I just look at technology over the last 50 years and see: decreasing prices, increasing capabilities. Barring temporary hiccups. Altman & co did a very smart thing by locking up the GPUs and the RAM, they effectively created a temporary near monopoly. At the same time: that incentivized the rest of the world to look more closely at efficiency negating some of the gains of that expense.

It's just like a utility with the slight difference that it can be sold at a distance without significant cost for transportation, in many ways it is the ultimate digital good. Unlike Spotify copyright is not an issue, unlike your electricity, gas, internet or water company there is no last mile problem. Unlike Google there are a ton of qualified contenders already (Google had to be better than Altavista, which wasn't hard, but they were so much better they managed to lock up search for long enough to cement their position).

Anthropic has none of the advantages and all of the disadvantages. On top of that all AI companies seem to be run by megalomaniacs.

The long-term is considerably easier because in the short-term there are limits on supply so prices aren't really able to move in a way where balance will be found.

In the long-term, the marginal cost of a request is zero so prices will plunge and a lot of the fixed investment will get stuck. The difference between the cost of training a model/infra and serving a request from that model is a clear problem for economics.

What happens with Anthropic specifically is less clear because they, as every company in the same position in the last 150 years has done, are going to push for regulation. A lot of the risk has also been shifted onto neoclouds. The value of the customer relationship is, however, effectively zero. A core part of their strategy has been to build stuff out like CC to differentiate that but, ultimately, that won't work.

The other strategy they have employed is trying to get other companies to stop training new models. Again, seems unlikely to work.

There is no scenario in which they become Google search outside of government intervention. Obviously this is a key part of their strategy now but at best, this will only work in the US and Europe (I would actually say they are much further along in Europe where there is a strong tendency towards corporatism and centralisation of power, they could be "Europe's AI").

The evidence we have so far points to LLMs becoming commodities. I can’t think of a technology in the last 30 years where so many companies are within ankle biting distance of the leader. There seems to be no data aggregation moat like with Google.

I have a little different take..

Top models, capable of discovering and inventing stuff, held back from public, and available only as for-rent research service (or whatever the relevant bureaucratic term is) to big corporations and government agencies like NASA, DARPA, NSF, NIST, DOE, etc. only.

I have a feeling that we, the public, will stop getting access to more intelligent models starting next year.. i mean, not the incremental ones, but, major ones.. like the one you got going from opus 5 to opus 5.5.

How is it not clear that LLMs are being increasingly commoditized and that the foundation model producers are in a bad race to the bottom with each other AND cheap open-weight competition?

Not even just from Chinese models, but from companies/labs like Mercury

I wouldn't bet on local compute. Even if it may end up cheaper in the long run, the convenience of just another SaaS subscription sure beats the high upfront const and continuous maintenance required with local deployments.

We've seen businesses moving everything to the cloud before AI, we'll see more of that as regulatory requirements are increasing.

>> What's the endgame once the dust settles?

Given the trillion-dollar spend, the theory is that every many woman and child in the western world will devote a substantial portion of their income (like 5-10%) towards some sort of AI subscription.

or...

That doesn't happen. All the AI investments go south and every man woman and child in the western world will devote a substantial portion of their income (like 5-10%) to bailing out an economic system that went all-in on AI. Either way, the money has already been spent.

I don't see a path forward for valuations like Anthropic, either the hardware requirements fall off a cliff (and why would people pay Anthropic so much if it runs on a cheap laptop?) or either they don't and they will never have the capabilities to serve all their customers (and why would they be valued so much then?).

It's a lose/lose situation here for Anthropic (and OpenAI) whatever happens.

I agree it's hard to predict.

It's very possible given the current environment that LLMs go the way of airlines. Low margin businesses that are essential to world economics with heavy price competition.

It's going to be all about the training set. Theres no moat in models trained on internet data.

There are tons of private datasets where a model will be fine tuned to provide useful domain specific responses.

>The problem for me is that it's really hard to project what this industry is going to look like 6 months from now, let alone 6 years.

i keep thinking how crazy it is we went from ChatGPT releasing in 2022 to multi-trillion IPOs 4 years later. the industry still hasn't agreed on how to generally use anything. crazy times ahead.

For me, thinking about risk profile, local is the way to go. With cloud computing many feared of trasnfering data to the cloud and loosing the ownership. While that had some merit, there were no infrastructure to actually abuse the situation where vendor can realistically understand or use your data.

In LLM world model provider has the opportunity to understand and use the information you value as a company. Append this with regulatory requirements and local, even if less able models, are the way to go

It's going to be "all of the above" because AI has a whole lot of use cases.

For certain AI use cases, AI will be like a utility. For other user cases you need to pay a premium for the frontier. For other use cases AI will be local.

