I don’t know how OpenAI screwed this up. They had the best tech, the largest installed base, the best brand recognition.
And somehow instead of prosecuting the lead in all areas, they got all hubristic and sloppy and just failed to iterate on the core product, while also failing to respond quickly when Anthropic showed that coding agents are the flywheel that makes the whole company faster.
It’s like they thought they had an unassailable monopoly and speedran to the lazy incumbent position, all in a matter of months.
Anecdotally, I would actually argue tbe opposite - Anthropic is overrated, ass-kissed way too much here for mediocre coding abilities (especially for Elixir). ChatGPT most of the time one-shots complex solutions in comparison. The only reason why people shit on OpenAI so much is because of the defence deal, but, it's not like Anthropic is a saint either:
I’m so confused. Your link shows they are pushing for guardrails, what is bad about that? It is consistent with Anthropic safety-first principles, and what Dario wrote and talked about for the past decade or so? Could you be more direct with your criticism? Otherwise it’s hard to engage with
Claude Code is IMO the benchmark today. For all of the various contexts I’ve used it in it has mostly oneshot the tasks I’ve given it and is very user friendly for someone who is not a professional software engineer. To the extent it fails I can usually figure out quickly why and correct it at a high level.
I think Codex is a better fit for professional software engineers. It's able to one-shot larger, more complex tasks than Claude and also does better context management which is really important in a large codebase.
On the other hand, I think Claude is more friendly/readable and also still better at producing out-of-the-box nice looking frontend.
I think this is where we might have differing opinions. I'm a CTO by profession and I know what bad code is, so it is quite easy for me, based on my professional experience, point out when Claude generates bad code. And when you point it out, or ask it why it didn't take the correct/simpler approach - the response is always along the lines of "Oops, sorry!" or "You're absolutely right to question that..."
Yeah my CTO says similar things. I usually just tell him to add it to the backlog and move on. At the end of the day these tools save us 3-4 eng hires and that’s what the board cares about
Why pick elixir specifically here? I’m using opus/sonnet via Claude code for a moderately complex personal project built on phoenix and have had a good experience
Yeah, I've been building a fairly complex app with Claude and it has been great. Backend stack is a Go service, with TS front end and a solver running or-tools in Python.
I do think I do a good job of being very structured at breaking down my requirements and acceptance criteria (thanks dual lives as a devops and SRE guy and then PM). Extensive unit testing, discipline in use of sessions and memories and asking it to think of questions it should be asking me before even formulating a plan.
Claude is good, I'm definitely not saying it's bad. But if you work with LiveView, it will tend to choose more complexity over simplicity. Weirdly enough I have a feeling it's trained more on Python/Ruby (Object oriented paradigms) style code than functional code, so it tries to get things done not so functionally.
Strange take, I wasn’t lionizing Anthropic at all, and certainly wasn’t being preachy or moralistic.
I shit on OpenAI because they had a completely clear path to being dominant in consumer and enterprise, in chatbot and coding agent, and in text, image, and video. And somehow they screwed it up by being unfocused, unnecessarily slimy, drama-ridden, and slow to react to the market.
OpenAI has a lot of good tech and some good product. They may yet succeed. But they embody the lazy hare partying with celebs while the tortoises pull ahead in the race.
no i cant. chatgpt is a mobile app/website, not a model or agentic harness. if you are confusing these things then sadly you have no idea whats going on.
Sam lost the plot for me. He took too many interviews which led me to not trust him. Last straw came with him standing by Anthropic one day then throwing them under the bus the next. He showed little awareness on why that is problematic.
That's why I changed as well. I got really irritated how Altman tried to get the social credit by having principles, only to change them the moment it was convenient.
> On podcasts his attitude is basically “oh yeah all of you are basically fucked our products will take everyone’s jobs in a couple years.”
I also appreciate his honesty, and don't really understand why the others don't emulate it because there's no cost to them to be honest. At every level of society we've decided to stick our heads in the sand and pretend like this very large tsunami isn't racing toward the coast, so as someone producing this technology you can be honest (and mostly ignored by people in denial), or be cagey and mistrusted (like Sam Altman).
Because it isn’t honest, it is investor hype that these frontier labs need people to believe despite obviously hitting the sublinear part of the improvement curve.
“It’s so dangerous, we’ve reached AGI, we just have to release models that are obviously incapable of abstraction for your safety”
> Sam lost the plot for me. He took too many interviews which led me to not trust him. Last straw came with him standing by Anthropic one day then throwing them under the bus the next. He showed little awareness on why that is problematic.
It should have become clear to all that he was an untrustworthy person when he was fired from OpenAI by its then-board. My understanding is their complaint was he was lying, untrustworthy, and manipulative; and enough stories came out at the time to confirm that.
I've however been on the receiving end of some very odd board behavior and most founders I know have a story or two as well. I don't always take board decisions at face value. Reality is often askew, personality disorders run rampant in places of power, and there can be unknown incentives at play.
Brockman and his wife were the biggest donors to Donald Trump's Super PAC, MAGA Inc., in 2025 with each of them donating US$12.5 million. Brockman and his wife also donated $50 million to Leading the Future, a super PAC dedicated to AI deregulation that he helped found with Andreessen Horowitz co-founders Marc Andreessen and Ben Horowitz.
It's clearly because they didn't hire me after I applied :)
In all seriousness, I use Codex for work and Claude at home, and I feel like nowadays they're actually pretty competitive with each other. I don't know that it's that far behind.
I agree that they clearly erroneously assumed that no one would be able to catch up with them, though. OpenAI had such a head start that that should have been by itself a moat.
Yea. In my opinion the value provided for 20$ is better. I wonder how much of antropic value is the hype around claude code coming from every snake oil sales man promoting claude code as best to use with open claw to summarize your emails.
Just like OpenAI's original moat, I don't think that's particularly durable. I've already seen plenty of people swing back to preferring codex, and it'll probably swap again with the next model drop. Openclaw is potentially better integrated with ChatGPT at this point because of the explicit subscription support.
Nothing about AI mindshare is durable. As Anthropic tightens rate limits and/or raises prices, everyone will move to something else.
This week I’ve been hitting CC rate limits, and I switched to Codex with virtually no disruption to my workflow. It’s not good for Anthropic that I was able to do that!
One of the things I’ve been wanting to see is basically an estimate of their minimum revenue to meet investor expectations, too. I’ve heard people talking about using even cheaper local models aggressively to save on tokens and it increasingly makes me wonder if they’re caught in a vice where prices need to go up but if they raise them they’ll just shed usage to the competition, especially since at least Google has a much longer runway.
Also not like it’s a particularly good piece of tech. It was the first to show a new category. But jeebus the design and security are a nightmare. Any of the numerous other claws are better choices for anything serious.
Aside from the fabricated drama and the trend chasing, OpenAI still has the best overall model and API service. Anthropic is really good, no doubt. But gpt-5.4 is a better model than even Opus, even if its a marginal advantage. I use both.
I dunno, my experience mirrors the parent posters: we use opus for all our coding, but gpt 5.4 for all of our enterprise agentic work via api (much bigger amount of tokens). it just seems to be more optimized for this.
Classic SV hubris. Talk to OpenAI people and they’re so convinced they’re untouchable, they don’t bother worrying about things like revenue, or product strategy. All they cared about was being the first to AGI. Well it looks like that isn’t happening soon enough. And now they have zero moat except brand recognition, which is quickly getting eroded.
Coding assistants won't win this game. They sure will win the hearts of developers, but to scale you need mass adoption and products for which users want to pay substantially. OpenAI is falling behind in the small features in their chat and app offering and have failed to innovate in their expensive offerings.
Codex btw is getting very competitive. It is fast and no longer far behind.
Uber doesn't have to convert hundreds of millions of users to a paid plan - it was paid to begin with. Much easier to raise prices from a low base than to raise them from $0. When push came to shove, Uber also cut payments to drivers and replaced them with a tip screen. Is Nvidia going to accept tip annuities on ChatGPT responses instead of full payment on its chips?
Most importantly, Uber did not have fast-depreciating cars on its balance sheet - OAI meanwhile is planning to spend the GDP of a mid-size nation on owning fast-depreciating datacenters, which they deem necessary to cope with the demand for their services.
The strategic playbook of the web era said: Get a huge userbase of normies, then figure out how to monetise them (usually via advertising). OpenAI stumbled into the userbase via ChatGPT, but it's unclear if the strategy or the economics apply to AI. Anthropic tried to compete in the consumer market, but couldn't, so focussed on coding and enterprise, and it looks like that's actually turning into a smart choice, at least right now, because it turns out people will pay subscription costs for agents that do their job for them.
There are three possible paths that sort of substantiate current valuations:
1) Business: LLMs become essential to every company, and you become rich by selling the best enterprise tools to everyone.
2) Consumer: LLMs cannibalize search and a good chunk of the internet, so people end up interacting with your AI assistant instead of opening any websites. You start serving ads and take Google's lunch.
3) Superhuman AGI: you beat everyone else to the punch to build a life form superior to humans, this doesn't end up in a disaster, and you then steal underpants, ???, profit.
