AI is too expensive
93 points
1 hour ago
| 19 comments
| wheresyoured.at
| HN
luk212
2 minutes ago
[-]
A lot of the current AI economics seems to depend on three assumptions being true at once: 1. inference costs fall fast enough 2. usage grows into very large recurring revenue 3. customers don't cut once handed the bill

We should draw a distinction between "AI is valuable" and "AI justifies its current investment levels." There's real productivity value in AI, especially for things like search, boilerplate, tests, refactoring, etc...BUT that doesn't mean every enterprise should let token spend grow without strict telemetry, cost-attribution and outcome-based measurements.

The teams that win here will not be the ones using the Most AI, but the ones that treat it like any other expensive production dependency, which means measuring unit economics, cap runway usage, properly align models with tasks(not just Opus everything), and scale workflows with ROI in mind.

reply
czhu12
33 minutes ago
[-]
I really respected Ed Zitron, but I feel like he's very much lost the plot on AI.

Scroll back not too far and he was publishing criticisms that no one wants to spend actual money AI. Anthropic has shattered all notions of that since then.

Then there was the idea that even if people want it, we have way too much GPU capacity to ever be saturated. Now almost all providers are hitting limits.

Now, its the next iteration that even if people want to spend money and GPU's are at capacity, its just never going to be profitable. This may or may not be true, especially with more capable open source models that can be served at cost. But at this point, he mostly just brings up anything possible to downplay AI

reply
bauldursdev
4 minutes ago
[-]
I'm not really an AI-futurist or anything, I think the truth is between the extremes, but it seems like a lot of people who are ideologically against AI just move the goal posts whenever a new development is made. They also seem to operate under the assumption that whatever the current state of the technology is is as good as it will ever be. "it can't even do x today, so it'll never be good".
reply
ant_li0n
21 minutes ago
[-]
I agree. He plainly has an axe to grind. I'm as AI-skeptical as the next guy, but I can't handle Ed Zitron. Doesn't seem like a good faith actor.
reply
voxl
5 minutes ago
[-]
Are you a bot? At what point has anyone claimed we have more than enough GPU supply. Why would that even make sense as an argument at any point in time against AI? As far as I'm aware Zitron has discussed at length that companies are buying currently non-existent compute at yet-to-be-built data centers
reply
changoplatanero
46 minutes ago
[-]
Imagine you were looking at Google, a sustainable and profitable business, and you thought you saw a once in a lifetime opportunity to compete with them and take their position as a leading tech company. How much money would you need to spend to make a credible attempt?

Google has had decades to accumulate intellectual and physical capital. Catching up quickly means spending >500 billion. If you can actually dethrone Google (admittedly not an easy task) then it will have been worth it. If not, I suppose it's wasted investment.

Now what happens when three or four startups vie for this opportunity at once? Well that's how you get $2 trillion in captial investments per year.

reply
jayd16
41 minutes ago
[-]
What a strange train of thought. Why would you need that amount in this hypothetical? Why would dethroning them alone be worth it? It would literally only be worth it if you could do so profitably.

More realistically, it seems like someone calculated that it could still be profitable up to several hundreds of billions of dollars which explains the initial investment. And continued investment can be explained by trying to salvage the existing capital spend. But even if it's the best option those companies have now as far as a hypothetical goes, it still might not have been worth it.

reply
mstank
3 minutes ago
[-]
I think the opposite is true. To dethrone the top tech company, you need to be able to spend much less than them, at higher efficiency and faster growth. Google didn’t catch up to Microsoft and Apple by spending more, they caught up by developing business lines and flywheels that were much more capital efficient.

