Remix.run Logo
Remix clone Hacker News
new | show | ask | jobsGithub
▲ jmyeet 3 hours ago

YEPPP... and I'm kind of shocked at how many people can't do simple math.

Let's put it context. Google's annual revenue seems to be north of $400B. So if OpenAI suddenly had Google's revenue, it would still be insufficient to recover their investment.

and it's a ticking time bomb because $1T in servers, CPUs, GPUs and memory is going to be worth $200B in 5 years. You can say they can keep using what they've got. Sure. But they're also not going to stop spending on new hardware. And the competitor that comes along in 5 years and spends $1T doing the exact same thing is going to have a huge advantage.

OpenAI at this point reminds me very much of the Russ Henneman pre-money hype cycle.

▲hansmayer 3 hours ago | parent | next [-]

This should be the top comment. Also, I think its not that many people, including our Simon here, are not good at math. Its more like, some of them seem to be incentivised to not be cough, cough, "good at math". How else will the hype sell?

▲simonw 2 hours ago | parent | next [-]

I thought my post was pretty free of hype. I said that this new revenue "Maybe even enough to start covering their costs!"

▲WhrRTheBaboons 2 hours ago | parent | next [-]

that statement is pretty high on hype relative to the actual financials though

▲hansmayer 2 hours ago | parent | prev [-]

Well, your title certainly was not, in any case!

▲chipotle_coyote an hour ago | parent | next [-]

I mean, a company that loses money on every widget they sell might technically have found "product-market fit." :)

It seems quite possible to me that developer tooling is going to end up being the biggest win from LLMs because there is a product-market fit -- and also quite possible that OpenAI and/or Anthropic end up getting bought for pennies on the dollar because their burn rate is unsustainable. AI may end up being this generation's "dark fiber."

▲simonw 2 hours ago | parent | prev [-]

[dead]

▲Imustaskforhelp 2 hours ago | parent | prev [-]

At a certain point, I genuinely feel like the best way this hype is being sold is by making people genuinely believe in it.

and in that sense, if Anthropic and OpenAI are able to create the projection that they can-be profitable despite finances seeming bubbly at best, I think that what happens is that these companies spew so much amount of content that people like Simon get into it too.

There is a deeper problem of people falling into AI psychosis too, in general, I am not sure if Simon has fallen into it or not

I think that the greatest point which can be made here is to not offload your thinking to others and to think about the situation yourself. Sounds familiar (looks like we are all off-loading our thinking itself to machines)

Side-note: As humans, we have a tendency to quickly judge or make quick decisions which stems from our times foraging and scavenging in jungles.

Another Side-note: at a certain point, I am unsure of how much to think about AI or not, certainly discussions about it that were happening 2 years ago weren't helpful in contexts that they are used now (well not in any way or form that a person discussing and getting into the weeds of AI 2 years ago is better than a person just getting into it say 2-3 months ago)

With the industry (moving so fast) [but that doesn't mean that you can't catch up with it, I feel like the fast word has made people think that they are falling behind which is imo wrong i suppose]*, It is basically unsure to me of any FOMO or anything if you aren't using AI already, I find this notion naive.

People might be making strong opinions (AI psychosis) and skills on the tools available at the moment the same done 2 years ago. We don't quite know about the tech as these are still black-boxes and how they progress and what these "AI skills" might survive or not in future. Heck, we aren't even sure if these tools might survive or not or wouldn't be made magnitudes more expensive simply to break even as they are given to us for the first time at percentages of the price.

I don't know if I should form (strong) opinions yet and also a question of its worth so much thinking efforts in the first place, probably just gonna do my own thing (the way I want to) which includes learning C at the moment. because learning is fun.

▲simonw an hour ago | parent [-]

I didn't exactly say that they were about to become wildly successful companies. I suggested that they had "found product-market fit" - not too impressive for more than a decade of work - and that their revenue may even be "enough to start covering their costs".

▲Imustaskforhelp an hour ago | parent [-]

Firstly thanks for responding and I wish you to have a nice day. your suggestions have value and I appreciate you writing the article. Perhaps enterprise businesses do end up becoming the fat and meat of the AI industry.

My question which I wish to ask: What would happen to these AI companies if they turn out to be anything but wildly successful companies, both to the investors who have already invested in it and to those who might be investing indirectly into it in the near-future (passive investors, retirement funds)

I would love to hear your thoughts on it!

Thanks and have a nice day :-D

▲simonw an hour ago | parent [-]

> What would happen to these AI companies if they turn out to be anything but wildly successful companies

I'm not nearly enough of an economist / finance person to answer that credibly, but I expect they'll go bust, and a lot of people will lose their shirts.

... and the model weights will be sold to other companies who will then run them at a profit, and eventually figure out an economically sustainable way to train new ones.

The 1800s railway booms are a good comparison here - a lot of companies went bust, a lot of investors lost money, and we still ended up with railways.

If the AI companies all go bust we're going to have a lot of spare data center capacity!