Even in the industries you mention (water and electricity) you have all sorts of businesses. Like people selling water in fancy cans for a large markup or batteries vs wall power.

> Or is Anthropic (or some other big AI lab) going to swallow everything and become the Google Search of LLMs, with everybody else fighting for scraps?

Of all scenarios, based on the data we have now, this seems to be the least likely one.

> Or is it local compute, models becoming more efficient and [...] companies hosting their own AI clusters like they would run SAP?

Already happening, which is requiring a lot of enterprise DC power retrofits.

When google became the king of search the incumbents were so far behind that catching up wasnt possible and they thought their position was entrenched anyway. With LLMs openAI's competitors are at or close to their capability already and everyone knows they need to keep spending to remain relevant. So unless the cash pipeline dries up for competitors that is unlikely to happen. The reason for that is anyone who has invested knows that either they give up and hand everything to their competitors or someone else does. Which means you need to keep spending to outlast your competitors. These companies are too rich and there is too much VC money around and enough billionaires with more money than they and their descendants could spent in many lifetimes, and a huge amount of greed. If the money runs out that is because some very bad shit has occurred and LLMs will be the leats of our worries.

>Is it going to be a utility like water or electricity and the cheapest offer will prevail

This

I don't particularly care why they do it, they do it. OpenAI is less open than the Chinese.

"The Chinese labs are not providing their open-weight models for free because they think it is good for humanity. "

Neither do Anthropic and OpenAI.

neither did google or any other american company did open source anything for the greater good there is no point in this comment.

> for free because they think it is good for humanity

I like Qwen as much as the next person, but that's a bit to credulous. Those models would not get the compute allocation nor distribution they do if it was not aligned with broader policy.

How long until China has its own chips?

It's a shame "AI" cannot be made to run on bitcoin mining dedicated hardware because they mastered that in only a few years

You aren't paying API prices. Anthropic (or OpenAI, SpaceX, etc) are almost certainly losing money on your subscription. How long they'll be able to continue to subsidize your usage is an open question, but they definitely won't be able to do it with everyone (currently, their solution is to make enterprise users pay per token, but that might change). What that means is that for the customers that the frontier labs need to be profitable, going with a closed weight provider isn't "marginally better intelligence for marginally higher (or even lower) prices" but "marginally better intelligence for much higher prices".

You are not the real consumer though. The real consumer is API, inside of products ingesting vast amounts of data at scale.

My last role swapping from Haiku/Sonnet to GLM dropped costs by 90-95% and increased quality.

Money talks.

Yeah. People have very strange ideas about competition and choice and moat. If all you needed to unseat the entrenched was a competitor then Bing would be going toe to toe with Google, and Edge with Chrome. They have a billion+ devices they can stick those options as defaults and still nothing. There are many examples.

People are naturally sticky. If they're satisfied with a product, they'll stick around unless something comes along that is so much better it gets people talking. Open AI's 1B+ customers won't give a shit about Chinese models. Enterprise customers won't move to Chinese models, especially just to save money. They are a non issue, especially if the M.O is just going to be playing catch-up.

At $100 / month (or $1200 / year) it may not be worth it, for you. But many of us would be spending a (large) multiple of that and then thinking about how much you spend, where you spend it and who you spend it on can make a big difference. For companies there are other considerations at play as well, for instance trust.

The $100/month guy for inference isn’t the prize. It’s the thousands of companies spending millions/month.

It’s not unusual for someone to spend $5k in a week. 2 guys at my place just spend $150,000 over 3 months building out a factory.

We are aggressively exploring open weight models now.

First off, $2k is the price of the new iPhone Duo. Suggesting that that is basically the same thing as a $500 Android phone is... strange.

But also, I don't think you should use individual consumer behavior as a proxy for how the overall LLM market behaves. Current estimates are that individual users represent a pretty small share of overall genai revenue. Most of it comes from enterprise contracts. Those might be more price sensitive because the CFO gets a say in which vendor they choose.

For a couple years now I've been accustomed to choice of foundation model being a heavily scrutinized design decision because it's just about the easiest way to manage COGS. I can certainly imagine scenarios where, given the option to pay 1/2 as much in inference costs but with a 5% reduction in output quality, the boss would say, "Nah, that's a bad deal." But I have yet to encounter one in the wild.

But if you’re a heavy user you’re probably paying $100 for something that costs $1,000. Hence Anthropic’s problem. If they charged real prices to the masses the massses would flee.

I spend extra on a smartphone because I use the thing all the time, see it all the time, hold it in my hand all the time, it's my camera, music player, note taker and so on.

For AI? Prompt comes in, tokens come out. The value of the 'ecosystem' is roughly zero to me. Switching costs between models are essentially zero. The UI and UX is the same whichever LLM runs underneath.