If it’s a spending game, the incumbent has a huge advantage.

reply
doctorpangloss
35 minutes ago
[-]
You: says someone has a strange train of thought, and then you also ask how can a company become profitable in a situation where it becomes a monopoly? Dude, the winner raises prices? "AI" is not expensive enough!
reply
giancarlostoro
26 minutes ago
[-]
Meanwhile Meta / Mark Zuck is in crisis mode trying to come up with some meaningful way to break into AI, instead of competing in what's currently hot, they'll probably remake Elons weird persona system, which is not exactly the hottest thing on the planet, and billions down the drain that could have gone into building more efficient LLMs instead.
reply
ethagnawl
24 minutes ago
[-]
> billions down the drain that could have gone into building more efficient LLMs instead.

Or any one of thousands of other ventures which could be more beneficial to humanity, the environment, etc.

reply
giancarlostoro
22 minutes ago
[-]
Agree, though if he's so fixated on AI, and Meta has released plenty of things in AI that have affected the industry in general, he would invest in making less resource intensive LLMs.
reply
fred_is_fred
36 minutes ago
[-]
Most disruptive start-ups don't come from a giant pile of cash, but from new ideas that the old players can't or won't adopt. It did not take $500B to build a digital camera.
reply
Forgeties79
17 minutes ago
[-]
No but the first one was the result of Sasson’s R&D while at Eastman-Kodak, a massive company that failed to capitalize on it years before anyone else was near it. They easily could’ve been the big player if they didn’t fear the impact on film sales. They had a solid decade head start and blew it.

The way digital cameras developed (hyuk hyuk) is arguably exceptional, definitely not a clean example.

reply
bensyverson
48 minutes ago
[-]
If you're older than 30, you've seen this play out before… This is just how the VC game works. Cross-town Uber rides don't stay $5 forever.

The bright side is: this is a golden era of subsidized tokens. It will not always be like this, so now is the time to churn out your passion projects.

reply
Aurornis
32 minutes ago
[-]
Uber is a good comparison because everyone was predicting the demise of ride sharing as soon as they tried to become profitable.

The subsidies went away gradually and the prices leveled out in a spot where the services are heavily used. Uber became profitable. Ride sharing is affordable.

I think our $20/month plans might become a little less generous and the $200/month plan won’t always allow non-stop vibe coding, but I don’t think the prices are going to rise so much that users are priced out. Like Uber, customers will grumble for a while and then adapt to the new normal.

The big difference is that compute hardware is getting better. I think we might overshoot with data center buildouts to the point that compute becomes cheap, while hardware improvements continue to lower the cost of serving models. Over time the same service becomes cheaper to operate, opposite of Uber where driver wages are creeping upward.

reply
WarmWash
12 minutes ago
[-]
AI will likely cost $60-$80 per user per month.

Akin to an average cellphone bill. The infrastructure costs are comparable and the ROI would be 5-10 years for the current insane build out.

Yes, chinese and local models exist. But so do $20 cell phone plans. People go with what is convenient, works, and is readily available.

reply
eieiewq
24 minutes ago
[-]
No it’s a bad example as uber had solid unit economics. Uber is more Akin to Amazon.
reply
bensyverson
23 minutes ago
[-]
I would love to see the data to support the idea that Uber had solid unit economics during their expansion.
reply
eieiewq
22 minutes ago
[-]
State your priors so I can determine whether you are qualified to have this discussion.
reply
bauldursdev
1 minute ago
[-]
"I know these facts, they just go to a different school. I'd bring you to them if your Mom would let you leave the neighborhood, but you're just a little boy"
reply
bensyverson
15 minutes ago
[-]
haha what

My "priors" are that I raised a search fund and analyze companies every day.

reply
SirMaster
21 minutes ago
[-]
I like writing my passion projects without AI. Is this so strange?

Using AI to write my software just takes all the fun out of it for me.

It feels like just reading a summary or recap instead of reading the actual full novel myself. Like it defeats the whole purpose of it. I write software because it's fun and it stimulates my mind and teaches me things and improves my skills.

reply
bensyverson
13 minutes ago
[-]
Not strange at all! I think there will always be people writing code by hand. I take photos with a 100 year old camera because it's fun.
reply
edwin2
35 minutes ago
[-]
Uber was an application of technologies sitting on top of many iterations of performance optimizations. (Think: the difference between 2009 internet speeds versus 2019 internet speeds. Or, 2009 smartphone specs versus 2019 smartphone specs.)

