None of this is going away without legislation.
> Slot machines have been around for generations and have not become any less effective.
Slot machines are not a growth market. The majority of people wised to them literal generations ago, although enough people remain susceptible to maintain a handful of city economies.
> They did not collapse, they moved to smartphones
Agreed, but the dark patterns being used are different. The previous dark patterns became ineffective. The level of sophistication of psychological trickery in modern f2p games is far beyond anything Farmville ever attempted.
The rise of live service games also does not bode well for infinite growth in the industry as there's only so many hours to go around each day for playing games and even the evilest of player manipulation techniques can only squeeze so much blood from a stone.
The industry is already seeing the failure of new live service games to launch, possibly analogous to what happened in the MMO market when there was a rush of releases after WoW. With the exception of addicts, most people can only spend so many hours a day playing games.
I think he implied AI generated porn. Perhaps also other kind of images that are at odds with morality and/or the law. I'm not sure but probably Samsung phones don't let you do that.
I'm sure a lot of people see "uncensored" and think "porn" but there's a lot of stuff that e.g. Dall-E won't let you do.
Suppose you're a content creator and you need an image of a real person or something copyrighted like a lot of sports logos for your latest YouTube video's thumbnail. That kind of thing.
I'm not getting into how good or bad that is; I'm just saying I think it's a pretty common use case.
AI porn is currently cringe, just like Eliza for conversations was cringe.
The cutting edge will advance, and convincing bespoke porn of people's crushes/coworkers/bosses/enemies/toddlers will become a thing. With all the mayhem that results.
It will always be cringe due to how so-called "AI" works. Since it's fundamentally just log-likelihood optimization under the hood, it will always be a statistically most average image. Which means it will always have that characteristic "plastic" and overdone look.
The current state of the art in AI image generation was unimaginable a few years back. The idea that it'll stay as-is for the next century seems... silly.
If you're talking about some sort of non-existent sci-fi future "AI" that isn't just log-likelihood optimization, then most likely such a fantastical thing wouldn't be using NVidia's GPU with CUDA.
This hardware is only good for current-generation "AI".
I think there are a lot of non-porn uses. I see a lot of YouTube thumbnails that seem AI generated, but feature copyrighted stuff.
(example: a thumbnail for a YT video about a video game, featuring AI-generated art based on that game. because copyright reasons, in my very limited experience Dall-E won't let you do that)
I agree that AI porn doesn't seem a real market driver. With 8 billion people on Earth I know it has its fans I guess, but people barely pay for porn in the first place so I reallllly dunno how many people are paying for AI porn either directly or indirectly.
It's unclear to me if AI generated video will ever really cross the "uncanny valley." Of course, people betting against AI have lost those bets again and again but I don't know.
> No. There's already too much porn on the internet, and AI porn is cringe and will get old very fast.
I needed an uncensored model in order to, guess what, make an AI draw my niece snowboarding down a waterfall. All the online services refuse on basis that the picture contains -- oh horrors -- a child.
Yeah, and there's that story about "private window" mode in browsers because you were shopping for birthday gifts that one time. You know what I mean though.
I really don't. Censored models are so censored they're practically useless for anything but landscapes. Half of them refuse to put humans in the pictures at all.
Sure, but those developers will create functionality that will require advanced GPUs and people will want that functionality. Eventually OS will expect it and it will became default everywhere. So, it is an important step that will push nvidia growing in the following years.
That’s not what I’m saying. I’m saying that the people buying this aren’t going to shift their bottom line in any kind of noticeable way. They’re already sold out of their money makers. This is just an entrenchment opportunity.
If this is gonna be widely used by ML engineers, in biopharma, etc and they land 1000$ margins at half a million sales that's half a billion in revenue, with potential to grow.
If they're already an "enthusiast, grad student, hacker", are they likely to choose the "plumbers and people that know how to build houses" career track?
True passion for one's career is rare, despite the clichéd platitudes ecouraging otherwise. That's something we should encourage and invest in regardless of the field.
Boring fact: The underlying theme of the movie Her is actually divorce and the destructive impact it has on people, the futuristic AI stuff is just for stuffing!
The overall theme of Her was human relationships. It was not about AI and not just about divorce in particular.The AI was just a plot device to include a bodyless person into the equation. Watch it again with this in mind and you will see what I mean.
The universal theme of Her was the set of harmonics that define what is something and the thresholds, boundaries, windows onto what is not thatthing but someotherthing, even if the thing perceived is a mirror, not just about human relationships in particular. The relationship was just a plot device to make a work of deep philosophy into a marketable romantic comedy.
OpenAI doesn’t make any profit. So either it dies or prices go up. Not to mention the privacy aspect of your own machine and the freedom of choice which models to run
Recent report says there are 1M paying customers. At ~30USD for 12 months this is ~3.6B of revenue which kinda matches their reported figures. So to break even at their ~5B costs assuming that they need no further major investment in infrastructure they only need to increase the paying subscriptions from 1M to 2M. Since there are ~250M people who engaged with OpenAI free tier service 2x projection doesn't sound too surreal.
If Silicon Valley could tell the difference between utopias and dystopias, we wouldn't have companies named Soylent or iRobot, and the recently announced Anduril/Palantir/OpenAI partnership to hasten the creation of either SkyNet or Big Brother wouldn't have happened at all.
That said, enthusiasts do help drive a lot of the improvements to the tech stack so if they start using this, it’ll entrench NVIDIA even more.