Competition drives prices down. We tried Kimi and GLM models after hitting 'cybersecurity' blocks on OpenAI and Anthropic. We found them to be much cheaper to access and helped us complete our work without 'cybersecurity' refusals blocking us.
They're not competing against each other, they're competing against free/local/open weight models that they're also lobbying very hard to regulate/ban.
As much as I hear this sentiment, and as much as all the smart software people around me echo the desire for this sentiment to be true; it feels like people say these things, but then at the same time say - well, Claude/Chatgpt is still marginally better for me now. But clearly they are catching up so maybe one day!!!
In that sense, I'm really not sure if Kimi/chinese models are really competing against Anthropic/OAI? Not to mention that for individuals, the subscription model is extremely different, and Kimi subs are (I think) worse value, and chinese API's maybe marginally end up cheaper than using american subscriptions.
The profit comes from the enterprise customers who have to pay for usage at API rates. Not the subsidized subscriptions, which both OpenAI have designed to be unavailable for large enterprises by withholding IAM and auditing features they need except on the usage based entperise plans.
As an individual I'm still paying for a ChatGPT Plus plan since $20/mo gives me a lot of value and I use many multiples of $20 at API rates per month. But in terms of capabilities, I am entirely satisfied when using my Opencode Go plan with open models like Deepseek Flash/Pro, Kimi and MiMo.
At this point keeping my ChatGPT subscription isn't because of a significant capability gap compared to open models. It's 90% about the value of the subscription priced well below API rates (10% the fun of trying latest models).
If OpenAI/Anthropic dropped their cheap subscriptions or drastically undermined the value proposition I'd be happy using open models exclusively and currently barely notice a difference switching back and forth for the vast majority of tasks.
I'm expecting the subs to increasingly provide less value as the US labs IPO and follow the SV pattern of moving from marketshare capture mode to profit extraction mode. Thankfully I'm already comfortable using OpenCode with open models and avoid lock-in to any specific harness so would be a very soft landing.
Openrouter stats seem to indicate that there is a lot of demand for ‘good enough at a good price.’ After all, not everyone is decompiling PS5 games. In fact, most people are just speedrunning notetaking apps in JavaScript frameworks.
But, Openrouter stats can only measure the usage they see. I'm not sure it is really all that representative of total LLM usage across the market.
I think that's the difference here -- OpenAI and Anthropic are competing more with each other (and Google Gemini) than they are with open weight models. While people on HN are familiar with Kimi and GLM models, I'm not sure how many casual users of LLMs know about them. I think the competitive risk open weight models pose to OpenAI and Anthropic is much further in the future.
There is a lot of usage that's only known by Anthropic or OpenAI that's not visible to the rest of the world.
> most people are just speedrunning notetaking apps in JavaScript frameworks.
I feel seen. But seriously, my app, with my preferences, written by a semi-trustworthy entity that isn’t trying to shove ads down my throat. That’s really great.
Borrowing money is not the only way to go into debt. They reportedly made binding and non-cancelable commitments. Once those commitments come due then they will be indebted to their counterparties until they pay up or the courts sort it out.
This is comparable to how, if I sign a 12 month non-cancelable lease and then don’t pay the rent, I will be in debt to my landlord.
That's effectively borrowing money with a different name though. The only way to go into debt is to sign an agreement to pay someone money in the future, regardless of the terms used to describe it.
No, it's really not. Do you have a war chest to drag out a legal battle and the ability to bankrupt your landlord by withholding payment? The rules are different at this scale, and "you can't enforce a contract if the attempt will bankrupt you" is only one of the possible games.
Here's another one: pay a supplier to cripple their deployment. Suddenly it's the neocloud's fault for failing to deliver, not your fault for failing to pay.
Did we all forget how Sam Altman turned a non-profit into a for-profit? Do we think he'll bring less sociopathy to this battle?
Apple is worth more than all the AI companies combined. You think LLMs are less valuable in the global market than slightly updated iPhone models every year?
LLMs? No. LLM companies? Yes. It's just math, data centers, and mass copyright infringement. The only scarcity is Nvidia, who just passed Apple in market cap, somewhat proving your point but not in the way you meant it.
If you consider Nvidia an LLM company, and I do at this point given how heavily they've invested in frontier labs and where much of their revenue comes from, they alone are worth slightly more than Apple.
Maybe not, but it is probably less likely that Apple will face stiff competition from freely available alternatives, given that AOSP has been a thing for the iPhone’s entire lifespan and nonproprietary Android phones have failed to capture significant market share.
Their costs could drop, competition could dry up, or they may only be unprofitable in the first place because they’re choosing to invest more money in growing their business than they’re currently making, a practice which need not and will not continue indefinitely.
It’s not at all unusual for companies to run at a loss for several years before turning a profit.
The government could bail them out, that's the most likely scenario if they don't implode IMO.
They could also continue to grow at the rates they had last year. I don't know where those customers would come from, but if they did pull that off OpenAI and Anthropic could sustain propping up the circular financing going around the AI industry.
Goverment bailing they out will be using people tax to rescue the companies that absorbed most of the 801k money so they can get people out of work... it's the ultimate slap to the face... I would strongly resist this, I would vote if they implode to give to whoever owned shares via 801k their money, to people and let them (openai, anthropic, et al) eat shit... I would love to buy one of those gpus to run an open model at discount price...
Completely agree it will be a huge slap in the face to the public, unfortunately that's a common solution.
When everyone is afraid the market will collapse, or more likely distracted because the market collapsed just enough to hurt the average person, the government will bail them out and either people won't notice or will welcome it as a solution to huge levels of fear.
Voting isn't usually much of a solution after the fact, the political machine doesn't allow for candidates that would fix anything to have a chance at any office that matters. There's a reason our politicians today all seem to have roughly the same goals and all seem to become extremely wealthy while in office.
A bailout would be a slap in the face when the correct solution is to nationalize them. If they are too big to fail then the government gets a controlling share of equity and we finally start the sovereign wealth fund we should have started decades ago.
I live in a state that votes heavily in one direction, most states are the same. Its fine for me to vote, but the winner of a majority of states is pretty well decided before a ballot is cast (by strong majority, I'm not claiming fraud).
If everyone who threw up their hands actually voted, the outcome would change. See also, the elections musk tried to buy in the last 18 months or so since he doged up the place.
Stuff like this that defies all logic happens all the time. Amazon operated at a massive loss first it’s first decade and still IPO’d years before they posted a profit.
I am technically not saying that. I am taking off from the parent comment and speculating.
Their data from September is leaked via IPO preview, and October is not yet complete, so how can someone even project that?
// I am a techie, not a finance guy, so apart from public chatter, no inside knowledge.
And I do feel their IPO is being rushed before anyone has enough information to make up their mind. The VCs or private equity have given them a lot of money and want to cash out. When the revenue does slow down, any investor should pause to take a look at whether they can right the ship, or has the landscape itself changed.
Their revenue taps, product releases and marketing campaigns were all strategic: to be able to show nice numbers in the financials.
Once they stop playing games with quality of output of these models and revenue's shrink - they will be in deep deep trouble. They are hoping they will conjure up a break through with the IPO proceeds.
I don't have the numbers in front of me but could have sworn I had seen quarterly revenue growth rates dropping drastically over the last year or so. Still a strong growth rate, but much lower than prior growth rates.
Edit: Data because silent down votes are useless, Anthropic Q2 2026 grew 147% while Q3 grew 45%
Yes but its growth rates with respect to expenditures, and in particular capital expenditures (R&D etc), which we have no idea about.
Gross margins here are irrelevant: if they can't charge a high enough price to cover all expenditures and rely on cash raised via fund-raises, they are not financially healthy.
I think your question is rhetorical, in which case I agree with the sentiment.
The US absolutely could (and very well might) make Chinese models illegal, possibly open weight models in general, but all that will do is keep them out of the hands of law abiding US citizens.
Law breakers in the US and the entire rest of the world would still have access, and nothing in the US will be any safer. People who think regulation will solve the problem should be embarrassed by their ignorance.
I don't get why people think Chinese models could bring openai and anthropic to their knees. There are zero signs of such a scenario. In fact the market is still growing at a gigantic pace. Quality of American models is still AAA. Chinese will have their place. But they are unmatched by their American rivals.
> But they are unmatched by their American rivals.
Kimi K3.1 is top 5 on AA and half the price of Opus 5.5, Qwen4 is coming soon. There is near zero difference for most tasks, so it matters less today and even less as time goes on. Price is now the top concern for a reason.
OpenRouter and other inference providers won’t be able to openly host the models, for starters. That should cut most (or at least a significant share of) non-local inference, and local inference only works with smaller, less capable models, plus the hardware is hella expensive when you don’t have scale to saturate.
In addition, most companies don’t want to be in a position where a disgruntled employee can report them to authorities and rack up massive fines or worse, even if it’s hard to tell from the outside.
Banning open weight models in the US would provide the rest of the world with a huge business advantage since these models are much more cost-effective than US models.
> If you are a legally operating company in the US just saying it's banned is enough, the threat of enforcement is enough pressure.
Is there precedent for the US actually "banning" software? I know for example the Kaspersky ban and tiktok ban, but both of those were either banning the sales or banning the service. IIRC neither made the actual software illegal.
What is an example of a specific software being illegal to possess? What is an example of a specific software being illegal to use if it is not used to break other laws (copyright, etc.) or talks to banned service providers?
Closest thing I can think of is restricting the export of cryptography in the 90's, but that was not even close to what you are suggesting.
Correct it’s 7% chance on kalshi, and those 7% of people are just wrong. I know everyone likes to larp that the US is some kind of dictatorship, but lol, lmao as they say.
Also on the “how” side so far we have “they will just announce that “they” are banned, hilarious.
I have a harness for my harnesses now. My Hermes manages Codex, CC or Cursor now and interact with me via Slack and PRs on GitHub.
It's nice because I can keep going from anywhere, other team members can see and interject in my Slack threads, and Hermes auto updates its skills with any adjustments that we make so we don't have to ask twice.
A few bucks will realistically last you forever.
I also bought a DGX Spark at launch. I've learned a ton. but it's not usually my daily driver. I do run overnight tasks on it, like "read this book and write up notes on each chapter..."
In retrospect, given the dirt cheap prices of apis and or renting hardware, buying hardware for inference was likely a mistake (but not enough of a mistake to sell it now).
For me, I used Zed Agent with Codex as it has native integration with Zed's ACP so it's pretty convenient.
However, Anthropic is pickier about the harnesses you can use with Claude Code and I noticed discrepancies when using Zed's External Agent feature with Claude Code, so I just use the Claude app. It's fine.
Personally I've found Zed's agent to be much superior to OpenCode. IDK why but it gets better results from the same model on the same task. Using with GLM 5.3 and DeepSeek 4.1.