You could imagine a Moore’s Law-esque cheapening of the tech that coincides with the business raising their prices. That might look like a continuation of simply “using the tools” on the surface, but on the inside it would spell a gradual, meaningful increase in margin

reply
skybrian
25 minutes ago
[-]
For Uber, paying the driver is unavoidable (for now) so this isn’t a good comparison at all.

A better comparison is with how much PC costs went down during the 80’s due to IBM clones and Moore’s law.

reply
Aeolun
43 minutes ago
[-]
I think the good news is that we’re not at peak cheapness for tokens yet, but companies like deepseek show that it is perfectly possible.
reply
neals
31 minutes ago
[-]
Token cost will come down in the future, it might even out
reply
bensyverson
30 minutes ago
[-]
It's hard to predict the future, but it seems likely that flagship tokens will become more expensive, but many people will use free tokens via local inference.
reply
giancarlostoro
23 minutes ago
[-]
I mean SubQ claims to reason as good as the frontier models, not better, but reasonably as good, and is 5x cheaper, so the future might not be all in on overpriced LLMs. Data science devs don't engineer for scalable performance, they engineer for the model's capabilities first and foremost.

SubQ was validated by at least one third party, not sure if we'll see more confirmation, but 5x cheaper costs is worth it. None of the frontier models care enough about cutting costs of their models, only being the best in benchmarks.

https://subq.ai/

reply
BoredPositron
41 minutes ago
[-]
Probably the worst comparison I have seen on the topic. gz.
reply
TheSkyHasEyes
30 minutes ago
[-]
Eventually you come to realize the more things change the more they stay the same.
reply
BoredPositron
20 minutes ago
[-]
Sorry but no. It's a flawed analogy to suggest that an emerging technology like AI follows the same scaling laws as the taxi industry. AI benefits from active R&D, rapidly improving hardware, compounding software efficiencies, and deflationary cost curves that simply have no equivalent in a labor- and vehicle-dependent service...
reply
dosinga
48 minutes ago
[-]
The counter argument (not mine): Software Engineers are willing to spend their own money on AI. The same people that wouldn't pay 10 dollars for code if there was a workaround that took hours.
reply
righthand
12 minutes ago
[-]
The same people that laugh at you when you mention paying for Kagi or alternative to an advertising supported web.
reply
nubg
46 minutes ago
[-]
Interesting perspective
reply
simianwords
42 minutes ago
[-]
I did a thought experiment: if you went back to 2019 and could use AI in your job at the current market price - like lets say using the latest Deepseek V4, would you pay for it?

Hell yeah obviously. There's close to no doubt. So why do we think its not true now?

reply
graydsl
33 minutes ago
[-]
Am I the only one that has AI or does everyone? If everyone has it what is the thought experiment? Why 2019?
reply
eloisant
39 minutes ago
[-]
The thing that everyone seems to be missing is that the US AI companies are focused on the frontier models, that are very expensive for diminishing returns.

If suddenly the money craze stops, meaning (1) AI companies investors want them to become profitable and (2) clients start being cost-sensitive to AI bills (which they are absolutely not currently), then everyone will switch to smaller, cheaper models that are enough for a lot of use case.

Sonnet instead of Opus. GPT 5.4 instead of 5.5.

Chinese models.

People keep comparing to Uber but Uber can't suddenly make it cheaper to operate.

reply
not_the_fda
12 minutes ago
[-]
I work in design services and see clients already being cost sensitive to AI bills. They have wait lists, rationing, push to lower tier / cheaper models. This is all with subsidized pricing.

People are going to decide its too expensive for everyone to use AI agents and un-subsidized pricing.

reply
d0100
36 minutes ago
[-]
> GPT 5.4 instead of 5.5.