▲Imustaskforhelp 29 minutes ago | parent [-]

> If the AI companies all go bust we're going to have a lot of spare data center capacity!

I can be wrong I usually am but an AI DC != compute DC or that it might decrease the prices of servers substantially because of it. (well not exactly, I hope you read my whole message so that I am able to better explain what I am saying.). AI DC's try to optimize for one thing: running GPU's for immense scalability and flexibility (0 to numbers>=large_number).

Currently, its actually way worse, the server providers are some of the worst impacted by the industry at the moment because each server requires ram and ram is well... increasing in its price exponentially. It's really a tough time to be a provider at this time (in certain respects) directly because of AI.

It is unclear to me if spare DC capacity will have any meaningful impact to it. I don't think that atleast within compute (and not GPU/AI DC), that space was too large of a problem.

Fun fact but one of the largest providers (BuyVM) had its datacenter price from where they colo'd increase because of the immense demand at the moment for spots in datacenters by many tens of thousands of dollars that they did the first price hike in at this point at decades! The situation is this dire :-(

Ram prices might come falling down and DC's might get cheaper but they can only get cheaper to limit, they still need to for example DC security employees

and I wish to suggest that if anything, investors might wish to re-coup their losses within the AI loss, they might want to make up with what little they might have (ahem DC)

For example, if you wish to want to take at an even more egregious example of what I am suggesting, there are many new york LLC's who would much rather leave the properties that they own empty rather than decreasing the price of what it costs (which they have set to some egregious amounts). I think that for them, somehow the math ends up working out in the end somehow, so there might be something more to it.

I wish I was optimist but I don't believe that the gains in spare data center capacity are worth even a fraction of fraction of the damage if AI were to go bust as you suggested with trillions of dollars vanished.

So, with the data I have at the moment, I am unable to suggest that compute would be cheaper. Heck, it was cheaper before AI and compute prices have never been something that people worry about because there are sometimes 10x cheaper options than AWS,GCP,Azure with things like Hetzner/OVH and others (yes its not a 1:1 situation but still its a 95% overlap and for all intents and purposes, great)

I can see a potential where GPU compute can get cheaper, oh boy, its so much more expensive than compute but I feel like GPU's aside from AI might still have a much more limited niche than generic CPU.

The issue wasn't ever the pricing. Simon, I own 7$/yr vps's which run my websites fine because they are written in golang. I doubt it can get cheaper than it. (You can get a 3$/yr vps if that is what you are interested with using Nat VPS + cf tunnels)

I would once again appreciate to hear your thoughts on it. The only thing I realistically see is if Ram producers ramp up their productions and create a ram price glut in the next few years, but imo the prices would even out over the long term.

I have seen the point of spare DC capacity being raised up multiple times but I finally ended up writing a message which hopefully captures the nuance, but once again, I don't know the future about it.

Waiting for your reply and have a nice day Simon (& other readers) and thanks for reading if you did, I appreciate it :-D

▲simonw 8 minutes ago | parent [-]

I think we are in agreement that if the bubble bursts a lot of people will lose a lot of money. I don't have a strong opinion on the data centers, my main point is that I don't think AI "just goes away" if the bubble bursts, which seems to be something that a lot of people assume.

▲mfuzzey 2 hours ago | parent | prev | next [-]

It's actually worse than that. It's not just financial depreciation or that the existing hardware becomes obsolete due to being less powerful than new hardware but also that hardware being run all the time at high load actually has a limited lifetime of a few years so it will physically break...

▲jmyeet an hour ago | parent [-]

I agree but it's even worse than that.

Data centers come down to performance-per-Watt. Electricity accounts for 20-30% of a data center's operating cost [1]. I don't know the exact breakdown but the GPU part of that is probably the majority given how power hungry GPUs are. The B200 is upwards of 1200 Watts [2]. The B200 is rated at ~4.5PFLOPS of dense FP8. So you're getting 3.75PFLOPS/W. We don't know what the next generation will look like. The A200 (Hopper architecture card that preceded the B200) had ~4PFLOPS apparently but also lower power consumption. Obviously this changes depending on whether you're looking at dense or spare and FP8 vs INT8 vs INT4 vs FP4, etc so we're just using FP8 as a yardstick.

Imagine a fictional B200 successor, the T200 that has 8PFLOPS of dense FP8 at 1000 Watts. Well then a DC built on that where the T200 will likely cost similar to what the B200 does now, you'll get nearly double PPW so the same size DC and same electricity load is going to be like 2 of your old DCs in operating costs. That's a big deal when you've laid out a trillion dollars.

[1]: https://iaeimagazine.org/electrical-fundamentals/how-much-el...

[2]: https://www.trgdatacenters.com/resource/h200-power-consumpti...

▲mountainriver 2 hours ago | parent | prev | next [-]

How could extremely capable artificial brains ever pay for themselves?

▲WarmWash 2 hours ago | parent | prev [-]

Prices are not going to stay where they are.

You have either never seen a tech cycle, or need to be reminded of that. The pressure to buy more expensive plans is already starting to form.