> hopefully this quells some of the "AI is unprofitable everyone is spending a dollar to make fifty cents" arguments.

I think everyone has known that inference is profitable overall for the major providers for a while (I'm sure not all user accounts, but in aggregate).

The major providers are unprofitable after you add in the training and research costs and other overheads.

This is why the open source models are such a threat if they become good enough - it will destroy the business model of just providing raw chat/intelligence.

Arguably that's also why a few of these providers are trying to diversify into whole-platform approaches rather than just relying on having their whole thing be a great model.

A lot of people have already replied to this point but I think I want to distill the problems a little more directly:

- If inference is this profitable, is it profitable enough to offset the enormous training costs, bot those already incurred and those in the future?

- If inference is profitable, can it cover other capital and operational expenditures the hyperscalers and neoclouds are committing to, on the timelines they're committing to, to scale out inference?

- If inference is profitable we can measure if the price/earnings ratio of current valuations is sane. Is it?

- If inference is profitable, is it - given the answers to the questions above - profitable enough to make the risks of lenders of the huge amount of debt companies are taking on to scale, rational? It's one thing to say as a hyperscaler "we can make this pay", it's another to say "I'm going to lend them the money to do that and I'm going to get my money back on the timescales we agreed, and with an ROI to boot".

- Finally, if inference is profitable why is Nvidia specifically lending money to companies to buy Nvidia cards to do that inference, and then hitting the credit default swap markets to cover the risk of never getting that money back, and doing so at ever increasing cost, when they could just go build a profitable inference business themselves?

I agree seeing the numbers is good. It does suggest profitability in some segment of the market. However I still have a lot of questions.

I can't read the original Chinese text, but if it's revenue, then you still can't say anything about profitability. Housing and powering one GPU should be way less than $38 per hour, so there should be a profit on inference. If that profit can cover training new models we don't know, but it would appear to not be the case.

For the past year or two I think it has generally been agreed that inference is profitable, but the LLM business as a whole is not.

> Amortized cost of owning a DC full of GPUs is far less than $38/card/hr

what do you think it is, out of curiosity?

I don't think i'm paying more than Deepseek price, ever

Until the Chinese stuff came online I was completely convinced this is just a bubble waiting to pop and nothing will come of this tech due to its absurd costs. The Chinese stuff changed this picture massively. I'm daily driving DeepSeek 4.1 flash (at API prices) and every now and then I give Opus 5.5 a try and don't notice anything missing for me.

My workflow is one where I am heavily involved in the process and I don't like giving the agent too much autonomy + I need to follow along. Opus 5.5 speak is still cause aneurysms for my brain. It also has this vibe to its prose... like it's looking down on me. This is Prolly my bias speaking though.

The railroads went bust, twice, not because there wasn't profit available, but because investment outpaced profit and wasn't sustainable.

AI as a business will be 10x larger than it is today within the next 20 years.

Present valuations are the bubble, not the business / industry. The exact same thing was true about the Internet bubble circa the late 1990s, except this is a lot more important than the Internet and the global economy is drastically larger.

Productivity is how you expand economy and wealth. It's how China can go from being one of the world's most backwards agricultural nations, unable to feed itself properly, to gradually becoming an agriculture juggernaut. AI is a hyper productivity tool, saving people time and bolstering capability are two of the primary productivity concepts.

We're not even out of the first inning yet.

With this new data at hand, I have to wonder how profitable would it be for the investors.

The cost of amortising a datcenter, the gpus, the engineering costs attached all to produce inference.

At how much percentage of its inference would it be profitable (if it does), how much returns would it make over the long term and what would be the profit of those inferences/ profit-revenue and how that compares to other industries.

Are there any statistics which show this. Also how efficient token inference has gotten over time?

Those arguments will go away when spending on AI infrastructure is no longer some multiple of gross annual software sales revenue, yeah?

> Amortized cost of owning a DC full of GPUs is far less than $38/card/hr

You missed the bit where they used the word "revenue".

I suspect if you factor in all the various overheads of running datacentres the actual value is lower. Revenue is vanity, profit is sanity as the old saying goes.

However the Chinese do have the benefit that China are pushing hard on renewable energies. So at least that major part of the equation is probably far lower than your average US/EU site.

You're looking for comfort in the wrong place.

From microns S10: BITS shipped and net revenue.

Q1 '20 - $0.10 profit per gb shipped (ram and storage)

Q1 '23 - -$0.024 (cyclic ram market)

Q1 '25 - $0.131 this looks normal were on the upswing.

Q4 '26 - $1.136 Were all fucked.