Anthropic is clearly betting on #1. Google decided to beat everyone else to #2, and they can probably do it better and more cheaply than others because of their existing infra and the way they're plugged into people's digital lives. And OpenAI... I guess banked on #3 and this is perhaps looking less certain now?
Any publicly available evidence to back that up? There have been post-exit blog posts from OpenAI employees on HN before and it did sound like the only black magic they use there is that many employees work 16 hrs a day during launch of new features. I know that some current Claude Code devs are doing interviews where they claim that they use Claude Code extensively but they clearly have a conflict of interest while they are still employed at Anthropic, so it would be like asking a barber if you need a haircut.
Despite what the folks here like to believe about themselves, I think the reality is we as attuned to what is in fashion and on trend as everyone else, just about different stuff. Last year it was Chatgpt, this year Claude is the new hotness. Things move so fast we barely have time to form our own opinion, so we fall back on what we read or hear from others. In 12 months who knows what it will be... Gemini? ¯\_(ツ)_/¯
Long term, my feeling is Anthropic's focus on enterprise is the most obviously lucrative but also least defensible application of LLMs. If (more likely when) open source models reach the point of being "good enough" then it's a race to the bottom on pricing. Maybe it will be like AWS vs GCP et al, but I kinda doubt it.
Is there any equivalent for turning an AI LLM into a sticky platform? Right now seems like it’s pretty easy to use harnesses and tooling across models, including open weight and locally running ones.
Investors do not care about the product, the users, etc. They care about cash. There are lots of ways to make cash that don't involve having a good product. But if you commit to spending a trillion dollars on hardware, then borrow hundreds of billions in the short term, and it turns out there's no way to recoup the cost, the investors go looking for better returns. This would've worked back in the old days of a bull market, angels looking for the next whale (with "modest" $5BN investments), and startups with no rivals. But in a bear market with multiple competitors trading on a commodity? Lol. Finally the bubble bursts.
Congrats to OpenAI for having this fake "beef" with itself through Anthropic, but Google is still going to win the most users. Users aren't so stupid that they will fall for this kind of obvious psychological manipulation in perpetuity.
Some of that, sure. But realistically, a lot people are just don't want to pay for every frontier model provider out there as they're released. Not just money, but also time trying them out. (Recommend people at least try out their multimodal model.)
It doesn't help that Google offers a bunch of confusing plans in multiple places. I ended up just pasting all their AI plan URLs, at least that I could find, into Claude so I didn't have to figure it out.
I pay for Google Workspace, so pretty much the Gemini Pro included with that suffices my use case. I can't say it will work for everyone, but I do use it for random tasks - all the way from wood working to building complex software projects, and so far I've rarely hit the limit too.
I guess it's access to Gemini, because that's the plan I'm on and you can access Gemini using gemini.google.com or their phone app (Android/iOS). I couldn't find information on the API, maybe it's on that page, but I couldn't find it from my phone browser. Anyway, I use OpenRouter pretty much, so I have no idea if the API is part of my plan.
I think Antigravity w/Gemini is a great product; it's been super useful on a bunch of my hobby projects. It's especially wonderful when writing firmware and needing to add support for a new chip. I can point it at a PDF datasheet and it'll do a much better job of reading it and parsing out all of the register fields than anything else. Saves me enormous amounts of time.
Thanks, been meaning to try it. I heard the limits on that is an issue and people are supposedly blowing the limits off way too easily? How has your experience been so far in this regard?
Yeah, in most threads you will see anyone recommending Gemini be downvoted. Ironically, Gemini (with Workspace subscription) is the only model that explicitly states it doesn't use your inputs to train their models right under the chat box. AFAIK no other provider does that explicitly - usually there is a hidden toggle in settings you will need to turn off.
TIL. That's cool. I mostly use Gemini 3 Flash for some background jobs because of the price/perf, but rooting for their models to improve. Competition is good.
I've finished (as in: it's done, it works, and I may never need to change it again) entire projects with ChatGPT and Codex. Sometimes it takes a lot of hand-holding to get there, but it does get there and (with the exception of 4o) it's been improving since the beginning.
In contrast: I can't even get Gemini Pro to give me any answers to the most primitive questions that aren't caked in prima facie lies without at least 4 interactions, in any context, ever. The output is consistently and ridiculously garish with its insessant self-contradictions. It seems to be impossible to actually get anywhere with it.
Anthropic is not meaningfully better. Their stance is “the good guys have to make money to be in the fight with the bad guys” and so they do all the things their perceived bad guys do. I don’t know how they can do any different, but we just trust them to be good? What is the difference?
Agreed. I don't like Dario. I enjoy his insights but he comes across as "we are the good guys" and "China is the bad guys". Therefore, we must eliminate our enemy in any way possible, including developing god so we can destroy their civilization first.
Just pure propaganda. In my opinion, not so much different than the opinions of the current US whitehouse regime.
If you look at the engineering, technology, operations, etc., this isn't really a fair fight. I think that's why Anthropic puts so much effort into the other narratives.
OAI can play the empire / Death Star angle and get away with it. They've actually got a scary battlestation and anyone interested in capitalizing on the galaxy is going to be drawn to the most powerful laser system in the fleet.
I don't have a problem with the non-engineering arguments, but I don't buy that one is more ethical than the other. All of these people are total maniacs. Not just Sam and Elon.
I just appreciate the honesty of Amodei telling us pretty much straight up we’re all fucked because the AIs are taking all the jobs in a couple years or less.
It's not pure hype. The linear trendline of AI in the last couple of years from chatbot to autocomplete to agentic coding does point to developer replacement in a couple years.
Now mind you a trendline is a predictor and the trends don't always travel in a line. The future is unknown but a trendline that makes the prediction of AI taking over in a couple of years is not an unrealistic prediction simply because of past progress. That is the most probably conclusion given the information we have.
Discarding that as just "hype" just means you're not being very rational or logical which is normal given we're on HN.
He's just another con-man. Fear of missing out is an amazing motivator for investors. The more he shouts doom, the more people with very deep pockets throw money at him. It's all evil.
> The large gap between OpenAI’s $852-billion valuation and Anthropic’s $380 billion has investors rushing to grab equity in the latter before it rises, according to Augment co-founder Adam Crawley.
Interesting, so there are a lot of people still eager to invest in valuations of well greater than a-quarter-trillion, but OpenAI's latest raise has sucked up all the oxygen for enthusiasm of that valuation going even higher.
Which could be a "dumb money" move ("competitor number lower, already-big-number is scary") or a "smart money" move ("Anthropic is gaining position-wise, and currently is lower valued, let's bet on the one we think is better positioned") or some mix of both.
OpenAI just raised a shit-ton so clearly there is plenty of money out there who don't think there's a bubble or even a blown opportunity there. But the wider community doesn't think they have the competition in the bag, while still being willing to invest in big-AI-cos at absolutely enormous valuations.
If local hardware/models get good enough to take 80%-90% of what people use subscriptions for today... hoo boy. Big-AI is a bet I wouldn't be confident placing billions on. Unless your horizon is more "wait for IPO or next raise or positive news, then get out ASAP" than "hold for 5+ years."
Did OpenAI really “raise” that much? The startup world is not my area of expertise but I remember reading language in the announcements that implied those dollar amounts where more of a conditional promise of money in the future instead of a check today.
Yeah if they sold 1/800th of the company for a billion dollars then they are valued at 800b even if they only have a billion dollars. It’s advantageous for investors to both buy in as cheaply as possible but also have future investors to buy in as expensive as possible to prop up a, perhaps inaccurate, valuation.
It's raised in the sense that some people made a pinky promise to give them cash. But those people also don't have the money and have to raise it from other places. It's largely SoftBank, Oracle, Microsoft and Nvidia, all of whom don't have big piggybanks full of hundreds of billions. They ask for loans based on the promise of making cash to pay for it, and that cash is based on people wanting to use OpenAI. So it's kind of a big financial circle jerk. (Debt, SPVs, loans from Nvidia (at high interest rates), etc)
> It's largely SoftBank, Oracle, Microsoft and Nvidia, all of whom don't have big piggybanks full of hundreds of billions.
Actually SoftBank, Microsoft and Nvidia literally have free cash sitting there.
NVIDIA for example had over $60B in audited, reported free cash flow in 2025[1]
> loans from Nvidia (at high interest rates),
Is this just something you are making up?
"NVIDIA intends to invest up to $100 billion in OpenAI as the new NVIDIA systems are deployed. The first phase is targeted to come online in the second half of 2026 using the NVIDIA Vera Rubin platform."[2]
The closes there is to waht you are saying is reporting that NVIDIA has discussed guaranteeing some of the loans OpenAI is taking to build data centers:
"Nvidia is discussing guaranteeing some of the loans that OpenAI is planning to take out in order to build its own data centers, The Wall Street Journal reported, citing people familiar with the matter."[3]
This of course is the opposite of NVIDIA loaning OpenAI money - if they did this they would be liable for OpenAI's debts.
