I don’t have hard evidence of this, which makes it a thought experiment and not an assertion.
Big-cap tech is utterly notorious for some of the worst anti-trust violations and other cartel-style behavior of any sector. This goes back as least as far as “Wintel” in the 90s and probably further that I didn’t watch up close. Suffice it to say that the Justice Department is extremely disincentivized to go after domestic economic Cinderella stories in a globally competitive world and has had to bring lawsuit after lawsuit on everything from bundling to threatening OEMs to flagrant wage fixing in print (I do have second-hand that the mass layoffs are coordinated aka Don’t Poach 2.0).
Crippling “gaming” (high margin but not famine-price gouging margin) cards, controlling supply tightly enough to prevent the market clearing at MSRP routinely, stepping at least up to and likely over the line on the GPLv2: these things or things like them have been ruled illegal before (though it’s hard to imagine that happening again).
It’s possible that tech is just a natural monopoly and we had it right with Ma Bell: innovation was high, customer prices were stable and within the means of most everyone, and investors got a solid low-beta return.
It’s easy to view the past with rose-colored glasses and the Bell Era wasn’t perfect, but IMHO this status quo is worse.
Market segmentation generally benefits consumers though. Enterprise is willing to pay much more than an average consumer, and they buy a lot more PCs/gpus/etc than the average consumer, so if you banned market segmentation tomorrow the response would be that gpus get a lot more expensive. You can see that happening when miners or enterprises buy up gaming gpus (which they are doing right now with 4090s for example). The clearing price wouldn't fall, it would rise.
The de-facto mechanism that's occurring with market segmentation is that consumers pay less than the clearing price and enterprises pay more. And consumers benefit from this subsidy.
If you banned it, you also wouldn't change the cost of developing new generations of products, so product development just would go slower, and the market would get even lumpier and less efficient (vendors squatting newer nodes would have an advantage, but it's inefficient to launch something immediately to compete with them, etc).
I've addressed this point elsewhere in this thread and in a thousand other comments but it bears repeating: there are either zero or one groups of people who understand economics mathematically depending on whether or not Medallion is a a Ponzi scheme that's lasted 35 years (I tend to think they invented attention models at the same time LeCun invented working convolutional networks). If someone makes more money than the indices consistently by "using math" on markets they are lucky as hell (in a manifold compactness no fucking way sense), have some grey edge in the mix, or work at Ren.
Price discrimination is generally prohibited in naked form: it often if not usually induces market failures. So companies spare no expense working around it (there's a reason why ITA software people made a mint approximating NP-hard problems acceptably well and rather cheaply to compute airplane ticket prices).
"Market segmentation generally benefits consumers" is exactly the kind of Wien's Law first-order approximation that sounds compelling but doesn't fit the data.
It does not benefit consumers that a 4090 has the same amount of GDDR6 in it as a 3090, which is incidentally 2.6Gb shy of `dolphin-8x7b-v1-q4_k_m` (or whatever the NVIDIA equivalent is, I use Apple gear now because inference is the game now), while the price of that RAM dropped. The fact that an H100 or whatever the next Hopper iteration is costs more than an M-series BMW on paper says nothing about what Meta paid for 150k of them or whatever it was, and it certainly doesn't help someone who'd like a decent chat bot without sending their data to eyeball-scanner database guy. When you see a leading edge x090 for MSRP or less at Best Buy, it's a good idea to stock up if you're in the market over the last 5 years (London has been running ETH2 for what, a year and change now?). That hardly seems a subsidy to the consumer paid for by those spendthrifts in FAANG.
NVIDIA pioneered serious "GPGPU" engineering. 5-10 years ago their microarch and fab-sourcing and software and the whole show were legitimately differentiated, no one had done it. And they made a pile, and that pile seems "conflict-free". But anyone can make a fused multiply-add unit these days, EE undergraduates do it in Verilog. Intel makes PCI express cards that cost $300 and have higher memory bandwidth in the <hand-wave>decoder-only language attention model</hand-wave> use case. And their drivers are documented and work on everything.
Now did Lisa Su decide to "concentrate on the supercomputing market with the MI300XYZ" and Jensen decided to "concentrate on AI with Hopper" independently to a degree where the market is perfectly partitioned? Who knows, I certainly don't have proof one way or the other. But if someone made a call being like "I'm thinking of focusing on X but don't really see our differentiation in Y. How's Cathy?", it wouldn't be the fucking first time.
But even that isn't critical to the point: this, the here and now, isn't good for consumers, or society, or our industry, and unlike e.g. Stallman's melodrama, it fucking matters this time.
I'm very flattered that you enjoyed my comment, it's been my observation that failed novelists are entertaining posters somewhere on the Internet :) "How's that novel coming Brian? Still working on that novel?"
Levity aside, unfortunately this stuff is as real as Tuesday and taxes, and as serious as a heart attack used to be and an off-color tweet is now. This is the "Star Trek: TNG" v. "Blade Runner 2024" moment: for the first time in history the endless cycle of tennis courts and guillotines and eventually a new crop of nepotism-fueled classism awaiting its turn at the guillotine might end up in a spontaneously broken symmetry.
There are like 5-10 unanimous markers of human achievement that machines can trivially reproduce or exceed, and everyone acknowledges Chess and Go and Atari and fan-fiction and ImageNet and ImageNet fan-fiction are in that box.
They're human in a really scary sense: there's an old saying that if you can only be good at one thing, be good at lying, because then you're good at everything. LLMs beat pick-up artists in bars and tech CEOs in all-hands meetings all to hell at vaguely riffing with a ruthless, amoral, and narrow goal. If I ask Dolphin to talk like a VC? It says "path-dependent" every third sentence like it's got a Substack and a Tesla and a very respectable 3-bedroom in Los Gatos or Atherton.
