DeepInfer: High-performance inference framework for DeepSeek models (github.com)

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

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.

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.

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?

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.

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.

> 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.

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.

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

> 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 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.

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.

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.

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

you actually have no idea what the cost is

> 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?

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.

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

But haven't the Chinese (and Deepseek specifically) put a lot of emphasis on running models efficiently (because they've been locked out of being able to obtain certain GPUs by the US)? Whereas the US frontier companies seemed to get huge cash inflows and didn't need to worry about efficiency as much and put more emphasis on frontier performance. So how do you know that this $38/card/hr is about the same that the US frontier companies are making?

>It's really refreshing to put some cold hard numbers on this stuff, hopefully this quells some of the "AI is unprofitable everyone is spending a dollar to make fifty cents" arguments.

Inference has always been profitable. The problem is that this profit shrinks as your model becomes less competitive. The labs are in a constant race to release the best model to maintain inference profit. Deepseek is not training their own models at the scale of OpenAI and Anthropic and mostly sustains off their efforts via distillation.

> It's really refreshing to put some cold hard numbers on this stuff, hopefully this quells some of the "AI is unprofitable everyone is spending a dollar to make fifty cents" arguments. Amortized cost of owning a DC full of GPUs is far less than $38/card/hr

They (well, most of them) are looking at loans and debt (in plain sight and hidden) of AI companies and asking "what will pay for that", not looking at how much a GPU costs and how much tokens you can sell off it.

Just putting a model on a GPU and selling it is profitable, period.

But spending more and more money on building and running infra to build frontier models that solve smaller and smaller piece of the pie might very well be unprofitable.

Models competing with the frontier labs gets more capable every month and more and more of the tasks can be achieved without reaching for the big guns.

We might very well get to the point where AI/LLM is profitable just fine but bubble still bursts and the big frontier labs are in trouble. And frankly that would lead to far more sustainable ecosystem.

I don’t know, I take this as there being a good amount of margin that is going to be suppressed due competition in the very near term. It will be interesting to see how these margins hold up over the next 12 months, I expect them to drop substantially

> "AI is unprofitable everyone is spending a dollar to make fifty cents"

you are just talking about ai inference? Are you assuming ai models themselves fall from the sky.

I still feel like there are some questions there, as in what is the actual amortized cost, plus the ongoing variable costs and even the non-training related overhead.

Plus there is the actual demand side to consider and how does it impact when paid utilization isn't 100%.

Now that the unit economics are laid out in the open, What stops any of the companies or even startups from anywhere in the world with some investment $ and competence from spinning up and offering their own version of the inference engines, undercutting any moat that the current leaders have?

I could even see F1000 companies having their own backstop intelligence for most of the tasks and only use frontier where it is absolutely needed or as much as it is under their IT budget.

Won't it be race to the bottom when they compete on cost? Is regulatory capture the only moat left for the market leaders now?

revenue != profit?

I think you have to take this on a case by case basis - there is no reason to suppose that the training costs and serving costs are the same for all companies for all models and for all price plans. Some may be profitable, others not.

DeepSeek's models are relatively small, and training costs will be correspondingly lower than for larger models. All you have here is a revenue number - no mention of (training + serving) cost, so while they may well be profitable you can't determine that from this number alone (if it can be trusted).

DeepSeek (investor meeting leak) have previously said that they price their models at a point where the GPU is paid for after 10 months, so something doesn't seem to add up here. If they were making $38/hr per card 24x7, then in 10 months they'd be making $270K per card, while the cards (Huawei?) only cost around $20K.

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

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

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.

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.

DeepSeek CEO has previously said that their finance team considers the optimal payback period for hardware purchase to be 10 months. That is, 10 months worth of revenue should be equal to the purchase price of the hardware. This doesn’t cover operational costs such as electricity. This is their method for API pricing.

Source: https://x.com/FuckAnthropic/article/2080268590033621295 (it’s a leaked meeting transcript so take it with a grain of salt)

I still struggle to wrap my head around a $30,000 GPU card. But impressive payback numbers.

Ultimately, more people serving.

- gpu market is starving, demand is higher than supply = leads to high gpu price

- number of openrouter shovel sellers rises = lower token prices

Exactly... I was going to post exactly this and you called it two minutes before me. On openrouter, it is shown that 600T+ tokens are provided per month, so 20T+ per day. Openrouter is not all the world but it seems that something is not right with this 155T figure... Unless everyone is idle everywhere, waiting for demand

From the Github benchmark numbers, one GPU can do about 26k tokens/s when serving 256 users, which is about 2.25B tokens/day. A cluster would need to be close to 70k GPUs to serve 155T tokens/day (which seems doubtful).