If you look at a chart of S3 prices we have all been living in the "Happy Time" where the cost to run our ever growing software business has gone down. Githb says 2025 billion commits - 14 billion this year, and 2x the actions run. That means AI driven software growth is 2x to 14x.

But those ram/storage prices are going to hit the operational budget of EVERYONE soon.

Can your average sas survive a 4x price increase on its AWS/Azure bill? How about 10x (cause thats what the margins say the premium is on ram).

If the circular financing is a concern, the circular firing squad should be a bigger one. Your paying for tokens to build more software to pay more to run your software is not a sustainable cycle. People are starting to use the words CapEx again - and it's a good thing.

you actually have no idea what the cost is

Today. As new supply comes online and demand levels off, that 38$ will shrink. Absent legal monopolies, the margin will shrink to a sliver more similar to any other service. That is the danger of open source models. They destroy monopolies and shrink margins.

We still need to know if that includes purchase and maintenance costs, which are substantial. And then of course all other costs associated with it.

You also have to realize deepseeks running costs are immensely lower than say any us lab by orders of magnitude.

Chinas grid is almost entirely renewable energy with strong investments in it. In the us not the case, some dcs are running on gas.

Also DeepSeek was smart. They strategically rolled out model sizes they could support even if the intelligence wasn't the world's best for that week in human history. Then they have incrementally and cheaply improved models since.

Not to mention they also built trust with open weight models rather then fear mongering. They are in the top 3 of the smartest most viable LLM labs out there.

Your estimation feels too high. It implies >1000% ROI on a GPU card (= $300k gross annual profit on a $30k card).

As a reference point, Z.ai is operating at 26% gross margin as of 1H26. So assuming ~90% gross margin is likely a bit off.

How is the math so good for data centres but so bad for home gear? A Mac mini capable of running these models ends up at around 10 years of subscription costs.

I think we need to collectively get a grip on statements like this.

There is no reason to think inference can or will remain interestingly profitable even if you assume pure-play inference providers who don't do any model training. Serving an LLM isn't child's play but it's also not so hard as to count as a moat. Inferencing is:

1. Selling a largely undifferentiated product.

2. Into a highly competitive market.

3. To customers who are primarily sensitive to price.

Industries like this have always existed and in the modern era tend to have margins of about 3% to 5%. Any level of margin higher than that gets competed into the ground.

If you want to claim inference is highly profitable in general, you're going to have to answer why it won't end up like supermarkets or airlines, both of which have bigger moats than inference.

The other problem with this analysis is obviously that you can't just stop training models. Even if you don't care about improving capability or performance, the knowledge in their weights gets stale. Cost of inferencing has to take into account the ongoing cost of training, you can't cleanly split one from the other.

Accounting for initial acquisition cost and typical depreciation/failure rate? I assume they already consider colo price, electricity, etc

Fascinating site. Thanks for linking. However note if you change to "open router" pricing in some cases even B200 which is ~2y old has negative margin (and AFAIK ppl are using longer amortization than 2y - I do not know how cost are calculated here though).

Why isn't ordered by profit % per GPU? It' surely the most important indicator

Or now, Anthropic will distill the knowledge from this engine and optimize their inference.

No, a cluster consists of 4x SM100, and $38 profit is per card, so after 10 months you already have $1.0 million.

I haven't accounted for 7 hours of peak-pricing per workday.

Of course, you won't have 100% utilization.

For four cards at 100% utilization it's $4712 per workday and $3648 per day on the weekends and holidays.

This is per card pricing, not per cluster pricing, so I think you’re off by an OOM or 2

> I think they do not believe that big corporations in the US or Europe would actually use chinese models in their business operations due to security, regulatory and/or business continuity concerns.

They'd be wrong. Especially about Europe. Many companies are in a bind about this: they can't upload their confidential files to a service provider but they do want AI. So right now they end up with a choice: do nothing, breach their contracts, use a service provider that pinky promises not to share their data (which we have already seen they will violate) and run local models.

The latter is the reason you can't buy powerful GPUs for reasonable prices because a local model (GLM 5.3, DS 4, etc) + some hardware is all you need to solve the problem in a way that avoids the legal issues.

> I think they do not believe that big corporations in the US or Europe would actually use chinese models in their business operations due to security, regulatory and/or business continuity concerns.

Quite failure by someone to assume that EU businesses would use US over Chinese, based on the current political atmosphere. US causes even bigger concerns than Chinese labs.

>I think they do not believe that big corporations in the US or Europe would actually use chinese models in their business operations due to security, regulatory and/or business continuity concerns.

Correct. And they're likely very correct in the belief.