Just because they have cash doesn't mean they're going to lay it all on the table for one risky business. Which is why the actual solution is mostly loans.
In being a guarantor, Nvidia isn't directly giving the money now, but it is promising to be liable for the loan if OpenAI defaults. That's a risk they have to budget for.
But in addition to being a guarantor, they're also loaning out chips:
"OpenAI plans to pay for Nvidia’s graphics processing units (GPUs) through lease arrangements, rather than upfront purchases.""By leasing the processors, OpenAI can spread its costs out over the useful life of the GPUs, which could be up to five years a person said, leaving Nvidia to bear more of the risk."
Nvidia is leasing them the GPUs, which is basically a loan.
"In addition to offering a cost-efficient way for OpenAI to access chips, Nvidia’s lease option and long-term commitment can help the company land better terms from banks when it comes to raising debt, a person said. "
"As a non-investment-grade startup that lacks positive cash flow, financing remains costly. OpenAI executives have called equity the most expensive way to fund data centers, and said that the company is preparing to take on debt to cover the remainder of the expansion."
"“Folks like Oracle are putting their balance sheets to work to create these incredible data centers you see behind us,” Friar said. “In Nvidia’s case, they’re putting together some equity to get it jumpstarted, but importantly, they will get paid for all those chips as those chips get deployed.”"
Nvidia leases (loans) them GPUs with a promise to pay for them later. Getting this lease (loan) from Nvidia helps them secure more debt.
"OpenAI’s path to a $500 billion private market valuation has been enabled by hefty investments from Microsoft and others that allow the company to burn billions of dollars in cash while building its AI models that power services including ChatGPT."
"Jamie Zakalik, an analyst at Neuberger Berman, said the Nvidia deal is the latest example of OpenAI raising money that it pours right back into the company providing the capital. Investors are concerned about the “circular nature of this deal goosing up everyone’s earnings and everyone’s numbers,” said Zakalik. “But it’s not actually creating anything.”"
They get a loan of chips, use it to get a loan of money from someone else, and spend that money on the company that loaned them the chips. It's a Ponzi scheme.
This is just a grab bag of different critisms to what you said.
Vendor financing is very standard across multiple industries. It's not a loan in the sense you originally stated because it's physical assets, not financing.
The circular nature of the deals is a valid concern but different to what you raised.
It's also a much smaller concern now with Anthropic hitting $30B ARR from non circular sources than it was in September 2025 when your article was written.
I am playing around with this at home right now. I think a lot of the latest improvements came with the harness, instead of AI.
The part I am working on is to have better tools and data to search over. Curated for my needs. Similar to the Karpathy post yesterday about his wiki. I am trying something similar and even qwen 3.5 is totally fine for most of what I do.
Disclaimer: I bought memory before the crisis started. Not sure if I would build my PC as is now..
These dumb models are just making people hooked on the thing. What you think they building all those datacenters, hm? When the real deal comes, all the zombies would want their even better cocaine intravenously at whatever exorbitant price.
Vibe coding requires the sota models to work at all, but someone who knows what they are doing and uses the AI more responsibly can absolutely use the cheaper Chinese models for coding, and they’re often faster too. If I was one of the big players my entire focus would be on lobbying for regulation and outright banning of local models.
Yeah, Qwen3 coder for Claude Code and 3.5 for OpenClaw replaced my full-stack use of Opus 4.6 already; it's fine for basic web apps, k8s/docker infra setup, optimizing AI models etc. with only slightly higher error rate than Opus. Upcoming 3.6 together with Gemma4 might make it even better (still to test). OpenAI's memory spot market play might have been directed at local inference as well.
Look for Deepseek 4 when it drops, I’m curious how good it will be.
The thing is, if you’re using AI responsibly today you’re already breaking down tasks to such a granular level that you don’t need the power of Opus. You can save that for deeper research tasks.
Based on the current DeepSeek website I suspect it's not going to be great as their current model (V3.4? V4-mini?) often forgets or changes facts explicitly mentioned in the conversation which R1 never did. It's better than R1 at math or coding, but nearly unusable for deep conversation. I suspect they pushed MLA or linear attention too much, or quantize a lot more than before.
> Vibe coding requires the sota models to work at all
I've been hearing that vibe coding is the future for at least the past year but somehow it always requires the newest model, and all the "old" models produce trash results. Why is that?
Because expectations on what could feasibly be vibe coded have changed. Before with older, worse models it was only toy projects that were expected but now with better models it's much larger enterprise apps.
Both of these valuations are absolutely absurd. I guess Anthropic looks good in comparison, but I don't want to hold that bag.
The Chinese models are catching up in quality while being a fraction of the price. The market will speak, how many devices that contributed to this thread were made in the USA?
Sure you can argue the Chinese companies are heavily subsidized, but no major LLM lab is remotely close to making a profit this decade.
I switched to Anthropic briefly to try out Code, it was great (having never tried such a product) and better than their chat bot which found prettier and nicer to talk to but far more often wrong.
Then shortly after Codex was released which made that more accessible and instantly preferred that compared to Code since much more generous allotments for $20 plan. Claude constantly kept hitting. And Code actually had a more robust UI and was more accurate when doing the same project side by side vs Claude Code.
But imagine many haven't tried Codex recently since all we hear about is Claude Code. So while they may have momentum, at least for me with no stake in the game, Codex finding far better, but I suppose that could all change again on a whim.
My loose understanding as someone adjacent to the AI model space is that you have good models that are costly and cheap models that are decent, so a lot of the publicly visible fights where Claude and ChatGPT leapfrog each other is the companies doing cost-benefit of how much optimization to do on the models before your userbase revolts because the agent "used to be great and now kinda sucks".
As a small business owner whose team is entirely in Google Workspace (Drive, Gmail, Chat -- so inbuilt RAG right there), I wonder if Gemini will be the darkhorse. As a user Gemini's a distinct third in "AI smarts", but most business owners aren't power users who are gonna setup Codex or Code to slurp up their work emails and internal docs/SOPs.
The article feels a touch clickbait-y since people love a good fight between the top players and OAI's lost a buncha public goodwill over the past year.
As a government based person who has witnessed multiple states (not in the US) move all operations off GCP because Google doesnt address sovereign risk, local data and hosting privacy requirements,and contracts well at all, where Microsoft and AWS do, I doubt Gemini will have as large a dark horse moment as it could. Copilot Enterprise can span across similar domains, and whilst it is very expensive per user it has the benefit of having existing contracts in place which it can bolt in to.
Out of curiosity, have you compared the relative effectiveness of ChatGPT and Claude vs Copilot? Given your existing enterprise contract, does copilot have a monopoly on your AI usage due to its superior compliance?
In the handful of contexts Ive seen across departments and states, no, ive not seen such a comparison. Ive seen state custom implementations of GPT (eg 4o), and Copilot, but not Claude. Claude is likely out there for the same reason Google is - navigating those same localisation / legal issues is complex for government departments.
Copilot has a far easier path to monopoly because it can be pushed down existing Microsoft contracts, MS has put in the work for international jurisdictions in terms of data compliance in existing products and thus can extend upon that with Copilot, and they have a foothold in a good many places. One thing that is interesting with Copilot is that if you set it to Work mode, it can treat your organisational SharePoint sites and OneDrive locations as sources, but do so in a manner that respects what you as a user has access to.
> large gap between OpenAI’s $852-billion valuation and Anthropic’s $380 billion
IIRC Anthropic's revenue is either roughly at parity with or larger than OpenAI's, and Anthropic is growing faster[1]. All indicators are that Anthropic should be worth more than OpenAI. Given that, one could reasonably expect the relative valuations to change a great deal. In any case, it's not clear why OpenAI would command such a price premium over Anthropic.
Anthropic's run rate increased from $12B to $19B in the period between February 12 and the end of the month. If the implied growth rates held through March, Anthropic may well be larger than OpenAI now.
Both companies have the same issue: the unit economics are bad, and there’s no moat.
I’ve been primarily on Claude for the last 6 months or so, but have been hitting rate limits. I switched work to Codex seamlessly, just like I could switch to any other provider seamlessly
The unit economics at this point are about utilization. Their cost is well below what they’re charging, but only when there is enough traffic to keep the GPUs busy. So the game is about increasing demand to level the load.
GPT has negligible moat because they gave up on all their integrations. Claude code is starting to develop one as people start to build things that require Claude Code specifically, not just any LLM.
> Claude code is starting to develop one as people start to build things that require Claude Code specifically, not just any LLM.
I hate to be a "source?" guy, but I'm curious if you have any examples of this. Skills and MCP are really the only extensions on CC itself I'm aware of, and these are both supported in Codex.
Things like Dispatch / remote sessions is something CC has that Codex does not, but these features are quite easy to replicate (and I expect Codex to do so in short order).
I agree that’s a great question which I don’t really know the answer to. These tools have been moving in lock step for some time now. One will innovate and within 2-3 weeks the others have that feature. Where I sit the mind share all seems to be going to Claude though. The moat develops when people build something that only works on one - even if the others have the same features it doesn’t matter unless they’re literally binary compatible. Skills are just prompts at the end of the day with nothing more specialized than a file naming convention.
