Posted by akrylov 21 hours ago
There's a significant amount of innovation happening, but if the market decides this AI thing is not worth funding then I think that'll dry up overnight.
1. https://thenextweb.com/news/anthropic-private-equity-venture...
In my eyes I would rather use the AI I can run on my own paid infrastructure, so if there's an outage its isolated, or I could potentially have a different region / DC to fallback on.
I'm still surprised that neither Microsoft nor Amazon have made their own models available on their cloud offerings. I guess Microsoft probably does have Phi on there, but it's not front and center, especially with something like Copilot for Devs (seriously Microsoft rebrand that damn thing to be clear what you mean by Copilot!) where they could use the cheaper compute by using something like Phi.
https://azure.microsoft.com/en-us/blog/introducing-anthropic...
https://docs.cloud.google.com/gemini-enterprise-agent-platfo...
Claude has been available on AWS Bedrock for a long time too.
The new "Claude Platform" announcement was about an Anthropic operated version on AWS (as opposed to self-operated on Bedrock). See the differences here: https://platform.claude.com/docs/en/build-with-claude/claude...
> In my eyes I would rather use the AI I can run on my own paid infrastructure,
Claude has been available like that for quite a while.
One of the reasons for the OpenAI divorce from MS was so they could become available on AWS where they see significant demand, and being available only on Azure was holding them back.
Yes, you can even choose regions, for EU they serve it from Belgium. With all the encrypted at rest stuff and other guarantees that vertex provides.
> Important: Accessing Claude models through Vertex AI meets the FedRAMP High requirements, and operates within the Google Cloud FedRAMP High authorization boundary.
I understand that America dominates in distribution, integration, enterprise contracts, ecosystems, infra... The article isn't wrong, it's just that that dominance is fragile and requires constant upgrading.
But what is the point of that if you have to infinitely scale because the opposition is right behind you at all times ready to usurp you... You CANNOT scale infinitely, the VC money will run out at some point and then everyone will have to downscale everything to meet the real costs associated with SOTA models, they'll have to be able to use subscriptions, and other monetization to cover those insane costs, we just saw SORA shut down because it was bleeding money far too fast while the Chinese released video models that far surpassed it back to back to back...
EDIT: Hell, one of the most critical aspects is integration of the models into other products, and even on this end open-source is keeping up (and will eventually outpace when the VC money dries out) with these big companies.
Citation needed.
All reporting is that they are profitable on the inference side and all the VC money is going to building more data centers to run more inference. (Note that the coding subscription models are probably only break even on average - the money is in the API)
> The Chinese models are keeping up with them, while offering the models for free and able to run on consumer grade hardware, and more importantly they train them for cheap.
No one is running DeepSeek v4 (a 1.6T token model) on consumer hardware.
They aren't much cheaper to train the US models. Training is subsidized by the big Chinese tech companies. They are slightly cheaper because they are smaller (and weaker) models than the 5T and 10T models the US frontier labs are training, and the US labs are paying for a more diverse set of RL data (which shows up in diverse benchmark performance).
> we just saw SORA shut down because it was bleeding money far too fast while the Chinese released video models that far surpassed it back to back to back...
Ironically this proves the point.
OpenAI didn't shutdown Sora, just the subscription version and weird social network thing. You can still access it via API.
The Chinese models are API models and probably just as profitable for them as the LLMs are for the US frontier labs.
[1] has prices for video models. There is a big range, but Google's Veo model and OpenAI's Sora are around the same price as the Chinese models.
Ask yourself if AI was so profitable, why don't any of the big hyperscalers break out AI revenue in their earnings. OpenAI and Anthropic both project huge losses for the next couple years, it's not hard to find.
The real problem is, as the GP comment pointed out, that they can never stop training. As long as they're committed to building these behemoth models, the second they stop training, someone else will catch up and everybody will switch over because it's trivial to do so.
No. Anthropic at least expects to be profitable this year:
> Anthropic expects its gross profit margin, which measures how much revenue it makes compared to the cost of producing that revenue—largely from running servers—to swing from negative 94% last year to as much as 50% this year and 77% in 2028.
> And yeah, if you subtract out all your R&D, payroll, sales, marketing, and other overhead, and get someone else to take on the debt or dig into their free cash flow to build the hugely expensive infrastructure on which you depend, it'd be pretty hard to not be "profitable".
I think excluding capital expense on infrastructure isn't unreasonable and is done in most industries.