When people on the internet claim that frontier labs are done for because Chinese open-weight models exist, they really don't appreciate that the people spending the money do not want to hear "we can just run it cheaper ourselves!" Any software engineer with even a few years of experience knows that phrase is absurd, and executives are simply not interested in the risks associated with such an endeavour when they can just buy the solution outright.

Consumers will have access to open-weight models just like how consumers have access to Linux. But you can just go look at the market share of operating systems to see why something being free and openly available doesn't matter to executives who just wants to get things done and is willing to pay a premium to a trusted business to do the work for them.

And none of that even touches the fact that these frontier labs have only ever been in-front of open-weight alternatives. Again, if you're an executive, and you're questioning if you should pay a premium for a better version of a tool which you presume is important for your company's ability to compete, then the answer is immediately obvious: pay the premium or risk falling behind competition who is willing to pay the premium. And when you're talking about global companies who can pay these bills without it being significant enough to even show up on a quarterly financial statement, then it's an obvious sell.

> they do not believe that big corporations in the US or Europe would actually use chinese models

This is exactly right. The only AI that matters from an investor perspective is the AI that is built into your phone, your car, ships with Windows and Mac, and built into legal and business tools like Lexus Nexus and Salesforce.

Everything else is hobby.

Many big investors may be too deep into this, so they prefer to extend and pretend rather than cut their (potentially massive) losses.

So, they choose to believe, lacking a better strategy. They will also try to lobby their belief into existence.

The US will move to ban it for most major businesses in some capacity.

You're entire company would become ineligible for gov contracts if you have use DeepSeek. Something like that.

If you believe those investor, I assume you mortgage and bet all your wealth on it. SpaceX? Anthropic? Chinese has talents, electricity and Huawei. I really believe those investors are tulip blind to pile money into those western AI. China even now world number 3 ssd+ram ahead of SK Hynix. So armageddon of RAM SSD LLM model at dirt cheap price. And as time progress, large ram and small gpu will be built into handphone (which again dominate by Chinese).

> the relevance that investors assign

I think there is also a perhaps the need to assign air-quotes to "investors" when talking about the US markets.

There's too much fluff and noise there. Too much (over)leveraged trading. Too much speculation.

What the real investors think is hidden behind the actions of the speculative traders.

> I think they do not believe that big corporations in the US or Europe would actually use chinese models

The Chinese models are open-weight. You can download them and run them on your own servers. Or you can outsource to Mistral, Privatemode or any of the many other US/EU inference hosters.

So the "security, regulatory and/or business continuity concerns" arguments are effectively non-existent. And whatever argument does possibly remain, also applies equally to Anthropic, OpenAI & co.

There were pretty specific rumors about IPOs happening early this year, up to apparent insider bets on prediction platforms. Don't you think the apparent delays in IPOs could have something to do with the open weights/open source competition? The monetary conditions (interest rates,...) were better early in the year, arguably predictably so, so what would be your alternative explanation why they chose to postpone?

Seems a toss up whether Europe trusts US or China more.

This isn't true, especially not with the inference engine becoming public, now you can build your on models on the same or similar architecture and ship massive gains on top of these stacks.

I think people don't realize the Salesforce SAP Palantir would love to ship these models, and they have/are shipping them.

The threat of exfilteration of data via these models is high but it's nowhere near as high as the margins.

For instance if you could cut your ai input bill by over 90% would you do it? Can you put some of that extra margin say 10-20% to make it more secure.

I know of atleast 10 fortune 500 companies that are entirely planning to or have moved to Chinese models for a large portion of their work.

It's insane people in finance are delusional enough to think that won't happen.

Maybe Govt will try to de-risk with bans on using open models but I don't think that's going to work at scale, people will literally spawn labs to launder DeepSeek/GLM models into a US model by some novel training etc approach or something.

I still don't think that it's as much a kami-kaze rather it's an attempt to normalize the world around the Chinese stack.

If DeepInfer is supported by everyone, it will be much easier to export stuff aka hardware that only supports it for now or is optimized for it.

Assume china starts selling hyper optimized Asics for inference at some point.

This feels like a race to the bottom as soon as possible because they have much lower costs, if there is a Wipeout in AI Labs in US chinese labs are largely insulated, and they can hopefully buy Nvidia chips for cheap if that does happen.

> I think they do not believe that big corporations in the US or Europe would actually use chinese models in their business operations due to security, regulatory and/or business continuity concerns.