Having written several orchestrators I’ll say that the code to invoke the tool is pretty equivalent but it’s the details that matter. Exact CLI flags and json fields.
let me know when they scrap the data centers, id love to get some good deals on hvac equiptment. these companies cannot possibly make enough money when you can run something on your own computer that works mostly as good
IMO This says a lot more about the investor herd mentality and FOMO than it does about the technological or any other kind of merit of Anthropic vs OpenAI.
I am not a finance pro, but is it not normal and expected for secondary equity markets to be hesitant against large unloads of pre-money company stocks, even in cases where the company in question isn't in water like OpenAI is? Does this not say more about the failure of the investors themselves in balancing their portfolio properly than OpenAI itself?
The latest Opus routinely tells me the latest GPT Pro responses are much better. The GPT responses cost 10x more than GPT at least. And GPT takes 10’s of minutes. So unless and until I’m needing and ready for a really expensive “math checker” it gets left alone.
People are so late to Anthropic, I used Claude 1.3 back when it released, and I've stood by Claude since the early days. I don't think people quite realized the potential of Anthropic
Silicon Valley valuation numbers are entirely made up to boost their IPO. Actual value is only determined after these companies go public.
Allbirds, the shoe brand, went public at $4bn in 2021 and sold its assets last week for $39 million.
> Neil Saunders, the managing director of the consultancy and market analysis firm GlobalData said: “Allbirds has gone from being a high flyer to a dead parrot.” He said its early success “was driven by Silicon Valley hype, more than deep popularity with consumers in the American hinterland”.
Odd timing...Everything I've read about Claude the last several days suggests that its users are disappointed, even furious at what's happened to its performance.
Self-edit: a couple thousand session hours is probably hyperbolic; it's likely closer to a thousand after discussing usage with another user in a separate thread.
Maybe a better metric: Across work + personal I have 1,000+ sessions in the last 30 days, longest session was 3d 20h
Same, mostly. However today in particular Claude can't do front-end development with any competence. Never seen this before. I think the rumor is they are rolling out a new model and have to divide their infra across the new model vs the current model.
not to mention the drama that's followed the company from the very beginning. It think its getting to a point where the character issues can no longer be ignored bc it's directly affecting business
I think the IPO and subsequent quarterly earnings where they will be pressured by analysts will pop it all.
I was watching a recent Jensen Huang Q&A with analysts and it was essentially “just trust me bro”. They’re all interconnected - once there’s a correction for one player, all get affected.
Can’t wait to finally get this over with so we can finally move on.
The gap between hype and reality needs to be corrected.
Nah you see, the NASDAQ is giving them a out just like SpaceX. Short-circuit the index joining time and allow them to get into index funds within days. After that you shift all the losses onto 401ks as the original investors cash out and the 401ks blindly buy the shares as part of the index.
I wonder how much of this is associated with Scam Altman's personal negative PR and Anthropic's recent PR wins.
I'm inclined to think there isn't much of an association becauss investors don't seem very concerned with morality, but I know ~dozen developers that either switched to, or started using Claude in the past month or so, while not knowing anyone that uses Codex.
They want to see the CEO communicate a path to profitability. Anthropic has - purely by focusing.
OAI in contrast is all over the place and they haven’t shown they’ve learned.
Zuckerberg got punished for his metaverse nonsense and investors were correct to be skeptical and reflect that in the stock price. Altman thinks he’s a god and the rules don’t apply to him. More fool him.
I think it's hard to grasp just how much money he poured into this folly. $80 Billion. For an idea that any random person on the street could have torn apart. He read Ready Player 1 and decided that was what the future should be. Then he tried to make it. He ruined the Oculus brand and stole all the oxygen from other VR development just to make a lesser copy of our own world, but one where he could be admin.
Seattle just opened a new transit line, spanning from downtown Seattle to Bellevue. It's 33 miles long and cost $3.7 billion. It was already carrying 2 million people per year [0] even before the entire line was finished and on the opening day after completion it reached 200,000 riders.
Further from home, the entirety of the Artemis II program cost $93 billion [1] and lists an estimated $4 billion per launch.
One last example - the panama canal would cost an estimated $50 - $75 billion if you built it today. [2] The panama canal reshaped trade routes and the global economy.
Where is Meta's moon landing? Where is their impact? Zuckerberg spent enough money to build a new transit system for an entire city. Enough money to land astronauts on the moon. Enough to connect oceans together. He poured this money down a hole chasing his own vanity and a vision of the future that no one wanted.
How, after all that, did he get punished? He's still CEO. "Meta" is still overvalued. They're still regarded as a serious tech company.
> Zuckerberg got punished for his metaverse nonsense and investors were correct to be skeptical and reflect that in the stock price.
Did they? It's been a little less than 5 years since Zuckerburg announced the all-in plan on the metaverse. In that 5 year period, Meta's stock price has gone up by 84%... which is middle of the pack for MAGMA, less than Apple (91%) and Google (157%), but above Amazon (24%) and Microsoft (47%). It's also comfortably above the S&P 500 (59%).
I don’t know how OpenAI screwed this up. They had the best tech, the largest installed base, the best brand recognition.
And somehow instead of prosecuting the lead in all areas, they got all hubristic and sloppy and just failed to iterate on the core product, while also failing to respond quickly when Anthropic showed that coding agents are the flywheel that makes the whole company faster.
It’s like they thought they had an unassailable monopoly and speedran to the lazy incumbent position, all in a matter of months.
Anecdotally, I would actually argue tbe opposite - Anthropic is overrated, ass-kissed way too much here for mediocre coding abilities (especially for Elixir). ChatGPT most of the time one-shots complex solutions in comparison. The only reason why people shit on OpenAI so much is because of the defence deal, but, it's not like Anthropic is a saint either:
https://www.cnbc.com/2026/02/12/anthropic-gives-20-million-t...
> it's not like Anthropic is a saint either:
I’m so confused. Your link shows they are pushing for guardrails, what is bad about that? It is consistent with Anthropic safety-first principles, and what Dario wrote and talked about for the past decade or so? Could you be more direct with your criticism? Otherwise it’s hard to engage with
Claude Code is IMO the benchmark today. For all of the various contexts I’ve used it in it has mostly oneshot the tasks I’ve given it and is very user friendly for someone who is not a professional software engineer. To the extent it fails I can usually figure out quickly why and correct it at a high level.
I think Codex is a better fit for professional software engineers. It's able to one-shot larger, more complex tasks than Claude and also does better context management which is really important in a large codebase.
On the other hand, I think Claude is more friendly/readable and also still better at producing out-of-the-box nice looking frontend.
> not a professional software engineer
I think this is where we might have differing opinions. I'm a CTO by profession and I know what bad code is, so it is quite easy for me, based on my professional experience, point out when Claude generates bad code. And when you point it out, or ask it why it didn't take the correct/simpler approach - the response is always along the lines of "Oops, sorry!" or "You're absolutely right to question that..."
Yeah my CTO says similar things. I usually just tell him to add it to the backlog and move on. At the end of the day these tools save us 3-4 eng hires and that’s what the board cares about
Why pick elixir specifically here? I’m using opus/sonnet via Claude code for a moderately complex personal project built on phoenix and have had a good experience
Yeah, I've been building a fairly complex app with Claude and it has been great. Backend stack is a Go service, with TS front end and a solver running or-tools in Python.
I do think I do a good job of being very structured at breaking down my requirements and acceptance criteria (thanks dual lives as a devops and SRE guy and then PM). Extensive unit testing, discipline in use of sessions and memories and asking it to think of questions it should be asking me before even formulating a plan.
Claude is good, I'm definitely not saying it's bad. But if you work with LiveView, it will tend to choose more complexity over simplicity. Weirdly enough I have a feeling it's trained more on Python/Ruby (Object oriented paradigms) style code than functional code, so it tries to get things done not so functionally.
Anecdotally, Claude has worked far better for our elixir team than the others we’ve tried.
Strange take, I wasn’t lionizing Anthropic at all, and certainly wasn’t being preachy or moralistic.
I shit on OpenAI because they had a completely clear path to being dominant in consumer and enterprise, in chatbot and coding agent, and in text, image, and video. And somehow they screwed it up by being unfocused, unnecessarily slimy, drama-ridden, and slow to react to the market.
OpenAI has a lot of good tech and some good product. They may yet succeed. But they embody the lazy hare partying with celebs while the tortoises pull ahead in the race.
cant take anyone seriously who thinks
> ChatGPT most of the time one-shots complex solutions in comparison
is an intelligible sentence.
Hmm. I can definitely understand what the author was saying.
Paraphrased: ChatGPT often completes complex solutions in one try whereas Claude does not (or performs less well).
I guess you can’t take me seriously?
no i cant. chatgpt is a mobile app/website, not a model or agentic harness. if you are confusing these things then sadly you have no idea whats going on.
Sam lost the plot for me. He took too many interviews which led me to not trust him. Last straw came with him standing by Anthropic one day then throwing them under the bus the next. He showed little awareness on why that is problematic.