I'm calling BS. I use Kimi, chatGPT (Astra/Sol), and Claude (Opus 5.5), and Kimi is by far the most expensive in terms of tokens and $/work done. Using OpenCode harness.
yea but I thought they had a massive compute shortage. Something doesn't add up. Imagine there being a beef shortage and the beef producers cut prices because "competition". why?
If the beef producers don't believe beef is a commodity and they want to get you addicted to their specific beef then it makes sense. Right now in these companies life cycles profit doesn't actually matter. The only thing that matters is how much of their beef is being consumed. They want to give people so much free beef they can't even imagine the idea of having a meal that doesn't contain beef.
DeepSeek claims $38 of revenue per GPU, per hour for Flash V4. That is $332,880 per GPU, per year. This matches their DeepInfer numbers: 25,600 output tokens/s per GPU sold at $0.60 per million -> $55 minus some other overhead to arrive at $38.
For the bigger Anthropic and OpenAI models (and associated higher prices) there are many unknown factors. But presumably they also work hard on optimizations, and it's not so crazy to think they had or still have a lot of wiggle room to adjust prices.
It's sad how far down you have to scroll to get the real answer. It's because Anthropic margins on inference are absolutely bonkers and it's easy for them to drop prices by 80% and STILL have positive free cash flow
Yes, but DeepSeek went absolute goblin-mode on optimization and I would not be shocked to discover that OpenAI and Anthropic were behind on that front.
OpenAI and Anthropic have more compute and the best frontier models to throw at the optimization problems. They have published articles on these efforts. For example, Dots, and Anthropic's upcoming "persistent agent" is because of these optimizations. Both companies often publish articles detailing a lot of their technical efforts. They just get buried among the flood of other news.
Sonnet 5.5 is 16x more expensive at $10 per million output tokens. Kimi K3.1 is also $10 per million output tokens, and its number of active parameters is only 6.5x larger than V4.1 Flash (104B compared to 16B). These things are not directly comparable of course, but just to give an idea.
Even if inference of those models needs 16x, 32x, or 48x the number of GPUs compared to V4.1 Flash, still seems like potentially great revenue per GPU.
Given that they open sourced their inference framework, how realistic would be to buy hardware and become a provider on OpenRouter? These numbers are incredible.
That would be like regulating who gets access to the best food on a sinking ship. The USA could allocate capital and attention toward much better things, for the country and for their people. I just hope the next couple administrations are wise enough to not experiment with the idea that ASI can get the country out of this 'fall of Rome' slow motion decline.
> You can’t show ROI to investors by cutting inference costs
You can as long as you can saturate hardware with profit that makes up for training costs (which are also in free fall).
Anthropic is most at risk since they haven't bought any hardware (afaik), however OpenAi is building custom silicon now. The rest of Mag7 are primarily hardware vendors
Any sources on training costs in free fall? My understanding was training costs are proportional to model size. Is it just that compute costs are falling?
> I believe that’s post-training, not pre-training
Aren't AI companies already butting against the limits of usable training data?
Outside of undiscovered techniques, additional costly pre-training will likely yield diminishing returns if you've already scraped the whole internet and destructively scanned as many 20th century books as you could lay your hands on.
Training costs are not in free fall, and not every "post training" process is equal, Everyone does this differently. The costs certainly haven't gone down, to train a new 10T param model from scratch, like at all.
You cannot point at Xiaomis tiny model and compare it to training a 10T param model. The levels of dunning krueger are getting out of hand.
You also arent calculating the salaries of the people involved, of which there's only a few thousand worldwide in your hiring pool.
> You also arent calculating the salaries of the people involved, of which there's only a few thousand worldwide in your hiring pool.
There were probably a few dozen 20 years ago, and the existing pool grew from somewhere. AI (nee machine learning) is not rocket science - but it is a science, so it rewards those who are literate, curious, can hypothesize, experiment and learn from results to gain an intuition. I bet most statistics-adjacent professionals (millions, worldwide) can move laterally into AI research and excel at it. Nevermind AI has been among the most-in-demand undergrad programs since ChatGPT dropped. The only limited pool is for celebrity AI talent.
you haven't cited any examples, nor acknowledged this is how everything in computing has gone, DK is a two way street I suppose
stable diffusion is an excellent example, the price to train today's or yesterday's frontier is the analogy, many more people will be able to train up what we have today, especially as more of the pipeline is open sourced and the knowhow distills
>stable diffusion is an excellent example, the price to train today's or yesterday's frontier is the analogy,
Dont use llms to draft your responses to people, extermely cringe. That explains the llm induced Dunning Krueger.
What examples... There are no public numbers regarding how much it costs to train a 10T model, atleast nothing concrete.
You should just use common sense and think about how you have to fit tensors in memory to train, and the larger the model is the more GPUs you need. Then you could also use common sense to think about how GPUs have gone up in price, not down.
> Dont use llms to draft your responses to people, extermely cringe
You are throwing around unfounded and incorrect accusations. Why? It's a weak ass rebuttal mate
common sense is using things like history to guide expectations instead just saying things without evidence that are contrary to the patterns in computing history
I think that the frontier labs knew what they were doing when they invested in cybersecurity starting in the spring (or maybe earlier). They knew it would create hype. They knew it fit with their narrative that they are building a dangerous thing (something they believe because it’s fun to believe in doom, because it’s a great story for investors, and because it’s not inconsistent with a moat - in fact it makes a moat via regulatory capture).
Regulate the bad actors, not the neutral tools. Lockpicks aren't illegal, but breaking and entering is. Imagine if MasterLock used a lockpick to break into a government building and steal stuff, then tried to lobby to make lockpicks illegal. That's what OpenAI is doing, and you're falling for it.
In the hands of 4th party system Progressive Republicans and 5th party system New Deal Democrats, government was an extremely powerful force for trust-busting.
In the hands of 6th party system Republicans, namely Ronald Reagan and Robert Bork, the government worked hard to usher in a a new era of Robber Barons. We live in the culmination of this project.
History says the result depends on the hands, not the tool. 6th party system Democrats have largely failed to undo the Reagan/Bork policies, but that's an act of omission not comission and even then Lina Khan was a big step in the right direction. This issue was on the ballot in 2024 and will be again in 2028. Hopefully we make better choices in 2028.
It is pretty funny you refer to these periods because they are the quintessential example of big business pushing for regulations to push out their competitors and become monopolies. And the "progressives" enthusiastically supported such monopolization.
If you are going to limit the 5th Party system to the first half of the long 1930s-1990s period of overlapping realignments (which is pretty reasonable), then the 6th Party system should be the second half of that period, and the period from about the end of the Clinton Administration on should be the 7th Party system; you can call now the 6th Party system if you call the entire long realignment period the 5th (which makes some sense, too, though there is a pretty significant difference between the New Deal dominant period and the Civil Rights dominant period of the realignment), but to split it somewhere in the middle but then lump the period after with it makes little sense.
If you want to call Obama and Biden 7th party system, I don't really agree but I also don't think it matters for my argument: we can just re-label them and move on.
If you take away all regulation you’d probably end up with a monopoly anyway, as companies would no longer be prevented from merging with their competitors or using underhand tactics to gain market share.
It's wild how businessmen will pitch you the 10 ways they will abuse the rules of the game to obtain a monopoly and extract rent, then they will turn around and say that the game rules are perfect and our only failing is not following them hard enough.
That's wrong, but why let reality get in the way of a pithy ideology?
Government regulations exist to regulate activity, to prevent, punish, and/or remediate harms.
Environmental regulations exist because businesses wouldn't clean up their own messes. Financial regulations exist almost exclusively because of harms and abuses committed by members of the financial industry. Building codes are updated to reflect lessons learned about people died, to prevent more people from dying the same way.
If AI regulations are passed, it will be because the AI industry decided not to self-regulate itself and stop illegally hacking computer systems, generating CSAM, or various other illegal activities that should easily be preventable if their systems were actually intelligent.
Also there are real trade offs. Food would be cheaper if, for instance, it were legal to cut it with any old cheap substance regardless of health effects.
But I totally agree that some regulations are dumb or are regulatory capture monopoly schemes.
Unfortunately like many other things there is no absolute simple rule to save us from thinking. Each case must be evaluated in depth hopefully with domain expertise.
DeepSeek hasn't proven anything except what they say. You're inherently trusting a company's statements over others. When an independent third party corroborates the numbers then we know for sure. Until then, every company is just selling a story for investors and users.
Does DeepSeek offer a heavily subsidized subscription that provides 20x+ more usage than the equivalent value in API tokens?
API usage being profitable is completely plausible, but almost nobody pays API prices for Claude and GPT, nor would they be able to afford them if the subscription suddenly stopped existing.
If you're actually asking the question and not being sarcastic, no, Deepseek does not offer subscriptions at all, subsidized or otherwise. You only can use them at the API rates.
In both places I've worked at in the past two years (F500), subscription plans were banned due to data privacy concerns. Everyone _had_ to use the provisioned API tokens (even if they had a personal subscription)
Not DeepSeek themselves, but their open models are run by third-party companies in a profitable way. The unsustainable interference theory has been debunked.
Define profit? If by profit you mean covering compute costs then yes you're correct. If by profit you mean covering spent RnD budget and future RnD budget allocations then no, by no means are any of these AI companies running at a profit.
Also these third-party providers are running magnitude smaller models than what Anthropic and OpenAI's power users expect from them. Yes the average chat message isn't being routed to the 1+ Trillion param model but people running CC or Codex on Ultra probably are. Those are objectively more expensive.
I am citing out of my head here, but think deepseek’s inference engine with their models create around 30 dollars per hours while it costs around 4 dollars, there might be other costs but if you look at the demand, it will become very profitable.
I don’t understand why people are pessimistic about the profitability (I don’t want them to be this successful either). BUT why would they care about profitability right now. They believe in AGI, so do the big money people. They get their investments. This train is not stopping and I believe their predictions are correct (IMO we should not debate about timescales apart from spreading urgency).
And for the lowering of prices, it is possible because they just can and hooking people now into their systems will be 10x more profitable in the future. In addition to all the juicy training data they get with real world clients.
They are scared the others win, they are not scared that they run out of money.
I got our pocket god ChatGPT to fact check this discussion. No lab is profitable, it is a moonshot to expect they'll ever recover the billions frontier labs have put in behind these DCs. I never understand this argument everyone tries to make, AI is too expensive -> It will get cheaper as they scale -> AI is too cheap and not sustainable -> No it's very profitable on API -> go back to 1.
I am not an AI denier, I use it and I find it very useful. That said I can disagree with the current economics of the industry and their business model.
Inference economics only work for those who build on top of the models lol. It should be said that Anthropic and OpenAI suck at product development too - claude for finance, Sora, OpenAI web browser - wtf?
Its difficult for people to accept this because many want the model producers to succeed badly.
They are selling inference at severely gimped capability. Even in openrouter the quality and speed of the model being served varies by provider. When you dig into the details (which most people don't do), not all providers are equal.
