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.
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.
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 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.
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.
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.
Not DeepSeek themselves, but their open models are run by third-party companies in a profitable way. The unsustainable interference theory has been debunked.
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.
> 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.
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.
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!
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
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 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.
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.
7 replies →
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.
25 replies →
That's why Anthropic are rushing the IPO. You can explain a single quarter slowdown, you can not explain a multi quarter one.
7 replies →
> 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?
20 replies →
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.
8 replies →
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.
1 reply →
What is the TPS of DeepSeek? All of the LLMs I have tried are slow as molasses. \
2 replies →
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.
1 reply →
> 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?
10 replies →
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.
1 reply →
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
1 reply →
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.
3 replies →
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.
3 replies →
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.
2 replies →
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.
3 replies →
Not DeepSeek themselves, but their open models are run by third-party companies in a profitable way. The unsustainable interference theory has been debunked.
6 replies →
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.
2 replies →
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.
2 replies →
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?
1 reply →
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.
2 replies →
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
8 replies →
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
1 reply →
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.
1 reply →
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)
6 replies →
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.
1 reply →
> 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.
3 replies →
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.
3 replies →
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.
1 reply →
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]