That's why I changed as well. I got really irritated how Altman tried to get the social credit by having principles, only to change them the moment it was convenient.
I have appreciated Amodei’s brutal honesty about their intentions.
On podcasts his attitude is basically “oh yeah all of you are basically fucked our products will take everyone’s jobs in a couple years.”
Altman is a lot more coy and comes across as saying what’s politically expedient at any given point in time.
> On podcasts his attitude is basically “oh yeah all of you are basically fucked our products will take everyone’s jobs in a couple years.”
I also appreciate his honesty, and don't really understand why the others don't emulate it because there's no cost to them to be honest. At every level of society we've decided to stick our heads in the sand and pretend like this very large tsunami isn't racing toward the coast, so as someone producing this technology you can be honest (and mostly ignored by people in denial), or be cagey and mistrusted (like Sam Altman).
Because it isn’t honest, it is investor hype that these frontier labs need people to believe despite obviously hitting the sublinear part of the improvement curve.
“It’s so dangerous, we’ve reached AGI, we just have to release models that are obviously incapable of abstraction for your safety”
We still think Amodei is honest and his hype recycling is not ultimately incredibly self serving?
Upton Sinclair[1] has something relevant for Amodei.
Funnily enough, the same thing can be used against those who criticize his view on AI. I wonder if there is a word for this.
[1] https://quoteinvestigator.com/2017/11/30/salary/
> Sam lost the plot for me. He took too many interviews which led me to not trust him. Last straw came with him standing by Anthropic one day then throwing them under the bus the next. He showed little awareness on why that is problematic.
It should have become clear to all that he was an untrustworthy person when he was fired from OpenAI by its then-board. My understanding is their complaint was he was lying, untrustworthy, and manipulative; and enough stories came out at the time to confirm that.
Yeah, it likely should have.
I've however been on the receiving end of some very odd board behavior and most founders I know have a story or two as well. I don't always take board decisions at face value. Reality is often askew, personality disorders run rampant in places of power, and there can be unknown incentives at play.
For me, it was when I found out Greg Brockman's MAGA donations. From wikipedia (https://en.wikipedia.org/wiki/Greg_Brockman#Personal_life):
Brockman and his wife were the biggest donors to Donald Trump's Super PAC, MAGA Inc., in 2025 with each of them donating US$12.5 million. Brockman and his wife also donated $50 million to Leading the Future, a super PAC dedicated to AI deregulation that he helped found with Andreessen Horowitz co-founders Marc Andreessen and Ben Horowitz.
It's clearly because they didn't hire me after I applied :)
In all seriousness, I use Codex for work and Claude at home, and I feel like nowadays they're actually pretty competitive with each other. I don't know that it's that far behind.
I agree that they clearly erroneously assumed that no one would be able to catch up with them, though. OpenAI had such a head start that that should have been by itself a moat.
Yea. In my opinion the value provided for 20$ is better. I wonder how much of antropic value is the hype around claude code coming from every snake oil sales man promoting claude code as best to use with open claw to summarize your emails.
Claude Code became the default brand for an AI coding harness, much like ChatGPT was synonymous with AI chat bot.
Even now when I hear Codex I have to stop and think “oh yeah that’s OpenAI’s competitor to Claude Code.”
Yep, exactly. It became industry standard.
Does it matter that codex is now as good as claude code?
Check dev spaces like twitter and discord and all anyone talks about is claude-code, openclaw, opus 4.6 etc.
The mindshare went to anthropic.
Just like OpenAI's original moat, I don't think that's particularly durable. I've already seen plenty of people swing back to preferring codex, and it'll probably swap again with the next model drop. Openclaw is potentially better integrated with ChatGPT at this point because of the explicit subscription support.
Nothing about AI mindshare is durable. As Anthropic tightens rate limits and/or raises prices, everyone will move to something else.
This week I’ve been hitting CC rate limits, and I switched to Codex with virtually no disruption to my workflow. It’s not good for Anthropic that I was able to do that!
One of the things I’ve been wanting to see is basically an estimate of their minimum revenue to meet investor expectations, too. I’ve heard people talking about using even cheaper local models aggressively to save on tokens and it increasingly makes me wonder if they’re caught in a vice where prices need to go up but if they raise them they’ll just shed usage to the competition, especially since at least Google has a much longer runway.
That is 100% what I think is going to happen.
I wrote about it here: https://tombedor.dev/open-source-models/
that's why openai bought openclaw
I mean they hired the guy who created it. It's not exactly like openclaw is a real product.
Sure it is.. you can download and use it right now. Helps you understand why people are interested in it, at the very least.
Also not like it’s a particularly good piece of tech. It was the first to show a new category. But jeebus the design and security are a nightmare. Any of the numerous other claws are better choices for anything serious.
Aside from the fabricated drama and the trend chasing, OpenAI still has the best overall model and API service. Anthropic is really good, no doubt. But gpt-5.4 is a better model than even Opus, even if its a marginal advantage. I use both.
The advantage is marginal and sporadic, and its slower. Why would I use it over anthropic?
I dunno, my experience mirrors the parent posters: we use opus for all our coding, but gpt 5.4 for all of our enterprise agentic work via api (much bigger amount of tokens). it just seems to be more optimized for this.
[flagged]
It’s much cheaper, for starters.
Do you feel the companies' positionings are only marginally different in the same way the product is only marginally different?
These days, yes.
Classic SV hubris. Talk to OpenAI people and they’re so convinced they’re untouchable, they don’t bother worrying about things like revenue, or product strategy. All they cared about was being the first to AGI. Well it looks like that isn’t happening soon enough. And now they have zero moat except brand recognition, which is quickly getting eroded.
Coding assistants won't win this game. They sure will win the hearts of developers, but to scale you need mass adoption and products for which users want to pay substantially. OpenAI is falling behind in the small features in their chat and app offering and have failed to innovate in their expensive offerings.
Codex btw is getting very competitive. It is fast and no longer far behind.
The reality is given how much OAI has raised, they have to get to a place where they are doing insane revenues…
We’re talking on the level of meta, google and probably more if they keep raising money.
They really went all in with hubris and they’re gonna get punished eventually.
Just like Uber did?
Uber doesn't have to convert hundreds of millions of users to a paid plan - it was paid to begin with. Much easier to raise prices from a low base than to raise them from $0. When push came to shove, Uber also cut payments to drivers and replaced them with a tip screen. Is Nvidia going to accept tip annuities on ChatGPT responses instead of full payment on its chips?
Most importantly, Uber did not have fast-depreciating cars on its balance sheet - OAI meanwhile is planning to spend the GDP of a mid-size nation on owning fast-depreciating datacenters, which they deem necessary to cope with the demand for their services.
The strategic playbook of the web era said: Get a huge userbase of normies, then figure out how to monetise them (usually via advertising). OpenAI stumbled into the userbase via ChatGPT, but it's unclear if the strategy or the economics apply to AI. Anthropic tried to compete in the consumer market, but couldn't, so focussed on coding and enterprise, and it looks like that's actually turning into a smart choice, at least right now, because it turns out people will pay subscription costs for agents that do their job for them.
There are three possible paths that sort of substantiate current valuations:
1) Business: LLMs become essential to every company, and you become rich by selling the best enterprise tools to everyone.
2) Consumer: LLMs cannibalize search and a good chunk of the internet, so people end up interacting with your AI assistant instead of opening any websites. You start serving ads and take Google's lunch.
3) Superhuman AGI: you beat everyone else to the punch to build a life form superior to humans, this doesn't end up in a disaster, and you then steal underpants, ???, profit.
Anthropic is clearly betting on #1. Google decided to beat everyone else to #2, and they can probably do it better and more cheaply than others because of their existing infra and the way they're plugged into people's digital lives. And OpenAI... I guess banked on #3 and this is perhaps looking less certain now?
But will they pay the unsubsidized cost when anthropic needs to turn a profit?
And they actually can’t increase the price much.
Token generation is the metric Jensen Huang keeps pushing to temper analysts, which also affect nvidia’s future expected cash flows of course.
If increasing the price causes that metric to drop, the whole narrative falls apart and fear will spread in the stock market.
They’re all racing very close to the edge. Some closer than others.
Agents increase the velocity of OpenAI and Anthropic; whomever has the best in-house agent moves the quickest.
Any publicly available evidence to back that up? There have been post-exit blog posts from OpenAI employees on HN before and it did sound like the only black magic they use there is that many employees work 16 hrs a day during launch of new features. I know that some current Claude Code devs are doing interviews where they claim that they use Claude Code extensively but they clearly have a conflict of interest while they are still employed at Anthropic, so it would be like asking a barber if you need a haircut.
Look at the number of features (PRs) being pushed by these companies.
Despite what the folks here like to believe about themselves, I think the reality is we as attuned to what is in fashion and on trend as everyone else, just about different stuff. Last year it was Chatgpt, this year Claude is the new hotness. Things move so fast we barely have time to form our own opinion, so we fall back on what we read or hear from others. In 12 months who knows what it will be... Gemini? ¯\_(ツ)_/¯
Long term, my feeling is Anthropic's focus on enterprise is the most obviously lucrative but also least defensible application of LLMs. If (more likely when) open source models reach the point of being "good enough" then it's a race to the bottom on pricing. Maybe it will be like AWS vs GCP et al, but I kinda doubt it.
