Hacker News Reader

Comment by therobots927 | Hacker News Reader

The price shock that’s about to hit the AI industry is gonna be very fun to watch.

I keep asking our product team what the plan is when our API/Infra costs eventually skyrocket and our AI features become unsustainable, but there's absolutely zero preparedness.

We throw in all these cute little AI features to fill out marketing bullet points because they're basically free, but if they had real cost we're going to have no choice but to take them away.

But you gotta have that AI strategy or your good numbers won't go up. You can't be left behind in this business of business. I hear everyone is already moving on to post-agentic AI, whatever that means.

Recently while interviewing for a SWE position, I was talking to the company's VP of engineering and expressed my concern over these LLM tools precisely because of the economics of it, because right now most if not all of these tools are being subsidized. Also, the environmental impact.

I was talking mainly about code generation tools, which can be completely shut down today without affecting production. Not even considering LLMs that are implemented in user facing features or customer service right now.

The guy's response: "oh but these things will surely get cheaper. Every technology gets cheaper with time. And about the environment, yeah... unfortunately there's not much we can do."

That's the level of forethought for a VP of engineering where I live.

That’s the level of forethought encouraged by investors and the CEO in all likelihood. VPs are just doing their jobs. Even if they might agree with you, that sort of sentiment is a quick way to get you on the CEO / board’s bad side. The incentives are perfectly aligned to keep shoving air into this bubble for the time being.

What is the real AI cost at which point this is an issue for a corporation you think? I contract at a place that gives $1,000/month budget per SWE (5 Max accounts) and I am sure if the price was 10x that no one would even blink.

It is basically cost-benefit analysis just like with any other cost (and there is always option to revert to local?)

Don't worry your users will rejoice when you rip them out. I know when Atlassian gets hit by this and has to disable their AI features their products will be a little bit less unresponsive. They'll still suck, but they'll suck a bit less

I think a lot of usage will move to the cheap open weight Chinese models once there is an incentive to do that. Everything below the highest end frontier models are becoming commoditized and I suspect the commodity segment will pass the "good enough" bar for most applications.

It'll hit all of us, not just the AI "industry"

Yeah, and I doubt that this is the last price hike. But I don't think this actually has a whole lot to do with ram prices. The thing is, AI cloud providers have generally been running at a massive loss, if you do the napkin maths, the procurement cost of these machines is equivalent to approximately a year of rent at these rates, for normal servers at AWS it is approximately a month.

As a user of Claude I have been worried about this.

As someone who can't stand these tools, I'm ecstatic. Y'all have been hooked by the early venture-backed "scale up at all cost" period. You're gonna get squeezed so hard once investors start demanding returns on their investments -- and this whole craze is, hopefully, coming to an end.

Maybe don't hold your breath on that. Even if the price for AI tools would double, or tripple...that's still a very small part of the actual cost of an employee.

Sure, but employers will want a return on investment too. And productivity studies which aren't backed by these AI companies haven't exactly been promising. With how stingy some companies can be with the basics such as getting a laptop with enough RAM and drive space, a lot of the spending on AI tools for employees is clearly driven more by hype and the false promise of huge productivity gains than the normal expense approval process.

And plenty of people get hooked on these tools through using them for free or almost-free in their spare time. Those people will balk at huge price increases.

... and almost no company of any size above startup counts cost like that.

But I agree no point holding breath, whether somebody jumps on wagon or not won't change if price per query doubles, either its this massive productivity increase where costs of llms are a rounding error in overall costs or it isn't.

I’m betting on “isn’t”

That's what people widely claimed about Uber: it was toast once the investor subsidies stopped. Now it's quite profitable.

People will pay more. Claude Opus 4.5 is worth more than $20 per month, as is Gemini 3 Pro. These services keep getting better. Another three years of improvement, why shouldn't that command $30 or $40 instead?

$20 is ~$10 in the year 2000 per the BLS inflation calculator (or $1.25 when priced in gold). Nobody would have thought that was expensive for such utility. These are inexpensive tools at present.

I think you're ignoring that most users of AI currently aren't paying anything, nor would they. I believe the value of a Facebook user was $70 per year in 2023, for the US and Canada. Assuming that the AI companies could make twice that from ads, that's still only $10 - $12 per month, and even less in the rest of the world. Obviously there's going to be some business users as well, so they can cover some of the cost, but would also be responsible for a larger portion of the running cost.

The question should be how many free users can the AI companies convert.

The cost of an Uber has also gone way up, and they basically have a monopoly in many areas.

The cost of free users is much lower because they are served lite models, hit quota limits quickly, and can't soak up tokens by using agents. The main capacity usage is from agent loops, which is universally behind a paid tier.

At its worst, Uber had a net margin of -~60%. The AI labs are all running at least negative triple digit net margins, some running negative quadruple digit net margins. This is why AGI has been "forecasted" to death by the labs, because investors need the promise of infinite automation to stomach the losses.

Anyway, in this instance, what you received for $20 in 2025 will run you somewhere in the range of $60-$90 in 2027/2028. In the interim, you will likely see that $30-$40 of service gets you what cost $20 in 2025. The most likely avenue for this will be reduction in subscription user limits, and for API customers premiumization through substitution. The latter being a situation where what would be the next Claude Sonnet model is now sold as Claude Opus, for example.

The only way the math works for the consumer is if the user base has become dependent on the service instead of remaining in a conventional cost/benefit relationship.

You're likely going to stay seething for a while (even if bubble pops). the product market fit of these tools is so strong that companies paying tens of thousands per employee for them isn't really that big of a deal

Reply on news.ycombinator.com