OpenRouter routes traffic through many different providers

https://openrouter.ai/deepseek/deepseek-v4.1-flash-1017#prov...

> It may take a minute or two for access for 90.8.226.75 to become effective

Exciting... feel like I'm back in '96, having to wait two minutes for the page to load.

making AI bubble burst (even if it is just frontier labs that got hit) while chinese companies keep making money off cheap AI would be massive economical hit to US

“ Or, are you saying this is a tactical move from a higher level (Chinese gov interests)?”

What country produces all of the ASICs used in contemporary bitcoin mining?

The faster we get to a good enough open model, the sooner it gets burned to hardware, and the sooner the transition to (relatively) low tech, low margin, high volume cards to run it on …

… and those are built in China.

US wants to ban non us models from as many places as possible. This move enables them to get users in more places.

Does that mean US labs are banned from using or even learning from this? Assuming they still want to be used by the government.

They last part is impossible to enforce really without forcing audits on everything.

Not sure why you're being downvoted. It's very clear to me in the order that even the open weights would be forbidden, even served through vllm or sglang etc.

This is also the word inside Defense sector contractors as well. No PRC origin anything - no code, OSS or no, no open weight models, no assistance from them.

No regs set in stone yet, but DoD doesn't need regs - they can tell you what the rule is and that's that.

Thats like sticking your head in the sand :)

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?

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.

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

What moat?

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.

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.

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.

> 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.

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?

> 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.

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.

> 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.

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.

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.

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.

> 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.

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).

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

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.

>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.

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.

lol big corporations are 100% using Chinese models in their business operations. They dgaf about security, speak to anyone in cybersecurity or cyber insurance

The labs pushing this cyber scare is an attempt to cry wolf. The corps don’t give a fuck

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.

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

Isn’t the Muse team just 200 people? I guess you don’t need that much work done.

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

> 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...

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.

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

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

no one has had to pay back any of the loans yet. the valuations right now mean precisely nothing.

if these hold a year after an IPO then we can talk.

Because we are still at the first stage of enshittification

I rewatched The Big Short recently, and all I could think about was AI labs.

I'm betting that China is betting their government can better handle the massive unemployment these will create within the white collar labor force.

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.

Aren't they betting on the valuations like a week from now?

It's possible to be correct that the AI companies have no future, but not be able to time the exact month they collapse.

there's hardly any volume there, so I wouldn't take it seriously. The last transaction was 4 days ago

Best leveraged debt on the US economy ever.

People only use Chinese models because it is opensource.

One very simple reason could be that the US has been very vocal about China being its geopolitical adversary (they may be right).

This really not that hard.

Deepseek started as Liang's side project funded with his hedgefund couch change compute. That set the economics for everyone else in PRC. If deepseek releases for free because Liang doesn't give a shit about making $$$ on his hobby, every other PRC player has to follow. And serendipitously, due to US export controls, PRC tech companies can't dig themselves into ruinous seppuku debt buying compute even if they wanted to - what they can buy, i.e. everything around DC infra including power, they get for fraction of the cost expediently.

TLDR Liang said fuck profits... which means every other PRC competitor had to fuck profits, which ends up fucking profits of western labs so deep in debt due to rampant financialization and can not afford to say fuck profits.

Sometimes a billionaire is fine just burning $$$ on a hobby while remaining a billionaire instead of trillionaire, and it just so happens the side effect might mean preventing others from being trillionaires as well. Bonus others is geopolitical adversary who is hedging economy on trillion $$$ returns.

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

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

Why is that btw?

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.

  First, adhering to the principle of openness and win-win cooperation while boosting innovation-driven development. Xi highlighted the importance of encouraging open-source, openness, collaboration and sharing to facilitate technological innovation, industrial development and scenario-based application of AI.

It’s really hard to vilify the Chinese now on this and make it stick! So far, they seem to be championing openness and freedom on this front, which considering the nature of the technology, is a huge deal!

If they change their stance later, we’ll criticize them the same.

It’s possible they don’t if they fundamentally believe the AI market is going to end up being an exercise in just mass manufacturing tokens, which China is going to easily win because they can generate way more electricity than any country and manufacture almost everything they need to do it domestically.