If you do use llm some, then the $20 tier makes sense, but that's not where the money is. If you use it a lot then a $100 sub probably makes sense. But if you're a BigCo that heavily uses it and pay API prices and suddenly you pay $200 million a month, well, maybe it's worth buying a bunch of hardware and stick it in a basement somewhere, and you won't upload all your secrets to your future competitor to boot. If US labs didn't care about the competition from Chinese models they wouldn't lobby so hard for banning the evil, unaligned, super dangerous communist models that want to steal our freedoms and make our gas expensive.

Ultra-cheap models like Deepseek are also very nice if you need to burn billions and billions of tokens.

It's really marginally better and in some cases actually worse. If 100 dollars is throwaway money for you, and prices don't increase sure why switch. For heavier users, and for people with less money(most of America is struggling with bills and food) it's not attractive at all.

Because the $100/m fee they're charging is only there because Chinese models will undercut them if they try to raise it further.

There's been a headline circlulating lately: 80% of Anthropic's revenue comes from 1% of their customers. The $100/mo or however much individuals are paying doesn't seem to be material.

I was playing around with OpenCode on Omarchy over the weekend, working on my Android app, and was struggling to spend a single dollar with DeepSeek. The quality difference from the frontier models I use at work was hardly noticeable (where I typically rack up at least a few hundred dollars per month).

LLMs are not a luxury brand that individuals buy. Companies are always looking to save money. And their enterprise consumers are all developing their own models, so idk where you are getting this "analogy" from.

Anthropic doesn't care about sub users. At all. The real target is pay pig enterprises that give their employees access to the API and allows them to go buck wild spending $1000 on fable in a day.

People buy iphones due to network effect, android was late to the RCS game and now it's a moot point anyway.

How many people use an iPhone to its maximum? Isn't it the same with models? I think engineer's like to flex that they need a top of the line frontier model for their shitty corp projects. A Chinese model will write your Jenkinsfile just fine.

The LLMs don't have the moat that Apple has. Apple has carefully built their walled garden to keep people in. Their messaging, their cloud services, their media store, it all keeps users in.

LLMs don't have that. For at least 90% of users, the model in the back end is completely irrelevant to them. Switching to Kimi won't lose them their iTunes music.

You won’t use a Chinese model, and I won’t either. But both of our bosses have much higher incentives than we do to routinely shop around, and unlike us they can convince an Anthropic sales rep to give them a discount.

Your opinions do not reflect the market and the motivations of its participants.

When the chinese or any other party releases a model that is Opus 5.5 level, there might be a shift to other providers as especially people in the EU are trying to get rid of their dependency on american stuff.

100/m for a subscription is a lot of money to most people. Like people who don't have a SWE salary.

Consumer AI is going to be paid with privacy, not money. CFOs don't pay 2k iPhones for their employees.

Competition is the only reason you can pay 100$ a month for like 4000$ worth of API credits.

That's a great way of putting it and something I have come to realize as well. I'd rather just pay for the frontier model if I trust that it is indeed at the frontier and works reliably. Same with the iphone.

> Why would i bother? I pay $100/m for the frontier model, understanding it's marginally better than much cheaper alternatives. I get so much value from the product it's not worth shopping around

Wait until you are forced onto an enterprise plan...

"I am spending a provably untenable amount for a services that will have to degrade to meet my price point, the cheaper options and processes are beneath me"

Deepseek has been a game changer for me. I don't use them right now but the model is very capable.

What a lot of people don't seem to realize is how much of the edge of the 'frontier' is in the tooling, the prompts and the harness. The LLMs themselves are already all reasonably close if you keep all the other stuff constant. Sure, one will have an edge in one domain or another. But that edge mostly translates to a degree of performance rather than some absolute. And if you allow for a few more toolcalls, a web search or a longer context then that difference is much smaller than it might seem at first glance.

The calculus may be that DeepSeek (or similar) + human is good enough and that the frontier simply won’t have the growth to get to the next level that differentiates them and justifies the token cost. They get stuck in a race to the bottom which is opposite of what they want.

I have a hard time believing that making “a really good harness” justifies pre-ordering half the world’s GPU compute for years in advance.

That might be true for general consumers, yes. Not so true for software developers though. For developers, the tooling is definitely not a moat, and that means the models are what must differentiate.

Claude code is very polished, but I can get the same jobs done using Cursor CLI or any number of open-source coding harnesses (I like Pi, personally). Most plugins/skills/mcps are portable between harnesses with minimal or no effort.

I use the same skills, rules, and MCPs across claude code, cursor, and pi. My tools/plugins aren't identical between harnesses, but porting them from one to another is just a prompt away.

>Anthropic is not in the model business. They were smart and they are building productivity ecosystem business... It's at least a Google Docs killer if not an M365 killer.