It's worth noting that AI infrastructure has turned out to be an unbelievably good investment. Inference on a 4 year old H100 chip costs more now than it did brand new! That makes the hyperscaler's depreciation schedules look very (and unexpectedly!) conservative (!!)
Literally not a single one of these AI companies, regardless of where they are in the world has any right to complain about someone copying their work.
> OpenAI’s counsel asked Musk whether xAI has ever “distilled” technology from OpenAI.
> Musk: “Generally AI companies distill other AI companies.”
> “Is that a yes?” Savitt asked.
> Musk: “Partly.”
From https://www.interconnects.ai/p/the-distillation-panic which is worth reading in full.
https://decaboy.fit for tracking progress at they gym
https://megaparley.com sports betting platform
A horse betting platform not published yet, still looking for an API odds provider
A car mechanic AI assistant not published yet
I've learned that the more detailed the initial prompt the better result I get. I can share any prompt if you want
But just for the sake of discussion, let me ask: Who is the service provider you're using to run Deepseek V4? Do you have any way of knowing whether that compute is happening in the US or abroad?
Article content: “The US are capitalizing on AI the best”
A lot of assumptions there that no one can actually verify as true right now. If commercialization into rent-seeking SaaS landscapes is the endgame, then yeah, the US is winning the AI race. If individualization, local LLMs, and consumer hardware are the endgame, China is winning the AI race. If it’s something entirely different - if LLMs are the wall and research is what grants the next breakthrough, or if compute and memory requirements take a dive, or whatever; then we have no idea who’s winning the race because that stuff is mostly happening behind closed doors.
It's only a proof that it's possible with 18+ years of training.
Those are much more specialized models with pretty mediocre tokens per second.
I think China is thinking more about the application layer on top of models as going to matter more than the models themselves, so they don't need to gatekeep the models as much.
If China could work at the frontier, I don’t know, I kind of think they would still be dumping a lot of resources into exploring the value side since they have that culture already in place.
That's zero ex oh (the letter) five
> LLMs strongly prefer word-level tokens, and word substitutions follow semantic similarity and not the more human auditory similarity.
Is this an elaborate joke or your full-word misspelling of writing is both agreeing with your statement (word substitutions) and contradicting it (not semantic but only pronunciation similarity)
? Claude, ChatGPT, etc are heinously expensive for tiny benefits lmao. Local + efficient is clearly the future
Unfortunately local inference is inefficient, 100s of times more inefficient than cloud. When you answer one request at a time you still have to fetch all active weights into compute units, once every token. When you run a batch of 300, you load it once and compute 300 at a time.
Compared to cloud, local inference is less flexible. You can't scale up 5x or 20x, can't have spikes, and pay for it no matter if you use it or not. But usage factor is very low, like 5%. And to run a decent model your system costs $2000 or more.
Even if so, if China is coming behind 6 months later selling laptops with hyper-efficient local models that are 80% as good as "frontier" ones, I imagine they'll get the consumer business AND a fair share of the enterprise business as IT managers look at their options during the next refresh cycle.
Given economies of scale, I think it's ultimately inevitable that the enterprise more-or-less follows the consumer on this, and the consumer is going to prefer local models. There's no ongoing cost after the initial purchase, and your data at least nominally stays within your control.
Like I don't need an H100 or a dozen to summarize a PDF. And that's most of what I use AI for.
Corporate America is where the money is, and corporate America will dictate what products are successful by virtue of spend. Individuals aren't going to be paying $100s or $1000s/month en masse for these models but businesses will be. Being local and efficient isn't that important at this stage but even so as American companies continue to scale and invest they'll be able to make those models more local and efficient if the market wants it. Sort of like how you had a big, giant desktop computer and now you've got a super computer in your phone which is in your pocket. Going straight to "local and efficient" means going straight to being behind because at some point, perhaps now even, the local and efficient model won't be able to keep up.
For some reason people think that they somehow know something that Google or Nvidia or whoever, with hundreds of billions of dollars of real money at stake don't already know and it's both amusing and bizarre to see this play out again and again in off-hand comments like "lol tiny benefits".
You buy an iPhone even though the cheap-o Wal-Mart Android phone for $100 "does the same thing". Except that in this case the Android phone just puts you out of business while those spending big money for "tiny benefits" beat you in the market.
Capital inflows are different from manufacturing outflows. The US has historically imported capital which is part of why we have such a large trade imbalance. I’d encourage you to do some more digging here.