I am exclusively using 5.4 because its only 1x and very good, but the github calculation showed my once $40 become a $680 billing

That is too expensive and not worth paying

reply
wagwang
6 minutes ago
[-]
You have to look at use cases and there are a bunch of slam dunk use cases that are wildly profitable at todays token prices, whether we keep finding use cases as intelligence goes up is another story.
reply
overgard
48 minutes ago
[-]
More people need to read Ed, especially tech journalists. I feel like he's one of the rare few people that are actually speaking about the industry honestly.
reply
nstart
14 minutes ago
[-]
I've been subscribed to ed for a long time. I commend his foundational ideas like what he laid out in "The Era of the Business Idiot" or "The Rot Economy". My recommendation line for him to anyone else is "if nothing else, he'll leave you with something to chew on for a while to come".

My issue with Ed is that he doesn't have the ability to draw the line. In the pursuit of making a point he goes so dogmatic that he is willing to make harsh statements that go beyond number backed predictions. Like in his piece "AI is really weird" he states about agents, "Probably the weirdest thing about this entire era is how nobody wants to talk about the fact that AI isn’t actually doing very much, and that AI agents are just chatbots plugged into an API.". That's a massive stretch to make. Just because he has a claim that the business doesn't make sense, he doesn't get to claim that agents are not capable of doing very real work. His assessment of cowork was "a chatbot that deleted every single one of a guy’s photos when he asked it to organize his wife’s desktop.". These statements damage his credibility and make it too easy to dismiss his writing as a rant of an angry man.

reply
skybrian
28 minutes ago
[-]
I’m paying API prices for my hobby coding due to the coding agent I use. So far I’ve switched from Opus to Sonnet to GLM 5.1. Looks like it’s about 25% of the cost and quality seems good enough so far.

I think competition is going to keep customer costs low if you’re willing to switch. Maybe people on expense accounts won’t care, though?

reply
sailfast
52 minutes ago
[-]
These are all just bets that eventually someone wins anyway, right? Adoption is good but marginal revenue doesn’t matter if and when these models and solving world hunger - or have created the next yakuza mega corp that governs the world - right?

Feels like an unspoken rule here. Everyone wants to own a chunk of nuclear weapons and it doesn’t matter whether it’s profitable. You just need the nukes to survive and have a seat at the table

reply
Havoc
49 minutes ago
[-]
It’s a similar bet as Uber. They also started out with numbers that make no sense - overpaying drivers and undercharging users

The math may look questionable but there are also senior people talking of automating all white color work in the next couple years. Even if that estimate is miles off on both time and % it’s still trillions. So crazy as the numbers seem it could still work out

reply
not_the_fda
7 minutes ago
[-]
Automating white color work in the next couple years will cause the greatest demand destruction humans have ever seen.
reply
empath75
34 minutes ago
[-]
Cars and gas and human labor do not get cheaper over time the way that computer hardware does.
reply
hypnodrones
50 minutes ago
[-]
The rug pull on users is bound to happen and it will involve advertising.
reply
lifestyleguru
18 minutes ago
[-]
The vibe coded software designed from the ground up to contain ads will be something regrettable. Will be like a doctor smoking cigarettes while prescribing opiates.
reply
add-sub-mul-div
38 minutes ago
[-]
Algorithmically and seamlessly weaving undisclosed advertising (or other editorial content) into conversational output is their holy grail. It's the endgame. There's a reason they're pushing so hard.
reply
hypnodrones
18 minutes ago
[-]
I'd even walk back on just calling it advertising, because we immediately think of the usual ads we see everywhere. The actual thing here might be much more subtle and worrying, you could call it undisclosed influence.
reply
sybercecurity
18 minutes ago
[-]
Probably endgame plus getting "too big to fail" and getting gov't bailouts if things don't work out. It's part of the lobbying theme that LLMs are the next great power struggle.
reply
WarmWash
9 minutes ago
[-]
I mean, we could just avoid this if people realize it's not morally wrong to pay for a service.
reply
pingou
29 minutes ago
[-]
The end goal of these companies is AGI, or even ASI. If you believe this is around the corner, and think AI can do the job of a human for less money, it makes sense to put all your money into working towards that goal and buying as much compute as you can. This is especially true since whoever gets there first (or is simply ahead and can use their AI to get even better) gets a big advantage.
reply
hoansdz
44 minutes ago
[-]
The cost of tokens used by AI in many fields is even greater than the cost of human services; people are experiencing FOMO, but once the wave passes, the market will stabilize.
reply
KoolKat23
43 minutes ago
[-]
The timing of this is great considering Google's rumoured Gemini 3.5 flash pricing spike.
reply
jqpabc123
1 hour ago
[-]
It's just getting started. You won't find out what the real, actual cost is until after you build it into your workflow.