The conditions for DeepSeek's profitability are very very narrow. A key piece being that they have cheap energy, a second key is that they have west/east clients and thus extremely high uptime.
Energy costs (and almost everything else) are a minimal part of providing inference.
Most of the cost is depreciation on the GPUs and the HBM.
The one number you need to figure out is how much it costs you to buy that GPU and HBM, then how many tokens you can sell from it over its lifespan which is generally amortised over 2 years.
> One outcome that would be genuinely ironic is that if these models are more compute efficient, it reduces the need for these massive data center buildouts, which could cause cascading economic failures. However, that likely won’t happen since inference is only one part of the compute mix and those data centers are still needed for other parts of the pipeline.
Don’t agree with the however. I think like you’re saying we’re going to get to see in real time what happens when something is commoditized and it gets massively overbuilt.
A historical episode that may arguably be related -- regarding stock market euphoria & overbuilding capacity -- was the dot-com era Telecom bubble & crash [1].
Even if something is a reasonable infrastructure project with a sound long term business plan, investors can still lose money if they pay too much for it. For the Telco bubble & crash, there were a few of these Telco companies engaging in financial shenanigans / securities fraud to juice their quarterly financials by leasing capacity to artificially inflate their operating revenue etc. E.g. we booked $100m of revenue by leasing X units of our capacity to $competitor! Sure, we're still extremely cashflow negative, but look at our operating revenue grow! Fine print: we also incurred a symmetric $100m expense to lease X units of capacity back from the same $competitor, but for some reason we didn't classify that as an operating expense, we've thrown it into a different accounting bucket.
That said, overbuilding fiber optic capacity by laying undersea cables is likely different to overbuilding GPU datacenter capacity in that most of the cost of laying a cable is the physical cost to lay the cable -- given you're already laying an undersea cable, the incremental cost to add a bit more capacity to it (more optical fibers) is negligible -- at least before you've sunk the cable! Whereas for a data center full of GPUs, maybe 50%-60% of the total cost is the GPUs, so it isn't cheap to massively overbuild capacity.
Well they did pull back on ZDR a few times and at least Anthropic has made their arguments (that I disagree with) that model outputs, no matter how much user context was fed, do not fall under their definition of "user data".
Are these prices subsidized by user generated outputs?
Can I even trust any answer to not be carefully worded and rephrase the response to "user data"?
Even so, what's to prevent some "oops" event that seems to be all too common?
At least some months back Bedrock seemed to be properly isolated, but I've heard of many getting non-ZDR notices for new models.
It’s probably more efficient caching methods and smaller models (6 sol is actually 6 terra and opus 5.5 is smaller than 5.1 etc). Don’t overthink this please
> It’s no surprise both companies are trying to build moats to defend against this trend, such as a focus on enterprise commitments and personal agentic assistants like OpenAI Dots and Claude Duet.
This probably won't be enough, because Openrouter now has to do age verification to verify their users are adults, and if they don't, they have to apply "teen mode" to all users.
> leaked financials from both OpenAI and Anthropic show unsurprisingly that they are losing billions due to R&D/scaling
No, they are losing billions because LLMs are not a viable business. Their real business is making bombastic statements which the credulous press prints unquestioning, fooling various investors and ... that's it. IPO is another of these. You really should note how the Anthropic IPO date keeps slipping because they can't make the S-1 public as it would give away the game immediately. The market was fooled by SpaceX once you can't do this a second or third time.
Are those statements audited or fact checked? Inference profitability is great but how much of that profit gets eaten up paying back training costs, R&D, and operating expenses.
From their own inference framework docs: "A single cluster currently serves 155 trillion tokens per day, with overall revenue of US$38 per GPU-hour (at valley pricing)."
If they keep it running at near constant usage, you'd only need 80-120 million tokens an hour to hit 38$ of revenue (at off peak pricing, half as much for peak pricing).
The cool thing about models designed for cheaper inference is that post-training (which often requires inference as a part of the feedback loop) becomes cheaper as well, so almost certainly cheaper than models that are less optimized.
The market has decided that SPCX is worth $147, as of Friday, 5pm. Not sure who's fooling whom, but that's the market speaking. What that says, you'd have to ask a financial analyst, but that's where it is. It IPO'd at $135.
SPCX had a pretty wild ride so far with a jump to 200 a few days after IPO and a drop down 108 a few months ago so I'm not sure the market really knows how to properly price SPCX right now.
The same evidence could just as possibly point to supply being too great for demand. The OP article references this as a distinct possibility: if models really have gotten more compute efficient, we may wind up in a situation where we have more inference capacity than we actually need.
To give a simplistic example: if a city as a ton of available AirBnBs, you might say, "Lots of people are willing to rent out their apartments, it must be profitable for them to do so." But the fact that all of those rooms are not already booked could also mean that way too many people have attempted to rent out their apartments.
And for what it's worth, it's possible for supply to outstrip demand while still leaving inference profitable. The question is whether it's (a) profitable enough to service massive debts and (b) sustainably profitable over the long run.
Edit: to add one note, it's possible for the LABS to be profitable on inference without inference itself being profitable. We don't know what the contracts look like, but it's entirely possible that labs received sweetheart deals from neoclouds so that the latter could flaunt big utilization numbers.
If inference wasn't profitable why are consumer GPUs that are only good for running open models so expensive? Why and how would these companies sell access to open models if it wasn't profitable? You can rent your own GPU and make money if your electricity is cheap enough.
>You can rent your own GPU and make money if your electricity is cheap enough.
I would be curious to know whether you'd make more profit from the same kWh of electricity mining bitcoin versus serving inference. I know they use different hardware, but since the "inference is profitable" discussion ignores the purchase price of the GPU, let's pretend you could have opted to buy an ASIC instead.
It’s both competition almost 3 way , cheaper open source , a clearly lagging behind a bit open ai that cuts prices to win Anthropic customers .
There was an interesting technical advantage by some models where Anthropic as a leader start loosing dramatically in speed too. 5.5 made a come back in that dimension.
Well, because after 2011, when Anderseen famously said "software is eating the world"... it only ate so much of the world. We have that data on verticals now... is it possibly the same thing for "AI?"
Genuine inquiry here.
However, my bigger worry is this is just a temporary lull, and that due to alignment, regulatory, and distillation pressures, the best models will be kept internal, and then all of capitalism fails because the labs could maybe eat all of knowledge work.
There are already no junior hires. The ladder is already collapsing. Embrace, Extend, Extinguish.
I am guessing something to do with Qwen, Kimi and Deepseek. Plus after the tokemaxxing wave died down, a lot of cancelled subs from enterprise? Just my 2ct.
Price war. Duh. There is no moat. Also them chinese competition. It’s going to be race to the bottom. It’s gonna be ugly for the stock market. Good for consumers though!
Solar panels added to a grid is not plug and play.
When you're connecting solar to a municipal grid (which is what most homes are connected to), it could theoretically be taxed or even subjected to a tariff. The "added" power (often excess) is monitored for billing credits/payout/other, among other reasons like grid integrity.
Solar panels are literally plug-and-play, though! If your market doesn't have plug-in "balcony solar" yet, it is because your lawmakers are blocking it.
And indeed, they can tax excess solar, but they can't tax the power you use yourself. Backfeed tax gets too high to use the grid as a battery? Buy a home battery, problem solved.
Plenty of grids have different sorts of net metering ratios, grid maintenance fees, etc. But it's nothing compared to what you'd be paying taking power off of the grid, and there is a break-even period often measured in single-digit years. I still do not understand the point of what the GP was trying to make w.r.t. "taxes".
Strange, I lived in very blue states (WA and CA) and they definitely punish EV owners with large annual surcharges on registrations. To offset “not paying gas tax”. Money is money, red or blue, they will all take as much as they can from you.
That would likely require innovation in a renewable liquid fuel that can work at scale.
Electrification will drive the demand for oil down, making the overhead more expensive per gallon, and eventually we will largely move off of fossil fuels, though I don't expect completely. It will be many decades due to the turnover times for industrial/commercial transportation systems.
Tokens per second? That only really matters for things like chat and sessions where you are monitoring a session as it runs which is going to be an increasingly small percentage of usage. That's bottlenecked by human attention. Agents in the background, long running tasks, unattended tasks, all that stuff is what is even capable of driving significant usage because you can just increase consumption massively by parallelizing without being the bottleneck. I wouldn't be shocked if already the vast majority of token usage was from unattended sessions. TPS is not that important there
It is impossible to make sophisticated software products at a high level with a low TPS. Try making a modern emulator or an award winning game. Both require a lot of iteration and not everything works well with multi agent.
DeepSeek is serving more TPS than anyone, about as much as OpenRouter does in a week in a day, at 38$/H800/Ascend so about 900k/~90% profit over 3 years (or ~80s% profit including training expenses).
Meanwhile, Oracle is handing Tencent 100.000 GPUs at 1.60$ hour, dangerously close to breakeven...
Doubt the race to the bottom is really good for consumers, unless you actually just want more stuff rather than better stuff. The bubble bursting is going to be pretty horrible financially for us all.
The main thing it seems to have done so far is made everything feel less trustworthy, and everything digital less safe. It has empowered scammers, that's for sure.
And for what benefit? Spunking out 7/10 code a bit faster with your brain turned off. Great!
It won’t be a ban, it will be some “certified safe” badge from a federal agency. Companies will be scared into not using anything that is not “certified safe” and business insurance will even require it. Whoever (openAI or Anthropic) gets to the politicians first will be the only certified safe model provider.
Maybe the FDA is a good analogy, you wouldn’t want to buy meat at the store not FDA certified would you? OpenAI and Anthropic want you to feel the same about AI.
I do not expect an outright ban. (1) the bots are too useful, open is the way, re: HF handling the incident (2) you can train without certain capabilities, see Kimi K3.1 (3) reminiscent of the encryption is dangerous debate of the '90s
There will likely be requirements like registering with the government (China already does this, every business already does everywhere), certain usages require more or have deeper regulations (this is already done with most technologies, depending on application)
I think it will not technically be a ban, but a de facto one.
Some variation on, "Open weight models must be A) trained in the USA B) have a public training dataset + training pipeline C) be reviewed by a government 'safety council', etc" Something which technically leaves a crack open for someone to release a less capable model than the commercial ones.
That kind of policy will hold a country back, and why I don't think it will happen. There are already many companies already using Chinese models.
Another confounding factor, if a US company fine-tunes a Chinese model, or vice versa, what is that country that should be assigned to that model.
You also have to consider the competition between the US and China. If the US did something like this, the world will move to Chinese hardware and Ai. The think tanks that talk with legislators understand this.