> The world where we could compete is gone.
Sigh no that’s just not true at all. We compete hard and fast all day everyday, economy is growing and will continue to do so, and no amount of leftist doomer, Chinese, Iranian, or Russian propaganda changes those facts.
No but money only has value because of a product of the human labor and production capacity it refers to. Money is not capital, it is a reference to/legal coercion of capital
> We compete hard and fast all day everyday
Sir have you ever been to the us? Lmao. We are only competitive in the industry of white collar work (financial/artisanal services), an industry that capital is actively gutting
These are just strings of words without meaning or importance.
> Sir have you ever been to the us? Lmao. We are only competitive in the industry of white collar work (financial/artisanal services), an industry that capital is actively gutting
Yes, I live here. Why are you posting obviously untrue and asinine statements like this? Go look at the Fortune 500. There ya go. What other evidence do you need? And not only are you writing dumb things here, your original post was wrong too! Please get off of social media or whatever doomscrolling news you are partaking in because it is bad for your health and perception of reality. The United States by any measure, as a matter of indisputable fact, a highly competitive and dynamic economy across pretty much all sectors. This is not up for debate.
People buy iPhones because of status signalling and network effects, neither of which appears to apply to AI model choice. LLMs are already rapidly on the way to being interchangeable commodities.
To the extent LLMs are commodity products you're right (so far), but that is limited to the main model providers, such as ChatGPT, Claude, Gemini, &c. with interoperability on cloud platform providers and other technology providers like an Apple offering you a choice of LLM with Siri or something.
If you want to suggest that some other model is in the same bucket as those primary 3, it goes back to the crappy, cheap phone analogy which is accurate. Yea you can make calls with it, but you make calls better with an iPhone.
I get your point but in what sense is that "free"? What mobile plan giving you an iphone doesn't come with explicit debt?
They run various schemes like this all the time, you can also trade in your existing phone a lot of times for pretty favorable terms. I've traded in phones that were a few years old and gotten $1000+ for them, especially when switching providers.
$729.99 purchase on device payment or at retail price required. New line req'd. Unlimited Welcome, Unlimited Plus or Unlimited Ultimate plans required. Less $730 promo credit applied to account over 36 mos; promo credit ends if eligibility requirements are no longer met; 0% APR.Taxes & fees may apply. Credits will appear on your Verizon Wireless bill.
If you think the iPhone is a status symbol you’re just wrong.
I'm just pointing out the statement:
> What mobile plan giving you an iphone doesn't come with explicit debt?
isn't invalidated by some Yahoo article pushing a marketing promo that when you actually do the math and read the fine print its not really a "free" phone, its always some form of debt or bill credit or something along those lines that makes the phone "free". You're still paying for the phone in the end if you read the fine print. In the end one commits to spending several hundred dollars over 36 months or whatever or you pay up front and they give you bill credits if you keep the plan.
People who prefer truth in advertising.
> Why be so argumentative over something so stupid?
I don't want people to believe untrue marketing statements and make poor financial decisions without actually bothering to read the fine print.
> some companies run free promotions
This just isn't true. They're not really "free". They come with lots of financial commitments.
> Apparently Verizon ran some promo in the past and may again in the future giving away iPhones
They still say they do on their website. If you're getting one "free" iPhone it comes with a commitment to spend at least $65/mo for 36 months. A commitment to spend $2,340 is a lot different from $0.
These are far from "free" phones. Can I go into a Verizon store, not give them a dime or sign any contracts and walk out with a phone free and clear to do whatever I want? No? Sounds like it's not really free then!
My point is if you're poor/homeless you're probably not looking to sign a 3-year commitment to spend a few grand to get a "free" phone. A lot of those people won't even pass the credit check to qualify to even sign up for one of these post-paid plans required to get the "free" phone. If you're really broke you would probably be looking at signing up for a lifeline plan and get yourself a cheap used iPhone instead of signing up for a $2,340 contract.
You’re anchoring yourself to one payment scheme and ignoring others and it’s besides the point which is that iPhones aren’t status symbols even if these schemes didn’t exist and iPhones weren’t extremely cheap or freely available.
I don’t have anything left to say here besides that I proved my point unequivocally.
I already said I largely agreed with this.
> major carriers can and do give them away in various schemes and did so in the past and will continue to do so in the future
They only do if you're financially illiterate.