Same thing happened to Blackberry. Tech head start wasn't that big and product wasn't that sticky.
iOS had an API and a platform.
Is there any equivalent for turning an AI LLM into a sticky platform? Right now seems like it’s pretty easy to use harnesses and tooling across models, including open weight and locally running ones.
> Is there any equivalent for turning an AI LLM into a sticky platform?
Not with AI specifically. Having a large, private, data store (e.g. GMail, Google Drive) is the biggest advantage.
Investors do not care about the product, the users, etc. They care about cash. There are lots of ways to make cash that don't involve having a good product. But if you commit to spending a trillion dollars on hardware, then borrow hundreds of billions in the short term, and it turns out there's no way to recoup the cost, the investors go looking for better returns. This would've worked back in the old days of a bull market, angels looking for the next whale (with "modest" $5BN investments), and startups with no rivals. But in a bear market with multiple competitors trading on a commodity? Lol. Finally the bubble bursts.
Congrats to OpenAI for having this fake "beef" with itself through Anthropic, but Google is still going to win the most users. Users aren't so stupid that they will fall for this kind of obvious psychological manipulation in perpetuity.
What kind of OI slop is this?
5.4 Extra high >> Opus 4.6
Depends on your work flow.
I find that for human in the loop Gemini beats both.
Been my experience as well, but generally the anti-Google sentiment here is pretty loud so you'll never see anyone praising Gemini here pretty much
Some of that, sure. But realistically, a lot people are just don't want to pay for every frontier model provider out there as they're released. Not just money, but also time trying them out. (Recommend people at least try out their multimodal model.)
It doesn't help that Google offers a bunch of confusing plans in multiple places. I ended up just pasting all their AI plan URLs, at least that I could find, into Claude so I didn't have to figure it out.
I pay for Google Workspace, so pretty much the Gemini Pro included with that suffices my use case. I can't say it will work for everyone, but I do use it for random tasks - all the way from wood working to building complex software projects, and so far I've rarely hit the limit too.
What Gemini Pro comes with Workspace? Is there api allowance? Different than free users?
I guess if you choose the business standard, you get "Pro access to features & models with enterprise-grade security & privacy"
https://knowledge.workspace.google.com/admin/getting-started...
It says "Gemini App" but I am not sure what it's referring to.
I guess it's access to Gemini, because that's the plan I'm on and you can access Gemini using gemini.google.com or their phone app (Android/iOS). I couldn't find information on the API, maybe it's on that page, but I couldn't find it from my phone browser. Anyway, I use OpenRouter pretty much, so I have no idea if the API is part of my plan.
Thank you.
I think Antigravity w/Gemini is a great product; it's been super useful on a bunch of my hobby projects. It's especially wonderful when writing firmware and needing to add support for a new chip. I can point it at a PDF datasheet and it'll do a much better job of reading it and parsing out all of the register fields than anything else. Saves me enormous amounts of time.
Thanks, been meaning to try it. I heard the limits on that is an issue and people are supposedly blowing the limits off way too easily? How has your experience been so far in this regard?
Is anti-Google sentiment still pretty loud? People seem excited about Gemini catching up + Gemma 4
Yeah, in most threads you will see anyone recommending Gemini be downvoted. Ironically, Gemini (with Workspace subscription) is the only model that explicitly states it doesn't use your inputs to train their models right under the chat box. AFAIK no other provider does that explicitly - usually there is a hidden toggle in settings you will need to turn off.
The corporate subscription for ChatGPT says the same thing. And I would be shocked if it wasn't the same for a corporate agreement for Claude.
TIL. That's cool. I mostly use Gemini 3 Flash for some background jobs because of the price/perf, but rooting for their models to improve. Competition is good.
The catch: If you don't pay any subscriptions to Google, they will use your data for training their models. Agreed on competition being good.
Google does not know how to sell.
They have multiple offerings, they probably will kill some of them very soon. There is no reason to waste your time and money on Google.
I pay for both ChatGPT and Gemini.
I've finished (as in: it's done, it works, and I may never need to change it again) entire projects with ChatGPT and Codex. Sometimes it takes a lot of hand-holding to get there, but it does get there and (with the exception of 4o) it's been improving since the beginning.
In contrast: I can't even get Gemini Pro to give me any answers to the most primitive questions that aren't caked in prima facie lies without at least 4 interactions, in any context, ever. The output is consistently and ridiculously garish with its insessant self-contradictions. It seems to be impossible to actually get anywhere with it.
What am I doing wrong here?
Anthropic is not meaningfully better. Their stance is “the good guys have to make money to be in the fight with the bad guys” and so they do all the things their perceived bad guys do. I don’t know how they can do any different, but we just trust them to be good? What is the difference?
The difference is no mass surveillance of US Citizens and no killing weapons without human supervision.
OpenAI is fine with those as long as they are "legal"... So pretty much they don't care at all.
I agree Anthropic is no saint but it's much, much better than OpenAI.
Agreed. I don't like Dario. I enjoy his insights but he comes across as "we are the good guys" and "China is the bad guys". Therefore, we must eliminate our enemy in any way possible, including developing god so we can destroy their civilization first.
Just pure propaganda. In my opinion, not so much different than the opinions of the current US whitehouse regime.
If you look at the engineering, technology, operations, etc., this isn't really a fair fight. I think that's why Anthropic puts so much effort into the other narratives.
OAI can play the empire / Death Star angle and get away with it. They've actually got a scary battlestation and anyone interested in capitalizing on the galaxy is going to be drawn to the most powerful laser system in the fleet.
I don't have a problem with the non-engineering arguments, but I don't buy that one is more ethical than the other. All of these people are total maniacs. Not just Sam and Elon.
I just appreciate the honesty of Amodei telling us pretty much straight up we’re all fucked because the AIs are taking all the jobs in a couple years or less.
Honesty? Sounds like investor hype to me.
I of course cannot read his mind but he has been very consistent, even before he was raising money
It's not pure hype. The linear trendline of AI in the last couple of years from chatbot to autocomplete to agentic coding does point to developer replacement in a couple years.
Now mind you a trendline is a predictor and the trends don't always travel in a line. The future is unknown but a trendline that makes the prediction of AI taking over in a couple of years is not an unrealistic prediction simply because of past progress. That is the most probably conclusion given the information we have.
Discarding that as just "hype" just means you're not being very rational or logical which is normal given we're on HN.
He's just another con-man. Fear of missing out is an amazing motivator for investors. The more he shouts doom, the more people with very deep pockets throw money at him. It's all evil.
Approximately same service for half the valuation is definitely better. That said, Anthropomorphic is also very overvalued
somewhat better leadership, I think
> The large gap between OpenAI’s $852-billion valuation and Anthropic’s $380 billion has investors rushing to grab equity in the latter before it rises, according to Augment co-founder Adam Crawley.
Interesting, so there are a lot of people still eager to invest in valuations of well greater than a-quarter-trillion, but OpenAI's latest raise has sucked up all the oxygen for enthusiasm of that valuation going even higher.
Which could be a "dumb money" move ("competitor number lower, already-big-number is scary") or a "smart money" move ("Anthropic is gaining position-wise, and currently is lower valued, let's bet on the one we think is better positioned") or some mix of both.
OpenAI just raised a shit-ton so clearly there is plenty of money out there who don't think there's a bubble or even a blown opportunity there. But the wider community doesn't think they have the competition in the bag, while still being willing to invest in big-AI-cos at absolutely enormous valuations.
If local hardware/models get good enough to take 80%-90% of what people use subscriptions for today... hoo boy. Big-AI is a bet I wouldn't be confident placing billions on. Unless your horizon is more "wait for IPO or next raise or positive news, then get out ASAP" than "hold for 5+ years."
Did OpenAI really “raise” that much? The startup world is not my area of expertise but I remember reading language in the announcements that implied those dollar amounts where more of a conditional promise of money in the future instead of a check today.
Yeah if they sold 1/800th of the company for a billion dollars then they are valued at 800b even if they only have a billion dollars. It’s advantageous for investors to both buy in as cheaply as possible but also have future investors to buy in as expensive as possible to prop up a, perhaps inaccurate, valuation.
The “comitted capital” is about as real as the billion dollars that Disney invested into OpenAI for Sora.
It's raised in the sense that some people made a pinky promise to give them cash. But those people also don't have the money and have to raise it from other places. It's largely SoftBank, Oracle, Microsoft and Nvidia, all of whom don't have big piggybanks full of hundreds of billions. They ask for loans based on the promise of making cash to pay for it, and that cash is based on people wanting to use OpenAI. So it's kind of a big financial circle jerk. (Debt, SPVs, loans from Nvidia (at high interest rates), etc)
This isn't right
> It's largely SoftBank, Oracle, Microsoft and Nvidia, all of whom don't have big piggybanks full of hundreds of billions.