You have more faith than I do. It would be bad policy, but hardly a first for a government who wielded tariffs like a cudgel in spite of obvious repercussions.
honestly the surprising thing at this point would be if america did actually pass a policy of any sort, good or bad, vs just relying on the president to yolo some executive orders now and then that change every 2 years. Plus at least for the next couple years nvidia likes open source models and the president likes the ceo of nvidia. There could be a falling out but nvidia has deep pockets to ... discourage ... a falling out.
It's worth remembering that the tariff stuff is not any sort of durable policy. It's one guy sort of doing whatever until 9 other guys a year later go 'hey wait' and then the first guy does whatever again anyway. The whole thing is just a recipe for whiplash. maybe congress remembers that they're supposed to do stuff in a few months and steps in. maybe they don't and then every 4 years we have wild swings in enforced based on which half of a percent of the voters in georgia had rainy weather where they lived in early november. More of a simultaneous "All of the above and also none of the above" deal until the dynamics of american government fundamentally shift
There are people in Trump's inner most circle that know this. I'm not worried about it. People trust those calling for regulatory capture (Dario / Altman) the least. (HN is not a good representation of the general population and AI userbase)
I fully expect token vendors to go the way of mobile data providers. Increasing throughput and efficiency, yet commoditized and not reaping the massive rewards of the economic gains their technologies enable.
Can't wait for the unlimited plans we will have in due time.
"Unlimited" for a consumer will probably take a form like limited parallel agents at max intelligence for X amount of tokens, followed by either single threaded agents or nerfed intelligence or both.
This is the same mechanism used in unlimited mobile internet for example.
> Can't wait for the unlimited plans we will have in due time.
I don't see how this can happen. The cost of serving a model grows linearly with use (energy). The cost of serving a mobile data plan is approximately zero.
I'm curious, why would this not turn into an unlimited electricity flatrate? In most parts of the world, consumer electricity is charged by kWh, not as flat payment.
As long as tokens are as energy intense to compute as they are right now, I don't see a flatrate being a viable model.
Something close to how we rent other cloud resources today makes sense. We don't worry about electricity costs, it's built into the hourly prices, use as much of the hardware as you can. Electricity rates vary over the course of the day as well.
People don't like complex pricing. This would be for businesses running larger workloads. For individuals, the unlimited plan seems the most probable to me. This stuff is going to get real cheap in the coming years.
The energy needed per token is in dropping rapidly, even though the energy per-GPU is going up. We are seeing something like 10x improvements per year.
I can't believe the actual answer is so far down the thread. All praise to our overlord DeepSeek for ironically helping American labs make cached tokens much cheaper.
though problem with proprietary LLMs it's difficult to know if the vendor's supposed "optimization" affects performance negatively because it's difficult to define an SLA
It doesn't really make sense huh. If they're so resource constrained and tokens are so precious, then prices should go up regardless of how much optimization has been done. You're either struggling to meet demand or egregiously over-provisioned. Which is it?
> It turned out that GPT-6 Sol was a dud performancewise, so a week later OpenAI announced GPT-6.1 Sol,
I strongly believe that GPT-6 Sol was only ever meant to be the web chat model, but it was a panic public release to respond to Opus 5.5. It's clearly been over fitted for drawing diagrams in the chat.
>They know what they’re doing. The naysayers are like the people that said this little internet thing wasn’t going anywhere.
There's a big difference between people saying OpenAI/Antropic wont survive and LLMs wont survive. A lot of big early players in the internet space died off as well.
Whether you think they have a future or not does not change fact that they currently are wildly unprofitable.
>internet thing wasn’t going anywhere.
This is conflating the utility of the technology with the economics behind it. Few people claim LLMs are useless. It's a question of figuring out a viable business model.
they are not horrible business, some of them literally get invested by governments and get extra debt from them too so they can never go bankrupt? if they are at national level airlines, they survive no matter what.
that is not bad business? that is just private profits, socialize loses perfect business?
Competition drives prices down. We tried Kimi and GLM models after hitting 'cybersecurity' blocks on OpenAI and Anthropic. We found them to be much cheaper to access and helped us complete our work without 'cybersecurity' refusals blocking us.
They just started competing against each other? No.
Anthropic is going public soon. Until September, OpenAI was going public. Neither are remotely profitable.
They're not competing against each other, they're competing against free/local/open weight models that they're also lobbying very hard to regulate/ban.
As much as I hear this sentiment, and as much as all the smart software people around me echo the desire for this sentiment to be true; it feels like people say these things, but then at the same time say - well, Claude/Chatgpt is still marginally better for me now. But clearly they are catching up so maybe one day!!!
In that sense, I'm really not sure if Kimi/chinese models are really competing against Anthropic/OAI? Not to mention that for individuals, the subscription model is extremely different, and Kimi subs are (I think) worse value, and chinese API's maybe marginally end up cheaper than using american subscriptions.
The profit comes from the enterprise customers who have to pay for usage at API rates. Not the subsidized subscriptions, which both OpenAI have designed to be unavailable for large enterprises by withholding IAM and auditing features they need except on the usage based entperise plans.
As an individual I'm still paying for a ChatGPT Plus plan since $20/mo gives me a lot of value and I use many multiples of $20 at API rates per month. But in terms of capabilities, I am entirely satisfied when using my Opencode Go plan with open models like Deepseek Flash/Pro, Kimi and MiMo.
At this point keeping my ChatGPT subscription isn't because of a significant capability gap compared to open models. It's 90% about the value of the subscription priced well below API rates (10% the fun of trying latest models).
If OpenAI/Anthropic dropped their cheap subscriptions or drastically undermined the value proposition I'd be happy using open models exclusively and currently barely notice a difference switching back and forth for the vast majority of tasks.
I'm expecting the subs to increasingly provide less value as the US labs IPO and follow the SV pattern of moving from marketshare capture mode to profit extraction mode. Thankfully I'm already comfortable using OpenCode with open models and avoid lock-in to any specific harness so would be a very soft landing.
Openrouter stats seem to indicate that there is a lot of demand for ‘good enough at a good price.’ After all, not everyone is decompiling PS5 games. In fact, most people are just speedrunning notetaking apps in JavaScript frameworks.
But, Openrouter stats can only measure the usage they see. I'm not sure it is really all that representative of total LLM usage across the market.
I think that's the difference here -- OpenAI and Anthropic are competing more with each other (and Google Gemini) than they are with open weight models. While people on HN are familiar with Kimi and GLM models, I'm not sure how many casual users of LLMs know about them. I think the competitive risk open weight models pose to OpenAI and Anthropic is much further in the future.
There is a lot of usage that's only known by Anthropic or OpenAI that's not visible to the rest of the world.
> most people are just speedrunning notetaking apps in JavaScript frameworks.
I feel seen. But seriously, my app, with my preferences, written by a semi-trustworthy entity that isn’t trying to shove ads down my throat. That’s really great.
Have you considered you are reading comments from different people?
[dead]
If this is mostly competition enduced, I have to assume either their usage has plummeted or they will implode financially in short order.
The frontier labs were already burning money on more expensive price models, and competition wouldn't have reduced their capex proportionately.
How could they do anything but implode financially? Genuinely curious. The numbers are not right.
Did the frontier labs take out debt or did they make flimsy unenforceable agreements to neoclouds who took out debt?
Borrowing money is not the only way to go into debt. They reportedly made binding and non-cancelable commitments. Once those commitments come due then they will be indebted to their counterparties until they pay up or the courts sort it out.
This is comparable to how, if I sign a 12 month non-cancelable lease and then don’t pay the rent, I will be in debt to my landlord.
People heavily underestimate the number of tricks available to corporates for taking on "debt": https://latentstate.net/the-hidden-ledger/
[dead]
That's effectively borrowing money with a different name though. The only way to go into debt is to sign an agreement to pay someone money in the future, regardless of the terms used to describe it.
No, it's really not. Do you have a war chest to drag out a legal battle and the ability to bankrupt your landlord by withholding payment? The rules are different at this scale, and "you can't enforce a contract if the attempt will bankrupt you" is only one of the possible games.
Here's another one: pay a supplier to cripple their deployment. Suddenly it's the neocloud's fault for failing to deliver, not your fault for failing to pay.
Did we all forget how Sam Altman turned a non-profit into a for-profit? Do we think he'll bring less sociopathy to this battle?
You're delusional lmao.
Default is default.
E.g. Capital leases are treated as debt.
Stop posting nonsense.
Yes.
Apple is worth more than all the AI companies combined. You think LLMs are less valuable in the global market than slightly updated iPhone models every year?
LLMs? No. LLM companies? Yes. It's just math, data centers, and mass copyright infringement. The only scarcity is Nvidia, who just passed Apple in market cap, somewhat proving your point but not in the way you meant it.
If you consider Nvidia an LLM company, and I do at this point given how heavily they've invested in frontier labs and where much of their revenue comes from, they alone are worth slightly more than Apple.
Maybe not, but it is probably less likely that Apple will face stiff competition from freely available alternatives, given that AOSP has been a thing for the iPhone’s entire lifespan and nonproprietary Android phones have failed to capture significant market share.
I don't think we have seen an Apple of AI yet. I wouldn't describe the frontier that way, would you?
No, and moreover it’s not entirely inconceivable that the Apple of AI may turn out to be Apple itself.
Their costs could drop, competition could dry up, or they may only be unprofitable in the first place because they’re choosing to invest more money in growing their business than they’re currently making, a practice which need not and will not continue indefinitely.
It’s not at all unusual for companies to run at a loss for several years before turning a profit.
The government could bail them out, that's the most likely scenario if they don't implode IMO.
They could also continue to grow at the rates they had last year. I don't know where those customers would come from, but if they did pull that off OpenAI and Anthropic could sustain propping up the circular financing going around the AI industry.
Goverment bailing they out will be using people tax to rescue the companies that absorbed most of the 801k money so they can get people out of work... it's the ultimate slap to the face... I would strongly resist this, I would vote if they implode to give to whoever owned shares via 801k their money, to people and let them (openai, anthropic, et al) eat shit... I would love to buy one of those gpus to run an open model at discount price...
Completely agree it will be a huge slap in the face to the public, unfortunately that's a common solution.
When everyone is afraid the market will collapse, or more likely distracted because the market collapsed just enough to hurt the average person, the government will bail them out and either people won't notice or will welcome it as a solution to huge levels of fear.
Voting isn't usually much of a solution after the fact, the political machine doesn't allow for candidates that would fix anything to have a chance at any office that matters. There's a reason our politicians today all seem to have roughly the same goals and all seem to become extremely wealthy while in office.
A bailout would be a slap in the face when the correct solution is to nationalize them. If they are too big to fail then the government gets a controlling share of equity and we finally start the sovereign wealth fund we should have started decades ago.
Maybe you should try voting harder.
Meaning what? Cast more than one ballot?
I live in a state that votes heavily in one direction, most states are the same. Its fine for me to vote, but the winner of a majority of states is pretty well decided before a ballot is cast (by strong majority, I'm not claiming fraud).