> You’re anchoring yourself to one payment scheme and ignoring others
I'm being honest and taking about the real deal instead of blindly repeating marketing bullshit and lies.
> freely available
A commitment to spend thousands of dollars isn't the same as freely available.
The bank gave me this free house all I have to do is pay this mortgage for thirty years. But hey the house was free!
Once again, was the deal that you could walk into the store, grab a new iPhone, and walk out without signing a contract or other form of commitment? If not, it's not really free. It's bad financial advice for people struggling financially to get one of these "free" phones, they're often more expensive than buying outright and getting a much cheaper (or potentially even subsidized!) plan. Especially if you're just needing one or two lines. Many of these postpaid plans only really make financial sense once you're at like 4+ lines on it.
I'm reminded of seeing all those cell phones in the RadioShack mailer ads back in the day. Only 99¢! Dad, can't I get one? It's only a dollar!.
If you spent hundreds of dollars on box seats to a sporting event and they had a complimentary buffet, is that food really free or did it cost you hundreds of dollars? Would you tell someone struggling with money they could get free food, they just need to go spend hundreds on sports tickets first?
Maybe one shouldn't be so willingly close-minded to the truth.
https://mashable.com/article/apple-messages-green-doj
https://www.sfgate.com/tech/article/apple-green-bubble-messa...
https://old.reddit.com/r/Anthropic/comments/1snorbg/the_bigg...
I don't know enough about distillation to understand how much this hinders/slows the process, but it sounds at least superficially plausible.
Honestly, I think its quite possible that models will be retrained with gaps in their knowledge. e.g. a coding model for commercial use probably doesn't need to have deep knowledge of biology, and training on biological sciences probably doesn't help those evals much.
Honestly, I'd welcome such an approach.
Strange reading that on HN and realizing I'm not on Facebook
The whole idea of the deep state is that it’s part of the state, ie government, so not private citizens, and they’re “deep” ie hidden below the layers of government. Thats the exact opposite of politicians and the ultra rich.
Also, your link specifically starts with:
""a hybrid association of government elements and parts of top-level industry and finance that is effectively able to govern the United States without reference to the consent of the governed as expressed through the formal political process."
which exactly how this was defined by your opponent.
OpenAI and Anthropic are beholden to the capitalist system they exist under and hence cannot compete on local models. Like you say, they must try to maximize shareholder value. China is unencumbered by that constraint.
But if you were in China, could you say you hate the Chinese Communist Party and China openly and as often as you like without imprisonment or worse?
We know the answer to that. So go ahead and trust China more than the U.S., but I think that is pure foolishness.
There was an outdated but relevant saying
'In America, you can criticize president Nixon anytime'
'Yes, but in Soviet Union you can also criticize Nixon anytime.'
The point is not that they're safer but that they're not a relevant concern in the same way. (According to OP)
Many technological advances weren't driven by capitalism, early computers and the internet were literally developed by the government.
Thats like Microsoft saying "Don't use Linux because selling an operating system is what matters"
But the thing is... I could be using any of the llms for my use - I'm using a middleware that lets me change providers only with a configuration change.
So it's going to be tough for USA ai companies to charge 5x to 20x (depending on what you're doing).
It begs the question because both its premise and assertion are already wrong. Has AI improved the industrial capacity of the US in order to improve the lives of its citizens? No it hasn't. Has AI increased the wealth of its citizens by being able to do laundry or any household task in a generalized way? No it hasn't. The only thing it's really done is make very narrow slices of white-collar work more fungible. In what way has AI been able to address existing shortcomings of the US?
https://www.federalreserve.gov/econres/notes/feds-notes/moni...
Based on a survey if the business uses AI "in any of its business functions". And for all uses of what they consider to be AI, not just LLMs.
You mean grand declarations like 'industrial capacity has increased'? Just because AI is present in the factory doesn't mean it's actually increased capacity.
Have you happened to purchase anything in the past 12 months, and looked at the Fed's inflation numbers?
The Fed doesn't issue inflation numbers. The usually cited headline inflation numbers (CPI) are from the Department of Labor’s Bureau of Labor Statistics, the ones used by the Fed as an input to monetary policy decisions (PCE) are issued by the Department of Commerce’s Bureau of Economic Analysis.
t. literally works on AI for industrial applications
How?