If this stuff was floating around 20 years ago? We'd throw out the donor class without a second thought on seeing that a MacBook can talk like any power-broker but better.
Today? The capture is so far along that we might end up handing the reigns of "alignment" over to "effective altruists", at which point the situation is pretty static.
To be honest this just sounds like a bunch of talking points stuck together, can you write a credible argument based on known case-law/precedent/etc... if you have a legal angle to critique?
Society isn’t shaped just by current laws, but also by where we want things to go. If the current situations is undesirable, changing laws and regulations is the way to change it. (Of course actually enforcing current laws and regulations can also go a long way in many cases…)
So what changes to the law do you propose? None of the little annoyances that Nvidia is occasionally creating amount to anything that could have resulted in their current dominance in AI processing.
They made a good product. The competition is responding too slowly. And they got lucky with this crazy (in a good way) AI boom. I don't see how this could possibly be outlawed or why you would even want to. It will resolve itself on its own given enough time.
How about right-to-repair. How about forcing better access, documented, attributable firmware. Open Drivers for Linux.
The government as a big costumers could demand these things, even without having a low. The military should demand these things to be open for various reasons.
I think Nvida can still make plenty of money in such a world.
In cases where an interface has absurdly high value for society if its a standard, the government could also 'buy' that and open it up. Just like they do with other infrastructure.
One could make the argument that the x86 interface should be public domain as it amounts to infrastructure that most of society builds on. How such a thing would exactly work is of course up for some debate. But the concept of the government 'liberating' common interfaces makes sense from a society perspective.
>How about right-to-repair. How about forcing better access, documented, attributable firmware. Open Drivers for Linux.
How would that have stopped Nvidia from dominating the AI market?
>The government as a big costumers could demand these things, even without having a low. The military should demand these things to be open for various reasons.
I agree that governments should use APIs with multiple competing implementations (or one truly open source implementation) where possible. This could make a difference in some cases, but I doubt it would have had a big impact in this particular case as demand for GPGPU is overwhelmingly coming from the private sector.
>In cases where an interface has absurdly high value for society if its a standard, the government could also 'buy' that and open it up. Just like they do with other infrastructure.
Agreed, but is there a legal reason why AMD and others are not allowed to create a clean room implementation of CUDA? Haven't they done exactly that with ZLUDA (which they have now defunded)?
I thought not supporting CUDA was more like a failed strategic move by competitors to prevent CUDA from becoming an industry standard.
I think we agree on all the relevant principles. I just don't see how any of these principles make a big difference in this particular case.
Also, I don't see the Nvidia situation as particularly problematic. They are not too entrenched to unseat. Some of their biggest customers (themselves huge oligopolists) are shaping up to be their biggest competitors.
Most of AI processing will be inference, not training. Bringing the costs down is absolutely key for broad AI use. My bet is that the hardware margins will end up being slim.
Sounds like it's just undesirable to you. (I don't own any Nvidia stock). Their growth to their current position in the market was organic. You don't just break down big companies for no reason other than them being big. If we did, there's a long line of companies ahead of Nvidia to be broken up first.
I don’t see what being organic has to do with market failures (if in fact there were no shenanigans, which seems a stretch to posit at zero).
A market failure is a market failure.
There’s a big lobby on HN who want to defend or minimize or justify leaving market failures be, which is weird given they’re really bad for most people on HN, it’s a free country.
I’m saying that we have “deregulated” big-cap tech to the point of not enforcing laws that we enforced even a decade ago. This is analogous to the last time a sector 100% captured the regulators which was high finance in the Greenspan era last time. Ended poorly.
The likely height of big-cap tech regulation was the regulated monopoly of ATT/Bell Labs/Western Digital all through the 20th century.
I prefer the outcomes in that era to the present status quo.
You can agree or disagree, and disagreeing is simple: “I prefer these outcomes because…”. I welcome such disagreement. And FWIW I’m not one of the many people who downvoted you: but they were justified in doing so because you moved a comparison of policies and their outcomes into a more abstract space of 1-bit generalizations, e.g. “talking points” vs “law/precedent”. I gave examples, it’s public record and trivially Googleable.
You didn’t fail to understand my point, you didn’t like it and took a cheap shot.
Le sigh. I said big-cap tech used to get in trouble with the DoJ a lot, but has gotten finance-good at lobbying and so they don't get slapped hard anymore. You said that I was stringing together talking points and not citing concrete examples.
And I never mentioned 'not citing concrete examples.', if your confused about the ask, I'll repost it: "can you write a credible argument based on known case-law/precedent/etc... "
i.e. Can you write down the actual argument that you propose to advance?
Big-cap tech is utterly notorious for some of the worst anti-trust violations and other cartel-style behavior of any sector. This goes back as least as far as “Wintel” in the 90s and probably further that I didn’t watch up close. Suffice it to say that the Justice Department is extremely disincentivized to go after domestic economic Cinderella stories in a globally competitive world and has had to bring lawsuit after lawsuit on everything from bundling to threatening OEMs to flagrant wage fixing in print (I do have second-hand that the mass layoffs are coordinated aka Don’t Poach 2.0).
Crippling “gaming” (high margin but not famine-price gouging margin) cards, controlling supply tightly enough to prevent the market clearing at MSRP routinely, stepping at least up to and likely over the line on the GPLv2: these things or things like them have been ruled illegal before (though it’s hard to imagine that happening again).
It’s possible that tech is just a natural monopoly and we had it right with Ma Bell: innovation was high, customer prices were stable and within the means of most everyone, and investors got a solid low-beta return.
It’s easy to view the past with rose-colored glasses and the Bell Era wasn’t perfect, but IMHO this status quo is worse.