If everyone who threw up their hands actually voted, the outcome would change. See also, the elections musk tried to buy in the last 18 months or so since he doged up the place.
Sure, and I hope that (a) everyone votes and (b) most of them agree with me. Bit what is your point exactly?
Stuff like this that defies all logic happens all the time. Amazon operated at a massive loss first it’s first decade and still IPO’d years before they posted a profit.
That's why Anthropic are rushing the IPO. You can explain a single quarter slowdown, you can not explain a multi quarter one.
What makes you say that Anthropic has multi quarter slowdowns?
My reason is their revenue chart is flattening.
https://techstartups.com/2026/10/15/anthropics-q3-revenue-hi...
I am technically not saying that. I am taking off from the parent comment and speculating.
Their data from September is leaked via IPO preview, and October is not yet complete, so how can someone even project that?
// I am a techie, not a finance guy, so apart from public chatter, no inside knowledge.
And I do feel their IPO is being rushed before anyone has enough information to make up their mind. The VCs or private equity have given them a lot of money and want to cash out. When the revenue does slow down, any investor should pause to take a look at whether they can right the ship, or has the landscape itself changed.
Have more confidence - you are correct.
Their revenue taps, product releases and marketing campaigns were all strategic: to be able to show nice numbers in the financials.
Once they stop playing games with quality of output of these models and revenue's shrink - they will be in deep deep trouble. They are hoping they will conjure up a break through with the IPO proceeds.
Let's see.
I don't have the numbers in front of me but could have sworn I had seen quarterly revenue growth rates dropping drastically over the last year or so. Still a strong growth rate, but much lower than prior growth rates.
Edit: Data because silent down votes are useless, Anthropic Q2 2026 grew 147% while Q3 grew 45%
Yes but its growth rates with respect to expenditures, and in particular capital expenditures (R&D etc), which we have no idea about.
Gross margins here are irrelevant: if they can't charge a high enough price to cover all expenditures and rely on cash raised via fund-raises, they are not financially healthy.
Could they be one day? Maybe.
I think we agree in a backwards kind of way.
Yes they do need to grow margins to cover expenses, eventually when Amazon had to turn a profit.
My point was only that the numbers we do know clearly show a drop in growth rate of revenue, I can't judge numbers that aren't publicly known.
> or they will implode financially
They are too big to implode.
> If this is mostly competition enduced, I have to assume either their usage has plummeted or they will implode financially in short order.
If we assume it's due to competition, then their usage has precisely not plummeted because they cut prices by 80%.
So much this. I’m using these models primarily for dumb hobby projects and I’ve still hit this a lot - although much less so in Opus 5.5.
I wouldn’t be surprised if US “bans” access to the Chinese models by the end of the year
How?
I think your question is rhetorical, in which case I agree with the sentiment.
The US absolutely could (and very well might) make Chinese models illegal, possibly open weight models in general, but all that will do is keep them out of the hands of law abiding US citizens.
Law breakers in the US and the entire rest of the world would still have access, and nothing in the US will be any safer. People who think regulation will solve the problem should be embarrassed by their ignorance.
It may solve the problem of anthropic and open ai going out of business with all major consequences that would have for the rest of the US.
Lol, that's funny in an awful way. Choose between popping the AI bubble or creating a duopoly, and still have all the cybersecurity and other risks.
I don't get why people think Chinese models could bring openai and anthropic to their knees. There are zero signs of such a scenario. In fact the market is still growing at a gigantic pace. Quality of American models is still AAA. Chinese will have their place. But they are unmatched by their American rivals.
"Bring them to their knees" is highly unlikely.
"Puncture expectations of a multi-trillion dollar valuation" OTOH seems quite likely to me.
If the capability keeps growing significantly, then I agree it's possible the US could hold their significant lead indefinitely.
However, capability could plateau or reach a point where most of the market can't tell the difference and isn't willing to pay a premium.
> But they are unmatched by their American rivals.
Kimi K3.1 is top 5 on AA and half the price of Opus 5.5, Qwen4 is coming soon. There is near zero difference for most tasks, so it matters less today and even less as time goes on. Price is now the top concern for a reason.
https://artificialanalysis.ai/#intelligence
“Ban open weight models” Lol how. How do you even prove a model that I’m hosting is open weights?
OpenRouter and other inference providers won’t be able to openly host the models, for starters. That should cut most (or at least a significant share of) non-local inference, and local inference only works with smaller, less capable models, plus the hardware is hella expensive when you don’t have scale to saturate.
In addition, most companies don’t want to be in a position where a disgruntled employee can report them to authorities and rack up massive fines or worse, even if it’s hard to tell from the outside.
There's a difference between something being illegal and being able to enforce that law.
I don't know man. Marijuana is/was banned in most of the US. LOL, how do you ban a plant?
Given the general availability, the answer appears to be “you mostly don't.”
Banning open weight models in the US would provide the rest of the world with a huge business advantage since these models are much more cost-effective than US models.
If you are a legally operating company in the US just saying it's banned is enough, the threat of enforcement is enough pressure.
You and I obviously can just use a VPN, but 99% of usage is likely as part of a company.
Presumably, legally operating companies aren't the dominant threat though. So banning open weight models doesn't solve anything.
Not if you realize the real threat to be solved is threat to shareholder value.
ding ding ding
> If you are a legally operating company in the US just saying it's banned is enough, the threat of enforcement is enough pressure.
Is there precedent for the US actually "banning" software? I know for example the Kaspersky ban and tiktok ban, but both of those were either banning the sales or banning the service. IIRC neither made the actual software illegal.
What is an example of a specific software being illegal to possess? What is an example of a specific software being illegal to use if it is not used to break other laws (copyright, etc.) or talks to banned service providers?
Closest thing I can think of is restricting the export of cryptography in the 90's, but that was not even close to what you are suggesting.
Correct it’s 7% chance on kalshi, and those 7% of people are just wrong. I know everyone likes to larp that the US is some kind of dictatorship, but lol, lmao as they say. Also on the “how” side so far we have “they will just announce that “they” are banned, hilarious.
[dead]
What harness are you all using? OpenCode?
I have a harness for my harnesses now. My Hermes manages Codex, CC or Cursor now and interact with me via Slack and PRs on GitHub.
It's nice because I can keep going from anywhere, other team members can see and interject in my Slack threads, and Hermes auto updates its skills with any adjustments that we make so we don't have to ask twice.
I found Claude Code to work best. You can use it with any compatible API.
Pi.dev at home.
And I’m slowly writing my own. But it’s garbage at the moment.
I wrote one and kept iterating by using it to fix bugs in itself
I still also use other harnesses, but mine is already at the point where I prefer it more often than not
You should totally keep working on yours. It’s a cool learning experience. And you might stumble on some good ideas that nobody else has had
Don’t write your own. Fork maki.sh and customize it like I did. It’s built in Rust and is absolutely amazing.
Thanks for the suggestion, but that’s not a fit for me.
I’m writing my own harness to learn. And I’m writing it in zig.
But, I’ll absolutely add maki.sh to my reading list. So, that’s a win.
What kind of hardware do you need to run a useful LLM at home? To train one?
> "What kind of hardware... "
At home, I mostly use openrouter.
A few bucks will realistically last you forever. I also bought a DGX Spark at launch. I've learned a ton. but it's not usually my daily driver. I do run overnight tasks on it, like "read this book and write up notes on each chapter..."
In retrospect, given the dirt cheap prices of apis and or renting hardware, buying hardware for inference was likely a mistake (but not enough of a mistake to sell it now).
There are a lot of tasks I have, e.g. integration with homeassistant, that I absolutely will not trust a cloud provider for. On prem or nothing.
I'm not willing to spend thousands for those features, so my house will remain dumber than it could be until prices come down in a year or two.
Schaefer says they write their own harness, not necessarily training or running a local model.
Take a look at the docs for https://github.com/niko1221/strata for some ideas, particularly https://github.com/Niko1221/Strata/blob/main/docs/MODELS.md#... .
For me, I used Zed Agent with Codex as it has native integration with Zed's ACP so it's pretty convenient.
However, Anthropic is pickier about the harnesses you can use with Claude Code and I noticed discrepancies when using Zed's External Agent feature with Claude Code, so I just use the Claude app. It's fine.
Crush is great and easy to hack on: https://github.com/charmbracelet/crush
Personally I've found Zed's agent to be much superior to OpenCode. IDK why but it gets better results from the same model on the same task. Using with GLM 5.3 and DeepSeek 4.1.
I've been pretty satisfied with OpenCode 2
> much cheaper to access
I'm calling BS. I use Kimi, chatGPT (Astra/Sol), and Claude (Opus 5.5), and Kimi is by far the most expensive in terms of tokens and $/work done. Using OpenCode harness.
yea but I thought they had a massive compute shortage. Something doesn't add up. Imagine there being a beef shortage and the beef producers cut prices because "competition". why?
If the beef producers don't believe beef is a commodity and they want to get you addicted to their specific beef then it makes sense. Right now in these companies life cycles profit doesn't actually matter. The only thing that matters is how much of their beef is being consumed. They want to give people so much free beef they can't even imagine the idea of having a meal that doesn't contain beef.
Massive increases in efficiency and https://en.wikipedia.org/wiki/Jevons_paradox
DeepSeek claims $38 of revenue per GPU, per hour for Flash V4. That is $332,880 per GPU, per year. This matches their DeepInfer numbers: 25,600 output tokens/s per GPU sold at $0.60 per million -> $55 minus some other overhead to arrive at $38.
For the bigger Anthropic and OpenAI models (and associated higher prices) there are many unknown factors. But presumably they also work hard on optimizations, and it's not so crazy to think they had or still have a lot of wiggle room to adjust prices.
It's sad how far down you have to scroll to get the real answer. It's because Anthropic margins on inference are absolutely bonkers and it's easy for them to drop prices by 80% and STILL have positive free cash flow
Yes, but DeepSeek went absolute goblin-mode on optimization and I would not be shocked to discover that OpenAI and Anthropic were behind on that front.
OpenAI and Anthropic have more compute and the best frontier models to throw at the optimization problems. They have published articles on these efforts. For example, Dots, and Anthropic's upcoming "persistent agent" is because of these optimizations. Both companies often publish articles detailing a lot of their technical efforts. They just get buried among the flood of other news.
Yes, I would not be shocked either.
Sonnet 5.5 is 16x more expensive at $10 per million output tokens. Kimi K3.1 is also $10 per million output tokens, and its number of active parameters is only 6.5x larger than V4.1 Flash (104B compared to 16B). These things are not directly comparable of course, but just to give an idea.
Even if inference of those models needs 16x, 32x, or 48x the number of GPUs compared to V4.1 Flash, still seems like potentially great revenue per GPU.
My friend you are looking at the wrong number.
Its cash flows to the GPU that matter. Not revenue.
What is the TPS of DeepSeek? All of the LLMs I have tried are slow as molasses. \
I pretty reliably get 200 tokens per second on deep infra.