Actually SoftBank, Microsoft and Nvidia literally have free cash sitting there.
NVIDIA for example had over $60B in audited, reported free cash flow in 2025[1]
> loans from Nvidia (at high interest rates),
Is this just something you are making up?
"NVIDIA intends to invest up to $100 billion in OpenAI as the new NVIDIA systems are deployed. The first phase is targeted to come online in the second half of 2026 using the NVIDIA Vera Rubin platform."[2]
The closes there is to waht you are saying is reporting that NVIDIA has discussed guaranteeing some of the loans OpenAI is taking to build data centers:
"Nvidia is discussing guaranteeing some of the loans that OpenAI is planning to take out in order to build its own data centers, The Wall Street Journal reported, citing people familiar with the matter."[3]
This of course is the opposite of NVIDIA loaning OpenAI money - if they did this they would be liable for OpenAI's debts.
[1] https://nvidianews.nvidia.com/news/nvidia-announces-financia...
[2] https://nvidianews.nvidia.com/news/openai-and-nvidia-announc...
[3] https://archive.is/Gpvq2#selection-1299.0-1301.181
Just because they have cash doesn't mean they're going to lay it all on the table for one risky business. Which is why the actual solution is mostly loans.
In being a guarantor, Nvidia isn't directly giving the money now, but it is promising to be liable for the loan if OpenAI defaults. That's a risk they have to budget for.
But in addition to being a guarantor, they're also loaning out chips:
"OpenAI plans to pay for Nvidia’s graphics processing units (GPUs) through lease arrangements, rather than upfront purchases." "By leasing the processors, OpenAI can spread its costs out over the useful life of the GPUs, which could be up to five years a person said, leaving Nvidia to bear more of the risk."
Nvidia is leasing them the GPUs, which is basically a loan.
"In addition to offering a cost-efficient way for OpenAI to access chips, Nvidia’s lease option and long-term commitment can help the company land better terms from banks when it comes to raising debt, a person said. "
"As a non-investment-grade startup that lacks positive cash flow, financing remains costly. OpenAI executives have called equity the most expensive way to fund data centers, and said that the company is preparing to take on debt to cover the remainder of the expansion."
"“Folks like Oracle are putting their balance sheets to work to create these incredible data centers you see behind us,” Friar said. “In Nvidia’s case, they’re putting together some equity to get it jumpstarted, but importantly, they will get paid for all those chips as those chips get deployed.”"
Nvidia leases (loans) them GPUs with a promise to pay for them later. Getting this lease (loan) from Nvidia helps them secure more debt.
"OpenAI’s path to a $500 billion private market valuation has been enabled by hefty investments from Microsoft and others that allow the company to burn billions of dollars in cash while building its AI models that power services including ChatGPT."
"Jamie Zakalik, an analyst at Neuberger Berman, said the Nvidia deal is the latest example of OpenAI raising money that it pours right back into the company providing the capital. Investors are concerned about the “circular nature of this deal goosing up everyone’s earnings and everyone’s numbers,” said Zakalik. “But it’s not actually creating anything.”"
They get a loan of chips, use it to get a loan of money from someone else, and spend that money on the company that loaned them the chips. It's a Ponzi scheme.
(https://www.cnbc.com/2025/09/24/nvidia-openai-investment-in-...)
This is just a grab bag of different critisms to what you said.
Vendor financing is very standard across multiple industries. It's not a loan in the sense you originally stated because it's physical assets, not financing.
The circular nature of the deals is a valid concern but different to what you raised.
It's also a much smaller concern now with Anthropic hitting $30B ARR from non circular sources than it was in September 2025 when your article was written.
It's a valuation, not the amount they raised.
I am playing around with this at home right now. I think a lot of the latest improvements came with the harness, instead of AI.
The part I am working on is to have better tools and data to search over. Curated for my needs. Similar to the Karpathy post yesterday about his wiki. I am trying something similar and even qwen 3.5 is totally fine for most of what I do.
Disclaimer: I bought memory before the crisis started. Not sure if I would build my PC as is now..
These dumb models are just making people hooked on the thing. What you think they building all those datacenters, hm? When the real deal comes, all the zombies would want their even better cocaine intravenously at whatever exorbitant price.
Vibe coding requires the sota models to work at all, but someone who knows what they are doing and uses the AI more responsibly can absolutely use the cheaper Chinese models for coding, and they’re often faster too. If I was one of the big players my entire focus would be on lobbying for regulation and outright banning of local models.
Yeah, Qwen3 coder for Claude Code and 3.5 for OpenClaw replaced my full-stack use of Opus 4.6 already; it's fine for basic web apps, k8s/docker infra setup, optimizing AI models etc. with only slightly higher error rate than Opus. Upcoming 3.6 together with Gemma4 might make it even better (still to test). OpenAI's memory spot market play might have been directed at local inference as well.
Look for Deepseek 4 when it drops, I’m curious how good it will be.
The thing is, if you’re using AI responsibly today you’re already breaking down tasks to such a granular level that you don’t need the power of Opus. You can save that for deeper research tasks.
Based on the current DeepSeek website I suspect it's not going to be great as their current model (V3.4? V4-mini?) often forgets or changes facts explicitly mentioned in the conversation which R1 never did. It's better than R1 at math or coding, but nearly unusable for deep conversation. I suspect they pushed MLA or linear attention too much, or quantize a lot more than before.
> Vibe coding requires the sota models to work at all
I've been hearing that vibe coding is the future for at least the past year but somehow it always requires the newest model, and all the "old" models produce trash results. Why is that?
Because expectations on what could feasibly be vibe coded have changed. Before with older, worse models it was only toy projects that were expected but now with better models it's much larger enterprise apps.
Doesn’t matter if the current model creates bugs they can’t fix - in 6 months the next model will be such a huge improvement that THEY will fix it.
What about the bugs that model makes? Well the next model will fix them of course!
Just keep giving us money and we’ll fix your bugs for you, no worries!
Both of these valuations are absolutely absurd. I guess Anthropic looks good in comparison, but I don't want to hold that bag.
The Chinese models are catching up in quality while being a fraction of the price. The market will speak, how many devices that contributed to this thread were made in the USA?
Sure you can argue the Chinese companies are heavily subsidized, but no major LLM lab is remotely close to making a profit this decade.
I switched to Anthropic briefly to try out Code, it was great (having never tried such a product) and better than their chat bot which found prettier and nicer to talk to but far more often wrong.
Then shortly after Codex was released which made that more accessible and instantly preferred that compared to Code since much more generous allotments for $20 plan. Claude constantly kept hitting. And Code actually had a more robust UI and was more accurate when doing the same project side by side vs Claude Code.
But imagine many haven't tried Codex recently since all we hear about is Claude Code. So while they may have momentum, at least for me with no stake in the game, Codex finding far better, but I suppose that could all change again on a whim.
> much more generous allotments for $20 plan
> imagine many haven't tried Codex recently since all we hear about is Claude Code
My theory is Codex allotments are higher because the demand isn't there (yet). Claude users hit their limits because there is so much more usage.
My loose understanding as someone adjacent to the AI model space is that you have good models that are costly and cheap models that are decent, so a lot of the publicly visible fights where Claude and ChatGPT leapfrog each other is the companies doing cost-benefit of how much optimization to do on the models before your userbase revolts because the agent "used to be great and now kinda sucks".
As a small business owner whose team is entirely in Google Workspace (Drive, Gmail, Chat -- so inbuilt RAG right there), I wonder if Gemini will be the darkhorse. As a user Gemini's a distinct third in "AI smarts", but most business owners aren't power users who are gonna setup Codex or Code to slurp up their work emails and internal docs/SOPs.
The article feels a touch clickbait-y since people love a good fight between the top players and OAI's lost a buncha public goodwill over the past year.
As a government based person who has witnessed multiple states (not in the US) move all operations off GCP because Google doesnt address sovereign risk, local data and hosting privacy requirements,and contracts well at all, where Microsoft and AWS do, I doubt Gemini will have as large a dark horse moment as it could. Copilot Enterprise can span across similar domains, and whilst it is very expensive per user it has the benefit of having existing contracts in place which it can bolt in to.
Ooh good point.
Out of curiosity, have you compared the relative effectiveness of ChatGPT and Claude vs Copilot? Given your existing enterprise contract, does copilot have a monopoly on your AI usage due to its superior compliance?
In the handful of contexts Ive seen across departments and states, no, ive not seen such a comparison. Ive seen state custom implementations of GPT (eg 4o), and Copilot, but not Claude. Claude is likely out there for the same reason Google is - navigating those same localisation / legal issues is complex for government departments.
Copilot has a far easier path to monopoly because it can be pushed down existing Microsoft contracts, MS has put in the work for international jurisdictions in terms of data compliance in existing products and thus can extend upon that with Copilot, and they have a foothold in a good many places. One thing that is interesting with Copilot is that if you set it to Work mode, it can treat your organisational SharePoint sites and OneDrive locations as sources, but do so in a manner that respects what you as a user has access to.