The models are just a distribution mechanism in that case, and it’s in their interest to get that into as many hands as possible.

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

Why the downvotes, lol?

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

You mean Claude? I have had zero problems with DeepSeek

There are cheaper ZDR providers on OpenRouter.

OpenAI/Anthropic does not train on your data? I'm sure they do

Dax has said time and again they don't quantize the models they offer. Assuming they trust their upstream providers (I don't think believe they run their own inference) and we trust them, what you get with OpenCode Go is exactly what you'd get from the DeepSeek API.

I am personally a very happy OpenCode Go user. $10 month has never felt more useful.

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

Still a vague argument. Why an open-sourced inference engine changes the game, drastically as some claimed?

This assumes the set of tasks we want to use AI for is static. I don't think that assumption is true, it definitely hasn't been true up to this point.

This is a wrong mentality imo. Because we are nowhere near reaching a good quality and feature level on software in terms of the generally known theoretical ceilings.

For example you can use a database and see how it is nowhere near reaching any hardware limit. Or you can use some arbitrary software and see how it fails miserably once you start playing with options.

There is so much room for being better that I don’t think anything is justified for cutting including programmers.

Whether these companies will be able to sustain their business or weather software companies will prioritize paying LLMs instead of programmer I don’t know but the ceiling is way above

You're confusing software development use vs mass population use. And what you're claiming is entirely unsupported by the actual facts.

GPT's pie keeps getting larger, not smaller. Anthropic is growing hyper fast. DeepSeek and Kimi were supposed to put an end to that and the exact opposite has happened. The HN doomsayers are at 100% on being wrong spanning years at this point. It should be embarassing to be so wrong at some point.

The userbase that GPT has is not going to migrate to Chinese models. They don't even know where to find them to use them for the most part. They don't know their names. They don't know the difference between any of the models. They know: Google, GPT, Microsoft, Apple, Android, Samsung, etc.

GPT's foundation and future is as a service model. Anthropic is the frontier player that is potentially heavily threatened by China's models. It will require that Anthropic stay out in front and remain price competitive, and so far they're growing at the fastest pace of any start-up in world history.

How does this change that? You could already run open models with vLLM (note models, not just a single model like this release), which many companies do. DeepSeek’s fork of that will surely improve the self hosting landscape, but I don’t see how it it’s some “nuke” or “kamikaze” as the top comments are implying.

Also note that this is not a complete solution for self hosting. I assume that parts of this will be upstreamed to vLLM which this forked a while ago.

>The vast majority of operations don’t need SOTA anyway.

I agree. But people rarely pay only for the minimum they need.

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

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.

> 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.

That's true and the Chinese vendor market for most things is sufficiently decentralized that individual vendors do act with a lot of self-interest rather than government-interest, but it would strike me that a provider whose key operatives are travel-controlled by the government of the place they live in are not in that category. I imagine having your passports held by the Chinese government assists in alignment.

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

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

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.

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

I have no strong views on whether US or Russia or China are the biggest hackers (and think it's impossible to k ow or judge objectivel) but feel compelled to point out that you seem convinced without evidence and proactively attack any replies contradicting you ("everything else is propaganda").

This is neither good discourse style nor does it help the accuracy of your own worldview to proactively decry and dismiss every contrasting position as propaganda. It's a surefire way to maintain incorrect worldviews and while it may hell your confidence does not make your positions true.

First convert the repo into 20 md files then prompt something like audio overview for very technical senior engineer

False. There are more Antilles (1.4MM) than Slovenian (400k) TLD registrations despite recent registration surges by pusillanimous poltroons; further, Sports Illustrated has nothing to do with most businesses.

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?

more like: Your account will get automatically flagged for output containing the word nmap (technically illegal software in germany), blackholed killing your unrelated prod app, then the german support will take 4.5 weeks to respond inaccurately; you will complain and this will next get escalated to an “architect” (common role in german it, usually university graduates who have never written a line of prod code in their life, and never will) who unfortunately happens to be only working 2 days a week and has 10 days of holidays booked and then once the fax has been approved and stamped by 16 departments someone will upload the correct line to an excel spreadsheet on sharepoint which will trigger a cronjob to add your email to a servicenow ticket about an unrelated issue.

At least, this is how the three separate business i’ve migrated off the dumpster fire that’s hetzner found out what they’ve been paying for.

Avoid them like the goddamn plague. I’d rather run production off my fucking iphone than deal with that rank bunch of incompetent, arrogant useless tossers again.

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

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