European business users (+ consumers) can access GLM-5.3 via Mistral.ai hosting. (with GDPR)

https://docs.mistral.ai/models/zai-glm-5-3

https://legal.mistral.ai/terms/get-started/

If they are open sourced, it means any random data center could host it domestically and AI becomes a utility.

Europe is going to be just as worried about using US models for large parts of government related functions and they're going to feel alot better about open source self-hosted chinese models than US models which can literally get turned off on a whim by Trump. They'd prefere home grown models if they can, but if they can't they'll go self-hosted and that'll be Chinese at this rate.

> I think they do not believe that big corporations in the US or Europe would actually use chinese models in their business operations due to security, regulatory and/or business continuity concerns.

Well they are wrong, I'm at a large EU startup and they are currently considering Deepseek as their default model, Opus/Fable would still be available for complex tasks.

There's very little understanding of business operations on HN, especially outside the tech echo chamber. The "software will eat the world" mindset.

Based on HN comments, every company would be running Gentoo and developing line of business applications in Haskell or OCaml.

"valuations" are irrational and make dot com _valuations_ look like a miniature caricatures

"the market can stay irrational longer than you can remain solvent"

I am of course eager to pay needlessly more to support poor Anthropic/OpenAI.

Well, what can I say? If you play dirty, then so can everyone else. The way I see it, consumers are the real winners here. Just another variation of classic competition. If such open models didn't exist, then make no mistake, OpenAI and Anthropic would be making sure you are bleeding dry with your wallet without much innovation.

> It is clear from the beginning that the Chinese Open Model are a proxy for economic warfare.

I mean, what would you expect? Like USA isn't openly planning to weaponize their current AI primacy [1, 2].

[1] https://www.whitehouse.gov/presidential-actions/2025/07/prom...

[2] https://www.whitehouse.gov/wp-content/uploads/2025/07/Americ...

No, that isn't going to happen. The US can't afford to lose the AI race and US companies that can't compete will get thrown under the bus and replaced with those that can.

The DHS order only bans buying DeepSeek products directly (e.g. doing inference on their China hosted servers). It does not apply to open weights at all. And DeepSeek just gave their response to the DHS order.

Yeah, but they can't ban the rest of the world from using it :)

Isn’t this a ban on it being used by the state, rather than people/civilians?

From government and regulated places perhaps but a large portion will not care

Thats like sticking your head in the sand :)

Considering how absurdly far ahead of everyone else SpaceX is, 2T seems low, and is more of a testament to how immature the space market itself is at this time.

If SpaceX was reset to 2016, they would still be ahead of everyone else even today. Their deployed capabilities today are ahead of the plans of most of their competitors.

Today the closed-source giants maintain a substantial cost-per-token advantage due to proprietary inference tricks like speculative decoding kernels, ultra-optimized FlashAttention derivatives, multi-GPU pipeline orchestration and dynamic prefix-caching algorithms.

I'm not sure whether the DeepSeek's open-sourced inference engine does all of these, but it appears that plenty of inference tricks have just been revealed for the open source community to learn from and start improving upon.

When the model is fungible and the switching cost is near zero (business logic lies in the harness), then there really isn't a moat for models.

What moat?

In theory, this can allow a small US lab to compete with O/A/G and take marketshare from them with small open models deployed centrally or (and this is think is the real moat) on-prem for enterprises.

Their success is subsidized heavily by distillation, and even then, the Big Labs are no more than an OOM larger.

Not really. Anthropic’s headcount isn’t huge, and only a subset of their employees are working on the frontier. It’s also easier when you’re a follower.

It doesn't matter anyway. Weights aren't IP so the licenses are bunk.

Some open model might be Linux, none of them will be Chinese though.

If the Chinese government has anything to say about it, which in China they typically do, they won't be able to close everything down since the official stance is to be more open than closed.

in terms of raw token throughput and serving users it is, but in terms of R&D and long term lab capabilities, training is still the goose that lays the golden egg

You mean Claude? I have had zero problems with DeepSeek

Why the downvotes, lol?

I do believe Claude Max 5x gives best quality/value at the moment out of all plans/providers.

What's your criteria for AGI? You think it hasn't been achieved with the most recent frontier models?

> On October 16, 2026, the Ministry of Commerce and the Ministry of Science and Technology issued Announcement No. 48 of 2026, publicly solicited opinions on the revision of the China Export Prohibition and Export Restriction Technology Catalogue (hereinafter referred to as the Catalogue), and implemented temporary management measures for pre-export registration of relevant technologies. Recently, the spokesman of the Ministry of Commerce answered questions from reporters on relevant issues.

https://m.bjnews.com.cn/detail/1792137269129878.html

Apparently globally recognized top 10 models (so Deepseek 4.1 Flash and GLM 5.3 Flash) would have fallen under this export registration.