In other words; right now, we're still in the "bait" phase. The "switch" comes later.

reply
b65e8bee43c2ed0
59 minutes ago
[-]
Chinese 1T+ models are being offered at a fraction of GPT/Claude cost, and the margin is healthy enough for dozens of providers to compete, so I find it highly likely that ClosedAI and Misanthropic sell tokens at massive markup. they just still bleed billions on their free tier and san francisco salaries.
reply
citrin_ru
54 minutes ago
[-]
Why we should assume that whoever offers these Chinese models makes sufficient profits and will not rise the prices eventually too?
reply
larme
35 minutes ago
[-]
Some of these models are open weight. You can try hosting them and do the price calculation yourself.

They also publish papers talking about how to save kv cache and computation powers. Because currently they don't have the most powerful nvidia cards, training and inference efficiency is very import for them.

reply
han1
47 minutes ago
[-]
Chinese models are state-funded and not concerned with taking profits.
reply
happyPersonR
1 hour ago
[-]
Next year is gonna be when the “switch” comes lollll
reply
gdulli
42 minutes ago
[-]
And the switch will continue, it won't be a one-time event. Like how the price of Netflix etc. keeps going up periodically.

If people's dependence on their streaming service keeps them captive, just wait until people have gone 5 years without doing real work.

reply
ReptileMan
56 minutes ago
[-]
Next year we will have new deepseek.
reply
happyPersonR
31 minutes ago
[-]
I see a couple of possibilities

1) someone deepseeks deepseek lol:

Generates their own weights and figures out a way to determine all of the intermediate states.

2) places realize there’s real risk with using a model that might have things baked into it that produce specific flaws that could be security bugs, but only under certain conditions.

reply
feverzsj
16 minutes ago
[-]
They are betting on AGI, in other words, Bullshit.
reply
slackfan
18 minutes ago
[-]
> If you liked this piece, you should subscribe to my premium newsletter. It’s $70 a year, or $7 a month...

This blog is too expensive too.

reply
chipotle_coyote
5 minutes ago
[-]
It seems to me (entirely anecdotal, YMMV, etc. etc.) that Ed Zitron’s blog posts started getting both longer and considerably more histrionic when he started moving most of them behind a paywall. I’m definitely in the “AI skeptic” camp and think Zitron has good points to make about both the shaky business models around AI and the unrelenting hype train, but it’s hard not to get the impression that he’s found a niche of preaching to the rabid AI haters willing to give him money to keep spouting increasingly repetitive vitriol toward Sam Altman and Dario Amodei.
reply
simianwords
55 minutes ago
[-]
If AI is too costly: bubble will burst because costs are unsustainable.

If AI is too cheap: bubble will burst because you can run them locally and data centrs are not needed.

If is it in-between, AI companies make too much money and they make too much profit which is bad!

I don't think this guy is a serious commentator.

reply
sophacles
27 minutes ago
[-]
This isn't an accurate summary, even at a 50Kft view. Are you a serious commentator?
reply
harimau777
51 minutes ago
[-]
I mean, those all seem like true statements.
reply