The main point here
> large gap between OpenAI’s $852-billion valuation and Anthropic’s $380 billion
IIRC Anthropic's revenue is either roughly at parity with or larger than OpenAI's, and Anthropic is growing faster[1]. All indicators are that Anthropic should be worth more than OpenAI. Given that, one could reasonably expect the relative valuations to change a great deal. In any case, it's not clear why OpenAI would command such a price premium over Anthropic.
1 - OpenAI says they are doing $2B/mo run rate https://openai.com/index/accelerating-the-next-phase-ai/
Anthropic's run rate increased from $12B to $19B in the period between February 12 and the end of the month. If the implied growth rates held through March, Anthropic may well be larger than OpenAI now.
Both companies have the same issue: the unit economics are bad, and there’s no moat.
I’ve been primarily on Claude for the last 6 months or so, but have been hitting rate limits. I switched work to Codex seamlessly, just like I could switch to any other provider seamlessly
The unit economics at this point are about utilization. Their cost is well below what they’re charging, but only when there is enough traffic to keep the GPUs busy. So the game is about increasing demand to level the load.
GPT has negligible moat because they gave up on all their integrations. Claude code is starting to develop one as people start to build things that require Claude Code specifically, not just any LLM.
> Claude code is starting to develop one as people start to build things that require Claude Code specifically, not just any LLM.
I hate to be a "source?" guy, but I'm curious if you have any examples of this. Skills and MCP are really the only extensions on CC itself I'm aware of, and these are both supported in Codex.
Things like Dispatch / remote sessions is something CC has that Codex does not, but these features are quite easy to replicate (and I expect Codex to do so in short order).
I agree that’s a great question which I don’t really know the answer to. These tools have been moving in lock step for some time now. One will innovate and within 2-3 weeks the others have that feature. Where I sit the mind share all seems to be going to Claude though. The moat develops when people build something that only works on one - even if the others have the same features it doesn’t matter unless they’re literally binary compatible. Skills are just prompts at the end of the day with nothing more specialized than a file naming convention.
Having written several orchestrators I’ll say that the code to invoke the tool is pretty equivalent but it’s the details that matter. Exact CLI flags and json fields.
let me know when they scrap the data centers, id love to get some good deals on hvac equiptment. these companies cannot possibly make enough money when you can run something on your own computer that works mostly as good
IMO This says a lot more about the investor herd mentality and FOMO than it does about the technological or any other kind of merit of Anthropic vs OpenAI.
Google “does Anthropic make a profit” and you will get endless responses about how they’re “earning revenue”. It’s a pretty obvious tactic.
What’s interesting is that they’re both still losing money on their models and are essentially giving away compute for free, although lossy.
They’ve bought up computing power and are now renting it back to the rest of us, with a decent HMI, while subsidising us to use it.
https://archive.ph/ExQkj
I am not a finance pro, but is it not normal and expected for secondary equity markets to be hesitant against large unloads of pre-money company stocks, even in cases where the company in question isn't in water like OpenAI is? Does this not say more about the failure of the investors themselves in balancing their portfolio properly than OpenAI itself?
The latest Opus routinely tells me the latest GPT Pro responses are much better. The GPT responses cost 10x more than GPT at least. And GPT takes 10’s of minutes. So unless and until I’m needing and ready for a really expensive “math checker” it gets left alone.
People are so late to Anthropic, I used Claude 1.3 back when it released, and I've stood by Claude since the early days. I don't think people quite realized the potential of Anthropic
Mind sharing with us? What is the potential being missed?
And x.ai is at $2T now, while qwen baba is under $200B…
Isn't XAi's valuation paired with X and SpaceX?
Silicon Valley valuation numbers are entirely made up to boost their IPO. Actual value is only determined after these companies go public.
Allbirds, the shoe brand, went public at $4bn in 2021 and sold its assets last week for $39 million.
> Neil Saunders, the managing director of the consultancy and market analysis firm GlobalData said: “Allbirds has gone from being a high flyer to a dead parrot.” He said its early success “was driven by Silicon Valley hype, more than deep popularity with consumers in the American hinterland”.
[0] https://www.theguardian.com/business/2026/apr/02/eco-shoe-br...
Idiot investors who will lose their money when Google wins. Or even if it doesn't, for that matter.
Odd timing...Everything I've read about Claude the last several days suggests that its users are disappointed, even furious at what's happened to its performance.
I probably use a couple thousand session hours monthly between work & personal; very pleased with the results I get.
But I’m not necessarily oozing about it online — vocal minority and all.
Self-edit: a couple thousand session hours is probably hyperbolic; it's likely closer to a thousand after discussing usage with another user in a separate thread.
Maybe a better metric: Across work + personal I have 1,000+ sessions in the last 30 days, longest session was 3d 20h
Now that they’ve relegated OpenClaw to pay-as-you-go API usage, I hope the compute overload issues they’ve been facing will ease up some.
It's been working great for me. I haven't seen any changes.
Same, mostly. However today in particular Claude can't do front-end development with any competence. Never seen this before. I think the rumor is they are rolling out a new model and have to divide their infra across the new model vs the current model.
OpenAI is being squeezed from both sides.
ChatGPT's chat quality has recently dropped hard. While Claude is pricier, it actually takes the effort to think through complex tasks.
All the while, Chinese models are providing cheaper alternatives.
[dead]
tl;dr nobody wants OpenAI shares and the secondary market has completely dried up.
“We literally couldn’t find anyone in our pool of hundreds of institutional investors to take these shares“
This doesn’t bode well for an IPO. The market is smelling a stinker.
Get your popcorn ready for a mad scramble to salvage investments if indeed the shark has been jumped.
not to mention the drama that's followed the company from the very beginning. It think its getting to a point where the character issues can no longer be ignored bc it's directly affecting business
I think the IPO and subsequent quarterly earnings where they will be pressured by analysts will pop it all.
I was watching a recent Jensen Huang Q&A with analysts and it was essentially “just trust me bro”. They’re all interconnected - once there’s a correction for one player, all get affected.
Can’t wait to finally get this over with so we can finally move on.
The gap between hype and reality needs to be corrected.
Nah you see, the NASDAQ is giving them a out just like SpaceX. Short-circuit the index joining time and allow them to get into index funds within days. After that you shift all the losses onto 401ks as the original investors cash out and the 401ks blindly buy the shares as part of the index.
I wonder how much of this is associated with Scam Altman's personal negative PR and Anthropic's recent PR wins.
I'm inclined to think there isn't much of an association becauss investors don't seem very concerned with morality, but I know ~dozen developers that either switched to, or started using Claude in the past month or so, while not knowing anyone that uses Codex.
The issue is investors want discipline.
They want to see the CEO communicate a path to profitability. Anthropic has - purely by focusing.
OAI in contrast is all over the place and they haven’t shown they’ve learned.
Zuckerberg got punished for his metaverse nonsense and investors were correct to be skeptical and reflect that in the stock price. Altman thinks he’s a god and the rules don’t apply to him. More fool him.
> Zuckerberg got punished
I think it's hard to grasp just how much money he poured into this folly. $80 Billion. For an idea that any random person on the street could have torn apart. He read Ready Player 1 and decided that was what the future should be. Then he tried to make it. He ruined the Oculus brand and stole all the oxygen from other VR development just to make a lesser copy of our own world, but one where he could be admin.
Seattle just opened a new transit line, spanning from downtown Seattle to Bellevue. It's 33 miles long and cost $3.7 billion. It was already carrying 2 million people per year [0] even before the entire line was finished and on the opening day after completion it reached 200,000 riders.
Further from home, the entirety of the Artemis II program cost $93 billion [1] and lists an estimated $4 billion per launch.
One last example - the panama canal would cost an estimated $50 - $75 billion if you built it today. [2] The panama canal reshaped trade routes and the global economy.
Where is Meta's moon landing? Where is their impact? Zuckerberg spent enough money to build a new transit system for an entire city. Enough money to land astronauts on the moon. Enough to connect oceans together. He poured this money down a hole chasing his own vanity and a vision of the future that no one wanted.
How, after all that, did he get punished? He's still CEO. "Meta" is still overvalued. They're still regarded as a serious tech company.
[0] https://en.wikipedia.org/wiki/2_Line_(Sound_Transit)
[1] https://www.msn.com/en-in/news/world/artemis-2-cost-explaine...
[2] https://www.mikegravel.org/how-much-would-the-panama-canal-c...
[flagged]
> Zuckerberg got punished for his metaverse nonsense and investors were correct to be skeptical and reflect that in the stock price.
Did they? It's been a little less than 5 years since Zuckerburg announced the all-in plan on the metaverse. In that 5 year period, Meta's stock price has gone up by 84%... which is middle of the pack for MAGMA, less than Apple (91%) and Google (157%), but above Amazon (24%) and Microsoft (47%). It's also comfortably above the S&P 500 (59%).
[dead]
> Altman thinks he’s a god and the rules don’t apply to him.
The Sam curse. Took down Bankman-Fried before him...
How did Zuckerberg get punished? The stock is way up since he started on it.
[dead]