The other benefit is that as more of the stack is open sourced we can finally see just how cheaply these models can be served. Which in turn gives us insight into what the frontier labs might be doing.

That, and drive down prices everywhere.

You underestimate it.

7% of Chinas total population are members of the CCP. 7%. So at DeepSeek there's a good few handfuls of them.

If you want to understand the impact of that, read up on reporting lines in the CCP.

Not every company needs sign-off from the government for an OSS release, but every OSS release is visible to the government.

I would assume DeepSeek is considered one of China's strategic assets and is therefore heavily policed.

Egypt has been around longer. That's kind of a silly political statement.

This seems like a garbled retelling of the shift from NMC batteries to LFP.

The latter are worse on some dimensions, so much so they weren't considered viable for cars in the West. But incremental improvement saw them take over the Chinese market. A classic "low end disruption" story.

They bought up the patents from A123, an American firm with a Taiwanese American MIT researcher founder. The firm floundered in the global financial crisis, a time that is a big inflection point for China's global domination of green tech.

https://en.wikipedia.org/wiki/A123_Systems

The problem with that theory is that BYD and a few other Chinese companies already for some years have had batteries that are better than anything made in the West, from the point of view of energy density, charging rate, temperature range, cost, so if there is some stuff that is patented and locked by the West, that is the bad stuff.

Living in the EU, I find it concerning that China has advanced so much in battery technology and in the West there are very few positive signs of attempting to keep up with them.

Closing your eyes and imagining that others are not really better, when they are, is not the right strategy to avoid remaining behind.

Which good (EV, I presume) battery provider is from "the West"?

From what I’ve heard from owners (anecdotal) those batteries are still utter garbage.

that doesnt say what cards the reference number used.

I would love to read the details about how datacenters do load managing across such diverse hardware, if in fact they really are plugging in any computer hardware they can get on the market. Like how do they balance and serve an even customer experience across a wide range of hardware... is it by clustering similar hardware together?

Sure, but that size/token is inherent to the model. Part of DeepSeek's hosting advantage, and thus a remaining part of their moat, is migrating KV cache from VRAM to host RAM to nonvolatile storage, and presumably routing requests to where KV cache has migrated. People here on HN say KV cache lifetimes can be 8+ hours with DeepSeek, and I don't know of any other provider that can match that. That infrastructure isn't part of this open-source release.

Yes, the theoretical approach is listed there. But the actual code is what's missing.

Yes, that's why I commented. I'm not sure I understand your point.

Interesting, thanks!

I was referencing that. İt doesn't help deepseek it helps the country.

But how does that benefit the Chinese companies to have other firms host their models and gain profit from the inference, without having to pay the training costs? Training by itself is not profitable at all, only serving the newly trained model (inference) is.

> The real loser in all this is obviously Europe, which has nothing competitive once the Chinese models are gone.

Those models are open source / open weights. Europeans can continue with Chinese models as a base same like Cursor used Kimi K2.5 a base for their composer 2.0 model or SWE-2 from cognition is post-trained Kimi K3

This is not a home/consumer GPU discussion.

But in that vein, you are in the same position as the chinese LLM providers. They cannot get hardware to host so they put the problem in the forefront by offering it open.

In a way, Deepseek is championing your cause.

Make no mistakes.

I did use deepseek via API and it can easily cost 3$ per day for my workload. On the other hand Luna 6 on 20$ subscription from OpenAI cover by usage for whole month. However, `ccusage codex` command suggested that my Luna6 token cost is around ~100$ monthly.

``` or by 5 opencode subscriptions, and have $800 of ds4.1-flash for $200. I'm skeptical that the average dev will manage to burn through the equivalent of $800 of ds4.1-flash if they spend even a bit of time optimizing their token usage. ```

This is my point. For most developer, subscription model (with subsidize) is preferable. And what I'm wondering is, as suggestted by very cheap cost-per-token from DeepInfer, why don't Deepseek persued subscription market as well ? What make subscription not make sense for them to provide ?

While I agree with the debt part, API pricing can also change at any moment. Well, anything can change at any moment, it's still wild west outta here.

Even the Alibaba $6/month plan supplies a reasonable amount of API use, and Qwen 3.8 Max is a killer model for long time frame agentic use.

Literally everything I do is Open Source software and Open Content writing so I am fine with the lack of privacy using direct Chinese APIs. And, for a little more spend, just use US companies like Fireworks.ai or Ollama Cloud for more privacy, or run on-premise locally and get the best grade of privacy.