Google Workspace revenues are something like $25bn/q, and that's the entire workspace division, (including enterprise gmail sales, so the productivity suite is meaningfully smaller than that figure). It's taken Google a long time to carve that slice out of Microsoft's entrenched market position. It's not small money, but they'd basically be going to war with two of the largest and most profitable companies in history on their turf.

I don't think they can credibly take that business plan to the market and say it'll be worth $1.75tn. They're all-or-nothing on being in the model business, and everything else is superfluous.

>claude code

The same one that leaked itself a few months ago?

I mean, a lot of people before them were "not in the model business" (Cursor et al) and it seems they got disintermediated by the model companies. Step 1 is still to have the best model, or hope for model progress to taper out.

i think when people say today’s prices they do not mean subscription tier, they mean the enterprise paying 10M+

as a person, i would absolutely not pay more than $1K for ai, because at that point i can just get myself a bartening job

What's it like co-editing a combo document/workflow with multiple coworkers and an AI agent on Amazon, Google, Microsoft, or Meta?

Like if you want to design a survey you can send out, and collaboratively edit and test the survey, ask for changes in layout, wording, etc, and design the statistical analysis to be used, before sending the survey out where end users can fill it out right in the doc, while internal users/admins can ask the agent about its results?

Honest question -- that all works pretty well in claude. Amazon, Google, Microsoft, Meta all pretty similar?

Lots of people seem to have pretty short memories (or are too young to have seen it), but the flight from "local service" to "cloud service" was intense.

Companies don't mind paying extra to have a another company that does nothing but said thing do said thing for them. It even actually often ends up being cheaper/more profitable.

I don't think AI will be able to sustain cloud compute-size margins. Inference is a commodity and value-adds can't be sold for anything. Why pay for an advanced harness when you can just ask the model to build one for you?

> Similar to how “the Chinese” haven’t undercut on cloud providers’ CPU offerings

No Western company will be allowed to choose a Chinese cloud provider by their legal department or by their stakeholders. Which leaves only the US cloud providers as viable sources (as us, Europeans, have been too busy building our revenues based on turism and other such non-tech thingies instead of building real stuff).

>The operational engineering that goes into serving a frontier model is _highly_ nontrivial, it is not commodifiable. This isn’t a SPA application.

I don't really see it frankly. Admittedly I never tried it (I only played a bit with local models) but conceptually how is it much harder than all the other cloud services that already exist? My educated guess is that building a competitor to Netflix, Youtube or Dropbox seem technically a lot harder to get running than offering inference on demand for instance.

It seems that the main barrier for competing in that market at the moment is getting your hands on the GPUs/RAM and a datacenter to put them in.

idk what is this you people do, but most of my day is not spent on writing code. most of my day is spent in meetings abt budgets, timelines, evaluations, compliance and marketing. i don’t think ai is going to automate corporate circus anytime soon

Pretty much. That's why Dario & co are always so big on 'what we have that isn't released is so much better than what you've seen from us'.

The point in mentioning the Chinese is that they are doing this today. In Europe there is Mistral but I think they're too far behind the frontier to really matter but in time they may catch up.

Open models by other providers are not subsidized, and anyone can spin up a competitor. That's the price ceiling.

what you mean why? most companies with spare cash payed 10-50x what they actually need in tokens, just to see how it goes. it’s useful, it’s not that useful, so while the price per token can increase, everyone is getting smarter about spend. and even price has a limit, you can’t make ai more expensive than a human, because then everyone would just hire humans.

> I always wonder when people chime in with Chinese models what they are actually doing with those

In my case: robotics, complex document analysis (typically 800K context).

I use Opus 5.5 daily at work and a mix of GLM 5.3 (Flash or not), Deepseek v4 (flash or pro) at home and while Opus 5.5 is more capable, in my opinion it matters for less than 5% of the work i do. These chinese models are simply good enough for the vast majority of tasks I run and when they're not I use Astra or Opus 5.5. Yes, leaving Opus 5.5 on all the time is comfortable when you pay the 90€ plan but it's just not necessary.

Yeah, I spent the last week with Deepseek Flash 4.1-1017, Kimi 3, and GLM 5.3 (and 5.3-flash). None of them were literally useless but also none of them were capable of making progress on the main project I am working on where Opus, Sol, and even Gemini Flash have all done very good work. Benchmarks might have you believe open models are only barely behind the big US labs in intelligence, but it feels much worse in practice.

I am running GLM 5.3 Flash locally at home for coding and from my point of view it is indistinguishable from the frontier models of 6 months ago. It can handle almost every task I throw at it.

i reduced my claude expenditure in 80% using other models.

I don’t know why you’re getting downvoted. One of my coworkers who was constantly touting Chinese models a few months ago now only uses opus 5.5. It is a huge difference.