I use TensorX (EU hosted) and get 330TPS
Don't forget that Deepseek pays salary in RMB which is 1/7 of the value of USD.
Does Deepseek say how much those Clusters cost?
Given that they open sourced their inference framework, how realistic would be to buy hardware and become a provider on OpenRouter? These numbers are incredible.
This is why Anthropic and OpenAI are pushing for more “regulation”. You can’t show ROI to investors by cutting inference costs.
Regulation might inconvenience them a little, but the upside is huge if it means open models out of China get banned.
That would be like regulating who gets access to the best food on a sinking ship. The USA could allocate capital and attention toward much better things, for the country and for their people. I just hope the next couple administrations are wise enough to not experiment with the idea that ASI can get the country out of this 'fall of Rome' slow motion decline.
Ah yeah, our lovely free market, always asking for tariffs, blocking competition, to build cartels and monopolies.
Free markets don't exist to help individual companies, they are explicitly meant to hinder them.
They are meant to help the consumer.
Unfortunately neither of the two parties in the US even pay lip service to free markets anymore.
I wouldn't say 'help' the consumer as that makes it seem like things are biased.
The idealistic perfectly competitive markets entail exhausting all gains from trade.
> You can’t show ROI to investors by cutting inference costs
You can as long as you can saturate hardware with profit that makes up for training costs (which are also in free fall).
Anthropic is most at risk since they haven't bought any hardware (afaik), however OpenAi is building custom silicon now. The rest of Mag7 are primarily hardware vendors
Any sources on training costs in free fall? My understanding was training costs are proportional to model size. Is it just that compute costs are falling?
MiMo RL'd their latest models for $3.5M (million, not billion). It's hardware, software, and algos
https://mimo.xiaomi.com/rl/
Also look to the history of stable diffusion
I believe that’s post-training, not pre-training.
> I believe that’s post-training, not pre-training
Aren't AI companies already butting against the limits of usable training data?
Outside of undiscovered techniques, additional costly pre-training will likely yield diminishing returns if you've already scraped the whole internet and destructively scanned as many 20th century books as you could lay your hands on.
yes, just post training, which is the most expensive phase now as I understand it
Training costs are not in free fall, and not every "post training" process is equal, Everyone does this differently. The costs certainly haven't gone down, to train a new 10T param model from scratch, like at all.
You cannot point at Xiaomis tiny model and compare it to training a 10T param model. The levels of dunning krueger are getting out of hand.
You also arent calculating the salaries of the people involved, of which there's only a few thousand worldwide in your hiring pool.
> You also arent calculating the salaries of the people involved, of which there's only a few thousand worldwide in your hiring pool.
There were probably a few dozen 20 years ago, and the existing pool grew from somewhere. AI (nee machine learning) is not rocket science - but it is a science, so it rewards those who are literate, curious, can hypothesize, experiment and learn from results to gain an intuition. I bet most statistics-adjacent professionals (millions, worldwide) can move laterally into AI research and excel at it. Nevermind AI has been among the most-in-demand undergrad programs since ChatGPT dropped. The only limited pool is for celebrity AI talent.
You should start a recruiting outfit then if its that easy, because you could be making 40-60k per hire!!!
Your average Stats/machine learning expertise has little to do with standing up training infra and pipelines across $50 million in gpus.
you haven't cited any examples, nor acknowledged this is how everything in computing has gone, DK is a two way street I suppose
stable diffusion is an excellent example, the price to train today's or yesterday's frontier is the analogy, many more people will be able to train up what we have today, especially as more of the pipeline is open sourced and the knowhow distills
>stable diffusion is an excellent example, the price to train today's or yesterday's frontier is the analogy,
Dont use llms to draft your responses to people, extermely cringe. That explains the llm induced Dunning Krueger.
What examples... There are no public numbers regarding how much it costs to train a 10T model, atleast nothing concrete.
You should just use common sense and think about how you have to fit tensors in memory to train, and the larger the model is the more GPUs you need. Then you could also use common sense to think about how GPUs have gone up in price, not down.
> Dont use llms to draft your responses to people, extermely cringe
You are throwing around unfounded and incorrect accusations. Why? It's a weak ass rebuttal mate
common sense is using things like history to guide expectations instead just saying things without evidence that are contrary to the patterns in computing history
I would think that access to power is more critical than compute
> This is why Anthropic and OpenAI are pushing for more “regulation”
Given everything happening with Black October[0], don't you think regulating models may be reasonable?
[0] https://en.wikipedia.org/wiki/Black_October_(information_sec...
I too would like security issues to stay unpatched because only three letter agencies have access to uncrippled tools for finding the bugs.
Your point is important enough to say without sarcasm.
The AI companies didn't create the vulnerability. It has always been there.
The tools to detect these vulnerabilities will not go away. The only question is who will be allowed to have access to them?
I think that the frontier labs knew what they were doing when they invested in cybersecurity starting in the spring (or maybe earlier). They knew it would create hype. They knew it fit with their narrative that they are building a dangerous thing (something they believe because it’s fun to believe in doom, because it’s a great story for investors, and because it’s not inconsistent with a moat - in fact it makes a moat via regulatory capture).
Well, OpenAI is at the root of Black October, so maybe they are the problem
If they are the problem, shouldn't they be regulated?
You remind me of those politicians who try to scapegoat people who point out security holes. https://www.theregister.com/security/2022/02/15/missouri-end...
Regulate the bad actors, not the neutral tools. Lockpicks aren't illegal, but breaking and entering is. Imagine if MasterLock used a lockpick to break into a government building and steal stuff, then tried to lobby to make lockpicks illegal. That's what OpenAI is doing, and you're falling for it.
“Black October” sounds a lot like Red Scare.
This is the goal of all government regulations and it's always resulted in higher prices and near monopolies dominating the market.
Yet, so many Democrats support it.
In the hands of 4th party system Progressive Republicans and 5th party system New Deal Democrats, government was an extremely powerful force for trust-busting.
In the hands of 6th party system Republicans, namely Ronald Reagan and Robert Bork, the government worked hard to usher in a a new era of Robber Barons. We live in the culmination of this project.
History says the result depends on the hands, not the tool. 6th party system Democrats have largely failed to undo the Reagan/Bork policies, but that's an act of omission not comission and even then Lina Khan was a big step in the right direction. This issue was on the ballot in 2024 and will be again in 2028. Hopefully we make better choices in 2028.
It is pretty funny you refer to these periods because they are the quintessential example of big business pushing for regulations to push out their competitors and become monopolies. And the "progressives" enthusiastically supported such monopolization.
If you are going to limit the 5th Party system to the first half of the long 1930s-1990s period of overlapping realignments (which is pretty reasonable), then the 6th Party system should be the second half of that period, and the period from about the end of the Clinton Administration on should be the 7th Party system; you can call now the 6th Party system if you call the entire long realignment period the 5th (which makes some sense, too, though there is a pretty significant difference between the New Deal dominant period and the Civil Rights dominant period of the realignment), but to split it somewhere in the middle but then lump the period after with it makes little sense.
If you want to call Obama and Biden 7th party system, I don't really agree but I also don't think it matters for my argument: we can just re-label them and move on.
If you take away all regulation you’d probably end up with a monopoly anyway, as companies would no longer be prevented from merging with their competitors or using underhand tactics to gain market share.
It's wild how businessmen will pitch you the 10 ways they will abuse the rules of the game to obtain a monopoly and extract rent, then they will turn around and say that the game rules are perfect and our only failing is not following them hard enough.
Goomba fallacy
Nope, same guy. Seen it a dozen times. Unenlightened self interest is a helluva drug.
This is the goal of some, but very few overall, government regulations. Not even close to "all".
That's wrong, but why let reality get in the way of a pithy ideology?
Government regulations exist to regulate activity, to prevent, punish, and/or remediate harms.
Environmental regulations exist because businesses wouldn't clean up their own messes. Financial regulations exist almost exclusively because of harms and abuses committed by members of the financial industry. Building codes are updated to reflect lessons learned about people died, to prevent more people from dying the same way.
If AI regulations are passed, it will be because the AI industry decided not to self-regulate itself and stop illegally hacking computer systems, generating CSAM, or various other illegal activities that should easily be preventable if their systems were actually intelligent.
“All” is doing a lot of work there.
Also there are real trade offs. Food would be cheaper if, for instance, it were legal to cut it with any old cheap substance regardless of health effects.
But I totally agree that some regulations are dumb or are regulatory capture monopoly schemes.
Unfortunately like many other things there is no absolute simple rule to save us from thinking. Each case must be evaluated in depth hopefully with domain expertise.
It’s a race to the bottom to see who blinks first and goes bankrupt. It’s not sustainable.
Is it? DeepSeek has already proved they’re selling inference at a profit.
DeepSeek hasn't proven anything except what they say. You're inherently trusting a company's statements over others. When an independent third party corroborates the numbers then we know for sure. Until then, every company is just selling a story for investors and users.
I'm not aware of Chinese law, but I suspect that lying to the public or investors carries risks.
The people running AI companies are famous for being risk averse
Does DeepSeek offer a heavily subsidized subscription that provides 20x+ more usage than the equivalent value in API tokens?
API usage being profitable is completely plausible, but almost nobody pays API prices for Claude and GPT, nor would they be able to afford them if the subscription suddenly stopped existing.
If you're actually asking the question and not being sarcastic, no, Deepseek does not offer subscriptions at all, subsidized or otherwise. You only can use them at the API rates.
I doubt that's true for enterprise.
In both places I've worked at in the past two years (F500), subscription plans were banned due to data privacy concerns. Everyone _had_ to use the provisioned API tokens (even if they had a personal subscription)
Nope, they don't.
Other providers also charge similar prices (who are obviously not subsidising anything) for running DeepSeek models on their own infrastructure.
OpenCode Go sort of provides a "subscription" which will let you use DeepSeek but they're extremely transparent about what they're offering.
Not DeepSeek themselves, but their open models are run by third-party companies in a profitable way. The unsustainable interference theory has been debunked.
Define profit? If by profit you mean covering compute costs then yes you're correct. If by profit you mean covering spent RnD budget and future RnD budget allocations then no, by no means are any of these AI companies running at a profit.
Also these third-party providers are running magnitude smaller models than what Anthropic and OpenAI's power users expect from them. Yes the average chat message isn't being routed to the 1+ Trillion param model but people running CC or Codex on Ultra probably are. Those are objectively more expensive.
I am citing out of my head here, but think deepseek’s inference engine with their models create around 30 dollars per hours while it costs around 4 dollars, there might be other costs but if you look at the demand, it will become very profitable. I don’t understand why people are pessimistic about the profitability (I don’t want them to be this successful either). BUT why would they care about profitability right now. They believe in AGI, so do the big money people. They get their investments. This train is not stopping and I believe their predictions are correct (IMO we should not debate about timescales apart from spreading urgency). And for the lowering of prices, it is possible because they just can and hooking people now into their systems will be 10x more profitable in the future. In addition to all the juicy training data they get with real world clients. They are scared the others win, they are not scared that they run out of money.
