Hacker News new | ask | show | jobs
by ern 39 days ago
A few days ago we were hearing about how the "free lunch is over", now we're seeing discounts and increased usage limits.
10 comments

This is clearly a well-timed loss-leading strategic market share grab! Anthropic have blown a lot of user trust in the last couple of months..

But, overall, the current AI pricing is completely unsustainable, across all AI companies, except via the exponential growth they are relying on. Dylan Patel did the most insightful analysis of this I've come across.. https://youtu.be/mDG_Hx3BSUE?si=nyJu4adwYCH1igbJ

Really feel like the current versions are for sure "good enough". Thats not how market capture is gonna function though and they are gonna keep pushing because the only moat is to stay ahead, so the problems gonna stay strange. at some point more compute isn't a reasonable answer, and optimization is, and my feeling is we are well past that point from a product perspective, but ipos etc etc
The only moat is the us trying to buy all the compute hardware in the world for the next two years. Then China, amd, etc are just making their own chips.
DeepSeek claims that they can run inference on Huawei chips. Not sure about training.
If not now then definitely soon enough
So I think the current generation of models are arguably all about the same in terms of capability. However, the requirement for exponential growth I mentioned is all about the economics.

AI companies are trying to ride a growth wave where the income curve lags the expense curve by 1-2 years, and at the same time investing 10x their historical income on next year's projected demand.

Everyone is selling their API calls at a loss, because to capture the investment required to scale the business up and the costs down, you need to grow your market now (in relative and absolute terms). And history shows, that in big tech you often have winner-takes-all situations, or, at least a couple of big firms will dominate, and the others will die. That's where market share becomes a key strategic goal.

But to secure that, they also need to be building next year's compute now. And if their anticipated compute needs are 10x this year, they've got a serious funding problem, one that can only be filled by capital with an appropriate risk appetite. You can only get this high-risk capital when the potential payoff is even more enormous, or, when it's a smaller bite of a much bigger pie. Hence, MS putting into OpenAI and so on. But the investment needs are getting so big we are starting to see some pullback from more conservative sources, but also record deals from others.

Now say an AI company does get the capital they need to grow. Well, they've still got a very serious supply problem. RAM, GPUs, water, electricity etc. Hence why there's a lot of deals and cross-investment going on - everyone is trying to secure resources and lower their overall risk exposure while keeping a foot in every possible door, so they can switch alliances whenever it's expedient, and because collaboration also helps the overall market to grow.

This all explains to me why the industry _needs_ the hype. These companies can't exist without it, because the money they need to sink in, in order to even be around in 18 months, far outstrips all reasonable financial practices. So it's capitalism on steroids or nothing. If you believe the AI story, then to that extent, it's rational.

But note that nowhere in this scenario does it suggest the actual consumers will be getting a consistent product at a consistent price!!!

We're subsidized by the Chinese government!

https://www.reuters.com/world/asia-pacific/deepseek-nears-45...

Cool go download qwen 3.6 and run it on a single GPU and you can avoid paying into a subsidized model
why are we pretending these are equivalents?

yes, single gpu open models exist. Now show me the one that can keep up with a SOTA api model on more than short code block evals.

Qwen 3.6 supports reasonable agentic programming. People are vibe coding with it. It's really not that far off. If you truly cannot make a model that was SOTA 6-12 months ago work for you today for free I don't want to know what your needs are.
People don't understand that deep seek is running a plausibly sustainable business. Like how qwen/Alibaba is.
Every AI vendor is trying to steal marketshare. For now the competition is good!
Free lunch? More like "free data". The fools who give their life data and most intimate Intellectual property over to the AI companies for free, yes that's a free lunch that won't be subsidized for much longer when the cost on them which has been unsustainable (their data being harvested for non-training purposes) come stop catch up with them.

Sincerely, - I see you AI companies harvesting our data giving us discounted subscriptions so we can not realize we are paying you to take our own data!

They need to build data centers and lots of them everywhere, preferably powered with renewable energy. Let the tokens flow like water. The models are finally getting to the point where the LLM just knows what you’re asking for and gives it to you.
there will be free lunch till they admit to themselves that there is no moat. Acquring customers at huge costs is a fools errand when models are mostly indisguishable.

Anthropic is learning that lesson now. Doesnt help that their ceo goes around antognozing everyone by claiming jobs are over and annoying boris does like 500 podcasts per week repeating "coding is solved"

I'm guessing there was a pullback in usage as the free lunch started ending. So we get some more subsidized usage.
* from Chinese labs
What advantage do you think they have?
I’m not happy with their privacy policy [1]. I’m unfamiliar with the phrase “Parties with Other Legal Rights”. Given the well-documented struggles of Anthropic and others to provide enough compute, I wonder if “Parties with Other Legal Rights” constitutes part of the advantage here.

[1]: https://cdn.deepseek.com/policies/en-US/deepseek-privacy-pol...

Just run a local model or run deepseek from another provider with a policy you like. The models are open weight and widely available. Still cheaper than chatgpt and anything else through 3rd parties
this is the pitch - it's open source, run it yourself. But >99% of people will not have the hardware needed to run these models at a high enough quality to be close to SOTA. So they will run the open-source models on CCP systems for a good price.
What I mean is you can use providers who also host deepseek models for pennies without touching deepseek itself.
Operating in a jurisdiction where US companies can't sue them.
a lack of existential threat in the form of pay-seeking and remediation from the people you stole training materials from that allows for an intrinsically different pace of operation than the Western competition
A sane government policy that invests heavily on innovative businesses.
I can't figure out how there's both too little supply (so a dramatic need for more data centers) but also too little demand (so labs subsidize inference).
There isn't too little demand. There is massive demand and many competing companies trying to capture that demand, so they are attempting to make better offers than their competition. Hence subsidy.
That, and:

- Every competitor is planning for the demand to be much higher in a few years than it is now, and aiming to capture as much of that as they can, which starts by getting companies hooked on their models now

- The data center capacity will get used no matter who captures the most demand

I can somewhat understand companies getting users depentant on their harnesses or workflow, but model vendors as in this deepseek case, I have absolutely 0 model loyalty when it's a simple config change away, and will always optimize for either capability or price (or whatever !/$ metric you can determine).
Depends what you’re doing. For example, Gemini is somehow still your only option if you need a model that can natively understand video and reference timestamps in its response.
You're right, of course, but I would qualify this under "optimize for capability".