On a personal level, I simply do not trust the US anymore. I won't host any of my personal data in a US company. I don't want the US govt invading my personal privacy, and their corporations are constantly leaking and selling private data. I consider US to be rapidly approaching complete autocracy (on par with China) so US-hosted AI is a non-starter. And let's not forget local inference keeps getting more efficient, with higher context and TPS in the same amount of RAM. Within a year even small consumer machines will run local models good enough for basic coding, and in 3 years RAM prices will lower and everyone will be able to afford a decent rig.
Finally, open weight models are now good enough for daily work. They may never be as good as SOTA (SOTA will just keep increasing indefinitely), but that doesn't matter; my car may not be as fast as a Porsche but it still gets me to the grocery store and back. So I use non-US hosted model providers which provide open weights, which are both significantly cheaper than Anthropic/OpenAI, and actually allow me to use my subscriptions without a moat.
But yes, Anthropic/OpenAI are absolutely the new Oracle. They will win for US govt and Enterprise contracts. But that's far from the only users of AI.
And US absolutely has been xenophobic for years, by official federal policy. I'm really surprised you're not aware of it, but here's a small selection of examples:
- Both our elected and appointed leaders are white nationalists. Our president called all Mexicans murderers and rapists, said African migrants were eating random pets in a rural US town (they weren't, obviously, but it was intended to exacerbate xenophobia)
- Our federal government has a mandate using ICE to try to eject anyone with a Hispanic name from the country (has already deported US citizens based on being hispanic/latino). We even boot people seeing asylum, often exporting them to foreign prisons even if they've never had a criminal record. We have concentration camps now, filled entirely with foreigners, and people who have lived here for decades but were foreigners.
- We stopped accepting new visas from 75 countries. We may even expel you for social media posts we don't like, or for attending a protest that our citizens can attend. We increased travel bans for people from majority Muslim countries. H1-B visas have been rolled back to only the highest paying jobs, and you may need to pay a $15,000 bond. We also now collect and store foreigners' biometric data indefinitely.
- Let's not forget the tariffs on virtually all other nations, to say nothing of "America First" and the new "Greater North America doctrine".
I think you got lost in the rhetoric somewhere.
Tariffs are just the US adjusting to reality which other countries are slow to do. Free trade died all on its own, because the pandemic showed that critical industries were hollowed out by free trade in a way that could be appreciated from a national security perspective. That situation was favoring China too much, so we need to unwind that some.
Tariffs already existed in many countries in practice, so it's not like the US reinvented modern tariffs.
Pew [1] suggests that the changes around the start of 2025 were due increased restrictions on asylum applications under the previous admin and EOs by the current one to restrict new immigration. Given the rough numbers [2] of about 40k asylum grants per year in the early 2020s, I doubt the previous admin's actions are playing much of a role here.
Stating that none of it (immigration acceptance) changed under this administration might technically be true - with respect to the number of countries applying, but misses this point.
[0]: https://www.census.gov/newsroom/blogs/random-samplings/2026/...
[1]: https://www.pewresearch.org/short-reads/2025/08/21/key-findi...
[2]: https://usafacts.org/articles/how-many-people-seek-asylum-in...
If you feel like formulating a good argument about immigration, I'll listen, but you haven't provided one.
It can happen in Europe too, but the full fall is not that close.
The structure of the US makes it basically the single most secure democracy anywhere right now or in history. No country in Europe or Europe as a whole is even competitive by comparison. The main issue we're facing is that we are by far the primary target for foreign funded activism and systemic attacks, because China and Russia hated NGOs promoting color revolutions.
That is also part of the rule of law issue, but the system is overall managing quite well. It's all moving in slow motion, but many important metrics are going in the right direction, which we need as that's part of deterring China.
How do you figure? I hear you have roving gangs of masked thugs beating up random citizens with the backing of your government, that doesn't sound very democratically secure, especially with what healthcare costs over there.
So secure, in fact, that it has secured itself even against the influence of its own citizens.
Also, we have guns. LOTS of guns. The U.S. military's first and sole responsibility is to the constitution itself. If any state or the federal government tries to get rid of their constitutions, the military can rightfully take it over and re-establish a constitution.
There is no other country that's even remotely close to this secure.
This is just not true. It is failing visibly and loudly fast. It used to fail slowly but the process speeded up.
American administration supports Russia now. It praises Russian, Chinese, Belarus leaders again and again. It praises Orban. It hates last bastion of democracy - Europe.
China is not detered. Its power is growing while American one is going down. Trump openly admires its leader. China is celebrating current state of America.