I got our pocket god ChatGPT to fact check this discussion. No lab is profitable, it is a moonshot to expect they'll ever recover the billions frontier labs have put in behind these DCs. I never understand this argument everyone tries to make, AI is too expensive -> It will get cheaper as they scale -> AI is too cheap and not sustainable -> No it's very profitable on API -> go back to 1.
I am not an AI denier, I use it and I find it very useful. That said I can disagree with the current economics of the industry and their business model.
Inference economics only work for those who build on top of the models lol. It should be said that Anthropic and OpenAI suck at product development too - claude for finance, Sora, OpenAI web browser - wtf?
Its difficult for people to accept this because many want the model producers to succeed badly.
SOrry folks...
They are selling inference at severely gimped capability. Even in openrouter the quality and speed of the model being served varies by provider. When you dig into the details (which most people don't do), not all providers are equal.
The conditions for DeepSeek's profitability are very very narrow. A key piece being that they have cheap energy, a second key is that they have west/east clients and thus extremely high uptime.
Energy costs (and almost everything else) are a minimal part of providing inference.
Most of the cost is depreciation on the GPUs and the HBM.
The one number you need to figure out is how much it costs you to buy that GPU and HBM, then how many tokens you can sell from it over its lifespan which is generally amortised over 2 years.
This is not correct. Their margins are extremely high, (~1000%), and none of the factors you listed would cause them to have negative margins.
Because there's limited demand, and GPU capacity is like airline seats use it or lose it.
This will only get worse.
At the end the author briefly contemplates this:
> One outcome that would be genuinely ironic is that if these models are more compute efficient, it reduces the need for these massive data center buildouts, which could cause cascading economic failures. However, that likely won’t happen since inference is only one part of the compute mix and those data centers are still needed for other parts of the pipeline.
Don’t agree with the however. I think like you’re saying we’re going to get to see in real time what happens when something is commoditized and it gets massively overbuilt.
A historical episode that may arguably be related -- regarding stock market euphoria & overbuilding capacity -- was the dot-com era Telecom bubble & crash [1].
Even if something is a reasonable infrastructure project with a sound long term business plan, investors can still lose money if they pay too much for it. For the Telco bubble & crash, there were a few of these Telco companies engaging in financial shenanigans / securities fraud to juice their quarterly financials by leasing capacity to artificially inflate their operating revenue etc. E.g. we booked $100m of revenue by leasing X units of our capacity to $competitor! Sure, we're still extremely cashflow negative, but look at our operating revenue grow! Fine print: we also incurred a symmetric $100m expense to lease X units of capacity back from the same $competitor, but for some reason we didn't classify that as an operating expense, we've thrown it into a different accounting bucket.
That said, overbuilding fiber optic capacity by laying undersea cables is likely different to overbuilding GPU datacenter capacity in that most of the cost of laying a cable is the physical cost to lay the cable -- given you're already laying an undersea cable, the incremental cost to add a bit more capacity to it (more optical fibers) is negligible -- at least before you've sunk the cable! Whereas for a data center full of GPUs, maybe 50%-60% of the total cost is the GPUs, so it isn't cheap to massively overbuild capacity.
[1] https://en.wikipedia.org/wiki/Telecoms_crash
Yes - exactly.
Because all these newly trained models using the compute are doing really well and growing!
Well they did pull back on ZDR a few times and at least Anthropic has made their arguments (that I disagree with) that model outputs, no matter how much user context was fed, do not fall under their definition of "user data".
Are these prices subsidized by user generated outputs? Can I even trust any answer to not be carefully worded and rephrase the response to "user data"? Even so, what's to prevent some "oops" event that seems to be all too common? At least some months back Bedrock seemed to be properly isolated, but I've heard of many getting non-ZDR notices for new models.
It’s probably more efficient caching methods and smaller models (6 sol is actually 6 terra and opus 5.5 is smaller than 5.1 etc). Don’t overthink this please
> It’s no surprise both companies are trying to build moats to defend against this trend, such as a focus on enterprise commitments and personal agentic assistants like OpenAI Dots and Claude Duet.
Also regulatory capture, like OpenAI calcifying its teen mode into law via Adam's Law. https://openai.com/index/supporting-california-bill-advance-...
Openrouter recently changed their TOS to exclude under-18 users, presumably to escape liability from the law. https://conductatlas.com/diff/openrouter/openrouter-terms-of...
This probably won't be enough, because Openrouter now has to do age verification to verify their users are adults, and if they don't, they have to apply "teen mode" to all users.
> leaked financials from both OpenAI and Anthropic show unsurprisingly that they are losing billions due to R&D/scaling
No, they are losing billions because LLMs are not a viable business. Their real business is making bombastic statements which the credulous press prints unquestioning, fooling various investors and ... that's it. IPO is another of these. You really should note how the Anthropic IPO date keeps slipping because they can't make the S-1 public as it would give away the game immediately. The market was fooled by SpaceX once you can't do this a second or third time.
Deepseek prices their models based on a 10 month hardware payback period. They report ~$38 revenue/GPU-hour on V4.1-Flash clusters.
Inference is clearly profitable. The question is whether it's profitable enough to justify the training research costs.
Are those statements audited or fact checked? Inference profitability is great but how much of that profit gets eaten up paying back training costs, R&D, and operating expenses.
At best, this tells you v4.1 flash is profitable, but can they really get $38 per gpu hour? Sounds like a lot.
From their own inference framework docs: "A single cluster currently serves 155 trillion tokens per day, with overall revenue of US$38 per GPU-hour (at valley pricing)."
https://github.com/deepseek-ai/DeepInfer/blob/main/doc/infer...
So this implies they actually hit about double that during peak pricing.
If they keep it running at near constant usage, you'd only need 80-120 million tokens an hour to hit 38$ of revenue (at off peak pricing, half as much for peak pricing).
V4.1-Flash is probably quite smaller and much more resource effective than the frontier models. Add to that much cheaper energy in China.
Clearly potentially profitable. Not clearly profitable.
How much did it cost to train Deepseek v4.1 flash?
Great question. Nobody knows.
The cool thing about models designed for cheaper inference is that post-training (which often requires inference as a part of the feedback loop) becomes cheaper as well, so almost certainly cheaper than models that are less optimized.
The market has decided that SPCX is worth $147, as of Friday, 5pm. Not sure who's fooling whom, but that's the market speaking. What that says, you'd have to ask a financial analyst, but that's where it is. It IPO'd at $135.
SPCX had a pretty wild ride so far with a jump to 200 a few days after IPO and a drop down 108 a few months ago so I'm not sure the market really knows how to properly price SPCX right now.
The existence of websites where you can rent a GPU to run open models is proof that inference is profitable.
The same evidence could just as possibly point to supply being too great for demand. The OP article references this as a distinct possibility: if models really have gotten more compute efficient, we may wind up in a situation where we have more inference capacity than we actually need.
To give a simplistic example: if a city as a ton of available AirBnBs, you might say, "Lots of people are willing to rent out their apartments, it must be profitable for them to do so." But the fact that all of those rooms are not already booked could also mean that way too many people have attempted to rent out their apartments.
And for what it's worth, it's possible for supply to outstrip demand while still leaving inference profitable. The question is whether it's (a) profitable enough to service massive debts and (b) sustainably profitable over the long run.
Edit: to add one note, it's possible for the LABS to be profitable on inference without inference itself being profitable. We don't know what the contracts look like, but it's entirely possible that labs received sweetheart deals from neoclouds so that the latter could flaunt big utilization numbers.
If inference wasn't profitable why are consumer GPUs that are only good for running open models so expensive? Why and how would these companies sell access to open models if it wasn't profitable? You can rent your own GPU and make money if your electricity is cheap enough.
>You can rent your own GPU and make money if your electricity is cheap enough.
I would be curious to know whether you'd make more profit from the same kWh of electricity mining bitcoin versus serving inference. I know they use different hardware, but since the "inference is profitable" discussion ignores the purchase price of the GPU, let's pretend you could have opted to buy an ASIC instead.
Ironic statements, given that you yourself are an LLM.
It’s both competition almost 3 way , cheaper open source , a clearly lagging behind a bit open ai that cuts prices to win Anthropic customers .
There was an interesting technical advantage by some models where Anthropic as a leader start loosing dramatically in speed too. 5.5 made a come back in that dimension.
Trynna sell at Faberge prices when users can only afford Walmart
Well, because after 2011, when Anderseen famously said "software is eating the world"... it only ate so much of the world. We have that data on verticals now... is it possibly the same thing for "AI?"
Genuine inquiry here.
However, my bigger worry is this is just a temporary lull, and that due to alignment, regulatory, and distillation pressures, the best models will be kept internal, and then all of capitalism fails because the labs could maybe eat all of knowledge work.
There are already no junior hires. The ladder is already collapsing. Embrace, Extend, Extinguish.
I am guessing something to do with Qwen, Kimi and Deepseek. Plus after the tokemaxxing wave died down, a lot of cancelled subs from enterprise? Just my 2ct.
Price war. Duh. There is no moat. Also them chinese competition. It’s going to be race to the bottom. It’s gonna be ugly for the stock market. Good for consumers though!
Can they do the same thing with fuel prices?
Solar panels (which continue get cheaper every year) on my house makes charging my low-TCO EV free in 5 years
Until you start getting taxed.
How do you figure that works?
Solar panels added to a grid is not plug and play.
When you're connecting solar to a municipal grid (which is what most homes are connected to), it could theoretically be taxed or even subjected to a tariff. The "added" power (often excess) is monitored for billing credits/payout/other, among other reasons like grid integrity.
Solar panels are literally plug-and-play, though! If your market doesn't have plug-in "balcony solar" yet, it is because your lawmakers are blocking it.
And indeed, they can tax excess solar, but they can't tax the power you use yourself. Backfeed tax gets too high to use the grid as a battery? Buy a home battery, problem solved.
Plenty of grids have different sorts of net metering ratios, grid maintenance fees, etc. But it's nothing compared to what you'd be paying taking power off of the grid, and there is a break-even period often measured in single-digit years. I still do not understand the point of what the GP was trying to make w.r.t. "taxes".
Voters vote for Republicans that drive up TCO for EVs, like higher annual registration fees.
Strange, I lived in very blue states (WA and CA) and they definitely punish EV owners with large annual surcharges on registrations. To offset “not paying gas tax”. Money is money, red or blue, they will all take as much as they can from you.
TBH, roads and highways are paid for by gas tax and it effectively functions as a usage tax. Something similar for EVs wouldn't be too wild.
If you make it easy for lots of people to make the fuel, yes. EVs and smaller scale power generation anyone?
