Hacker News

Jensen Huang's fall hits 'open' models in boost to Anthropic and OpenAI

46 points by teleforce ago | 24 comments

stuaxo |next |previous [-]

Not sure on the logic here.

yumraj |root |parent |next [-]

Exactly. Don't Anthropic, OpenAI use Nvidia hardware also?

Also, Jensen leaving does not mean that Nvidia is suddenly gone tomorrow, sounds more like FUD.

jpardoe |root |parent [-]

Jensen was allegedly facilitating illegal sales to China and that won’t be happening any more.

bartleeanderson |root |parent [-]

China doesn't need to buy the hardware when it can rent it. Its already doing that from what I have read.

deltoidmaximus |root |parent |next |previous [-]

Open models are the main blocker against OpenAI/Anthropic's regulatory moat play. They can maybe get them banned in the US but good luck extending that elsewhere.

China produces the majority of open models.

Jensen will probably be replaced with some one who plays ball on the regulatory moat whereas he was more of a put eggs in multiple baskets kind of guy. Which means less eggs for China.

|root |parent |next [-]

Danox |root |parent |previous [-]

That makes no sense the only basket Jensen was concerned with was himself, China is a tough competitor because at this point in time, they seem to be willing to iterate and plan long-term, and they have been executing pretty well over the last 50 years.

Thank you, Nixon and Kissinger, in addition, thank you to all the other companies in the West that were happy to help them out for a short term profit.

I hope America can turn the tide around. I would love to see some of those big buildings that are being built across the country. Would be buildings designed for making things inside and I’m not talking about AI.

deltoidmaximus |root |parent |next [-]

If you are nvidia and worried about China being a competitor in the future you should keep selling to China as long as you can since it will reduce the amount of domestic funds available to any new competitors they create in the short term and delay their products becoming competitive AND you'll make more money in the short term. If you stop selling altogether then an otherwise uncompetitive domestic offering will get a funding infusion allowing them to iterate faster and stay alive longer. Necessity is the mother of invention and if Nvidia is around ChinaGPU isn't really that necessary.

If it is a national priority then in the long term they'll make something competitive anyway, like you said. All you control is the speed knob. And who knows? Things may change during that delay and China may no longer be able to afford the development and cancel it.

Now imagine you can only sell to OpenAI/Anthropic. Your other markets are withered away and you only have a couple big high margin customers. Now they have you by the balls! Think of how Apple were notorious for abusing ram suppliers for years because of how big of a customer they were. It'd be much better if you sold to multiple AI companies, as many as possible really. Then they have much less individual leverage.

jpardoe |root |parent |previous [-]

I’m no economist, but if America never started outsourcing to China surely that would leave them uncompetitive against other countries that did?

hollerith |root |parent [-]

Washington let China sell freely into the US as a bribe to pull Beijing away from Moscow during the Cold War. Mao was too dumb to take advantage of this bribe, but his successor Deng was not.

Many countries were bribed in this way during the Cold War. Also, many in the US leadersip class hoped free trade would prove a liberalizing and democratising influence on the world, and this hope persisted after the end of the Cold War. Clinton was probably the most influential champion of this hope.

American prosperity is not dependent on international trade. It has the most self-sufficient economy on the planet, the big exception to which was its former dependence on oil imports, which lasted from the 1950s to about 2015. China in contrast remains one of the least self-sufficient economies although it has become a little more self-sufficient over the past 2 decades (because of wise planning by Beijing).

Since ocean shipping is such a huge part of international trade and since the US Navy has dominated the world ocean since 1944 (i.e., can often dictate the terms and rules of ocean shipping) the promotion of international trade has been a big part of US foreign policy. But Washington does it to enhance US national security (and to advance the liberal ideological agenda), not because the US economy needs it.

jpardoe |root |parent [-]

Doesn’t the US economy largely rely on the strength of the dollar and being able to buy cheaply from elsewhere? The people arguing against the trade deficit seem to think this, and that it comes at the expense of American citizens and is causing the issues in the economy right now.

whynotmaybe |root |parent |next |previous [-]

Is this some coping mechanism to justify anthropic's IPO?

Danox |root |parent |previous [-]

It’s called wishful thinking, coping, desperation to find anything good out of this whole Nvidia fiasco. It sounds like the fear that the start of the AI bubble is upon us.

kylehotchkiss |next |previous [-]

China isn’t slowing down and they know a few more iterations of improvement in the <64gb memory space is going to make a dent for a lot of American companies.

Synthetic7346 |next |previous [-]

Is open synonymous with Chinese here?

SaucyWrong |root |parent [-]

"open" refers to any models that release their weights publicly so you can run them locally and fine tune them.

China happens to produce many of the open weight models, some are also produced in other countries.

OpenAI and Anthropic don't, and probably never will, produce an open weight model.

AppalachianEng |root |parent |next [-]

That's not true, there are gpt open weight models

jbergler |root |parent |previous [-]

aitchnyu |root |parent [-]

Is it a dead end?

dimator |next |previous [-]

Sorry but if open models' success rested on the fate of one person, then they were on weak footing to begin with. That is certainly not the case, though, and they will continue to improve to the point where people actually crunch the numbers (which is happening) and realize they're being taken for a ride by the big shots.

All of this frontier chasing just reminds me of gaming PC perf maxxers, bragging about their gfx cards. But in that space, let's not forget that entire empires were created on more commodity, every-man hardware.

The same thing is going to happen here with capable open models. DeepSeek flash can update your shitty bash script or occopy a place in your workflows just fine, you don't need to spend 15x.

ece |next |previous [-]

Companies selling hardware and those doing post training will always need open models. Open models have only gotten better even as hardware to run them has gotten more expensive, lots of inference providers rely on open models. It would be hard to argue their progress or use will stop, but the circular financing in this space has people confused.

villgax |next |previous [-]

says who?

2OEH8eoCRo0 |next |previous [-]

Why wouldn't the AI bubble popping be bullish for open models when it would mean affordable compute and memory for personal computers?

MetaverseClub |previous [-]

Who benefits is the criminal!