That would likely require innovation in a renewable liquid fuel that can work at scale.
Electrification will drive the demand for oil down, making the overhead more expensive per gallon, and eventually we will largely move off of fossil fuels, though I don't expect completely. It will be many decades due to the turnover times for industrial/commercial transportation systems.
Yes if China somehow gets the fuel and subsidize it to sell it for cheaper in the global world, even the gulf state price will go down.
The moat is high TPS. No one has rolled it to average consumers though.
Tokens per second? That only really matters for things like chat and sessions where you are monitoring a session as it runs which is going to be an increasingly small percentage of usage. That's bottlenecked by human attention. Agents in the background, long running tasks, unattended tasks, all that stuff is what is even capable of driving significant usage because you can just increase consumption massively by parallelizing without being the bottleneck. I wouldn't be shocked if already the vast majority of token usage was from unattended sessions. TPS is not that important there
It is impossible to make sophisticated software products at a high level with a low TPS. Try making a modern emulator or an award winning game. Both require a lot of iteration and not everything works well with multi agent.
DeepSeek is serving more TPS than anyone, about as much as OpenRouter does in a week in a day, at 38$/H800/Ascend so about 900k/~90% profit over 3 years (or ~80s% profit including training expenses).
Meanwhile, Oracle is handing Tencent 100.000 GPUs at 1.60$ hour, dangerously close to breakeven...
What does this tell you?
Doubt the race to the bottom is really good for consumers, unless you actually just want more stuff rather than better stuff. The bubble bursting is going to be pretty horrible financially for us all.
The main thing it seems to have done so far is made everything feel less trustworthy, and everything digital less safe. It has empowered scammers, that's for sure.
And for what benefit? Spunking out 7/10 code a bit faster with your brain turned off. Great!
Wait what, OpenAI is trustworthy now?
Just wait until the government ban on the dangerous open models materializes.
It won’t be a ban, it will be some “certified safe” badge from a federal agency. Companies will be scared into not using anything that is not “certified safe” and business insurance will even require it. Whoever (openAI or Anthropic) gets to the politicians first will be the only certified safe model provider.
Maybe the FDA is a good analogy, you wouldn’t want to buy meat at the store not FDA certified would you? OpenAI and Anthropic want you to feel the same about AI.
I'm not an American, but I thought the FDA didn't certify stores
Not just models, won't be long before hardware is regulated too.
Sounds like an eventual Supreme Court case.
I do not expect an outright ban. (1) the bots are too useful, open is the way, re: HF handling the incident (2) you can train without certain capabilities, see Kimi K3.1 (3) reminiscent of the encryption is dangerous debate of the '90s
There will likely be requirements like registering with the government (China already does this, every business already does everywhere), certain usages require more or have deeper regulations (this is already done with most technologies, depending on application)
I think it will not technically be a ban, but a de facto one.
Some variation on, "Open weight models must be A) trained in the USA B) have a public training dataset + training pipeline C) be reviewed by a government 'safety council', etc" Something which technically leaves a crack open for someone to release a less capable model than the commercial ones.
That kind of policy will hold a country back, and why I don't think it will happen. There are already many companies already using Chinese models.
Another confounding factor, if a US company fine-tunes a Chinese model, or vice versa, what is that country that should be assigned to that model.
You also have to consider the competition between the US and China. If the US did something like this, the world will move to Chinese hardware and Ai. The think tanks that talk with legislators understand this.
You have more faith than I do. It would be bad policy, but hardly a first for a government who wielded tariffs like a cudgel in spite of obvious repercussions.
honestly the surprising thing at this point would be if america did actually pass a policy of any sort, good or bad, vs just relying on the president to yolo some executive orders now and then that change every 2 years. Plus at least for the next couple years nvidia likes open source models and the president likes the ceo of nvidia. There could be a falling out but nvidia has deep pockets to ... discourage ... a falling out.
It's worth remembering that the tariff stuff is not any sort of durable policy. It's one guy sort of doing whatever until 9 other guys a year later go 'hey wait' and then the first guy does whatever again anyway. The whole thing is just a recipe for whiplash. maybe congress remembers that they're supposed to do stuff in a few months and steps in. maybe they don't and then every 4 years we have wild swings in enforced based on which half of a percent of the voters in georgia had rainy weather where they lived in early november. More of a simultaneous "All of the above and also none of the above" deal until the dynamics of american government fundamentally shift
There are people in Trump's inner most circle that know this. I'm not worried about it. People trust those calling for regulatory capture (Dario / Altman) the least. (HN is not a good representation of the general population and AI userbase)
There will likely be requirements like registering with the government
Sadly, my GPUs were lost in a tragic boating accident.
Because they have found a way that cost them less to have you use a lesser capable model o less expensive
I fully expect token vendors to go the way of mobile data providers. Increasing throughput and efficiency, yet commoditized and not reaping the massive rewards of the economic gains their technologies enable.
Can't wait for the unlimited plans we will have in due time.
They will not come. You can only spend so much time speaking on the phone.
But if you get unlimited tokens, with scheduled/automated jobs and subagents you will burn a LOT of money.
"Unlimited" for a consumer will probably take a form like limited parallel agents at max intelligence for X amount of tokens, followed by either single threaded agents or nerfed intelligence or both.
This is the same mechanism used in unlimited mobile internet for example.
I rarely use more than 25% of my subscription's weekly allowance. So, it's already unlimited.
> if you get unlimited tokens, with scheduled/automated jobs and subagents you will burn a LOT of money
Maybe, maybe not. There could be a saturation point of compute demand from ordinary consumers. (See: hot-water analogy.)
> You can only spend so much time speaking on the phone.
Tell that to my dial up modem as I played Active Worlds in 1996.
Yeu were still limited to about 4GiB/month usage as a time/bandwidth ceiling
Not where I am, it's 20G mobile and internet is unlimited, the level below mine has 1Tb cap.
The parent poster was talking about the amount of data transfer you could do on dialup over a month in 1996.
Mobile video data and tethered devices would tell a different story. Usage growth has been significant.
I expect they will be "unlimited" in the same sense the mobile / ISPs do that 99% of people never reach the limits.
We could have "unlimited" electricity too, if we choose, by printing solar panels (like the Chinese) and hopefully fusion before long.
> Can't wait for the unlimited plans we will have in due time.
I don't see how this can happen. The cost of serving a model grows linearly with use (energy). The cost of serving a mobile data plan is approximately zero.
> The cost of serving a model grows linearly with use (energy)
The cost of processing tokens however does not. Throughput and efficiency are improving around 10x per year. Hardware is not the only factor.
I'm curious, why would this not turn into an unlimited electricity flatrate? In most parts of the world, consumer electricity is charged by kWh, not as flat payment.
As long as tokens are as energy intense to compute as they are right now, I don't see a flatrate being a viable model.
Power and memory are apt to become the biggest billing constraints.
Something close to how we rent other cloud resources today makes sense. We don't worry about electricity costs, it's built into the hourly prices, use as much of the hardware as you can. Electricity rates vary over the course of the day as well.
People don't like complex pricing. This would be for businesses running larger workloads. For individuals, the unlimited plan seems the most probable to me. This stuff is going to get real cheap in the coming years.
The energy needed per token is in dropping rapidly, even though the energy per-GPU is going up. We are seeing something like 10x improvements per year.
[flagged]
Because they have been making inference a lot more hardware efficient.
By using Chinese research
I can't believe the actual answer is so far down the thread. All praise to our overlord DeepSeek for ironically helping American labs make cached tokens much cheaper.
They're competing with NVIDIA (huggingface)
competition. that's good for consumers.
though problem with proprietary LLMs it's difficult to know if the vendor's supposed "optimization" affects performance negatively because it's difficult to define an SLA
credits goes to china ai companies
credit goes to china ai companies
Think of all the open source projects in DeepSeek
Duh, China?
We also see improvements like 7x between the current and last Google TPU generations, with different chips for training vs inference.
Part of the cheaper costs, if not the majority, probably are KV cache optimisations. https://insufferable.dev/posts/the-ai-race-just-got-awkward/
This is literally just some rando blogger’s (overconfident) theory.
Microeconomics may give us a clue
Lack of demand and/or commodity product.
Either doesn't bode well for a multi-trillion dollar valuation for either company.
It doesn't really make sense huh. If they're so resource constrained and tokens are so precious, then prices should go up regardless of how much optimization has been done. You're either struggling to meet demand or egregiously over-provisioned. Which is it?
> It turned out that GPT-6 Sol was a dud performancewise, so a week later OpenAI announced GPT-6.1 Sol,
I strongly believe that GPT-6 Sol was only ever meant to be the web chat model, but it was a panic public release to respond to Opus 5.5. It's clearly been over fitted for drawing diagrams in the chat.
Moore’s Law.
People think OpenAI and Anthropic are just these wildly unprofitable companies with no future or business.
They know what they’re doing. The naysayers are like the people that said this little internet thing wasn’t going anywhere.
>They know what they’re doing. The naysayers are like the people that said this little internet thing wasn’t going anywhere.
There's a big difference between people saying OpenAI/Antropic wont survive and LLMs wont survive. A lot of big early players in the internet space died off as well.
Whether you think they have a future or not does not change fact that they currently are wildly unprofitable.
>internet thing wasn’t going anywhere.
This is conflating the utility of the technology with the economics behind it. Few people claim LLMs are useless. It's a question of figuring out a viable business model.
correct.
Commercial airlines are of great value to our lives. Who's questioning or denying that?
Horrible business to be in though.
they are not horrible business, some of them literally get invested by governments and get extra debt from them too so they can never go bankrupt? if they are at national level airlines, they survive no matter what.
that is not bad business? that is just private profits, socialize loses perfect business?
A good business is measured by the returns it generates.
They do not generate economic profits.
Yeah, I think AI model provider would goes the same way commercial airlines does
the money is infrastructure (hardwares, electrics, data etc)
Might want to Google Global Crossing. The internet survived, but most giant companies did not.
The Internet was profitable, but I'm not connecting to this website from my Solaris Sun Microsystems machine using Netscape Navigator
Or OpenAI and Anthropic are AOL and Yahoo.
> The naysayers are like the people that said this little internet thing wasn’t going anywhere.
The same rhetorical tactic was used to diminish blockchain critics.
It sounds like you're just a "yessayer", though? Not much different from naysayers, just on the opposite side.
One thing that naysayers have going on for them is the fact that those companies indeed aren't making profit right now.
People get obsessed with wanting companies to profit ASAP.
That's not how you grab market share. Early on, any profit you aren't reinvesting into the business is basically hurting your growth rate.
What value has market share if switching costs are close to zero, and alternatives continue to be available?
Grabbing market share by making losses only helps if it actually wins you a genuine monopolistic market position.
So did the bankers in 2008.
Crazy how delusional people are.
They are in too deep - they have to believe it all works out and act accordingly. lmao.
[dead]
[dead]