189 pointsby w10-16 days ago14 comments
  • rendang2 days ago
    Typically tech companies that end up with fat profits are near-monopolies with dominance in market share or with moats on their expense side, like Apple's vertical integration or Amazon's logistic network.

    What's the scenario look like where multiple LLM providers like Anthropic and OpenAI both live up to their valuations? Claude being 1.1x better at code and OpenAI 1.1x better at some other stuff doesn't seem like enough differentiation to do it.

    • swatcoder2 days ago
      There doesn't need to be any such scenario, and there almost certainly isn't one.

      NewHype valuations reflect how confident current investors are that later investors are still coming and have nothing to do with specific profit models. Maybe the hype will pay off or maybe it won't, but during this phase all I need to believe is that I can get out before the tide turns.

      At this point, the carrot that keeps investors coming is that many many people still believe at least one of these companies will become the next IBM/Microsoft/Apple/Google in one way or another and we don't know which one yet.

      It's just buying tradable bets on a horse race that won't see its end for several more years. One horse could win, all the horses could drop dead, and none of it matters today: the game is in deftly trading those betting tickets as the odds shift.

      • pj_mukh2 days ago
        OR, the pie is big enough that a monopoly isn't required. The effect on the economy is so large that all of these companies become money printing machines.

        I personally don't think this is possible unless these AI Companies translate from just bits to atoms but that's not out of the question[1].

        [1]: https://venturebeat.com/ai/openai-has-begun-building-out-its...

      • throwup2382 days ago
        This is the best take here. We’re in gold rush mode, everybody thinks there’s gold there and that the potential reward is worth the risk, even knowing that there’s a significant chance that their claim won’t find enough gold to make the investment worth it.
      • TechDebtDevin2 days ago
        With 8 billion being their largest investor being Amazon, who has proclaimed they're going to spend 100bb on similar investments, these valuations and investment sizes, aren't even large relative to the investor's bankroll.
      • lawn2 days ago
        Also called "selling to the greater fool", which has fueled bubbles for as long as we've recorded bubbles.
    • blindriver2 days ago
      It's not functionality. It's access to content.

      The real target is replacing search with AI, as well as locking up all content creators into exclusive deals so that other LLMs don't get access to the new data.

      Just like how streaming has locked video content under different vendors, LLMs will split information across multiple vendors and we will need to get subscriptions to different companies in order to search things in the future. And it's going to cost a lot of money for us peons to get access to data in the future.

      • nradov2 days ago
        Right. OpenAI exhausted all public training content and now they're hiring expert human authors to write proprietary training content in targeted areas. Frontier model competitors are probably doing the same. Over time we'll probably see quality differences between those models in particular domains depending on how much funding they put into generating novel training data.

        To draw an analogy it's sort of like the stock market. You'll never achieve sustained above-market returns by trading based on publicly available data because everyone else has access to all the same data. In order to get an edge over other traders you have to do proprietary research (or be faster or cheat).

      • fourside2 days ago
        > locking up all content creators into exclusive deals so that other LLMs don't get access to the new data

        It seems like most (all?) foundational models were trained on stolen content. I’m not sure what content exclusivity looks like for LLMs. It’s not like Joe Rogan signing on with Spotify where they have exclusive distribution rights for a particular piece of content.

        • blindriver2 days ago
          This is exactly what it's going to be. Free information is dead, we just don't know it yet. It's dead because why would someone produce free content, and then have all their page clicks hijacked by LLMs that just return the data directly? I think ad/search ecosystem is dead in less than 5 years. If it's not dead then OpenAI fucked up.
        • cruano2 days ago
          There was already a $60M/year deal between Google and Reddit
          • blindriver2 days ago
            Exactly. And Reddit is doubling down on monetizing content it doesn't create itself. Not only are the mods completely free and unpaid for who curate and hone the best posts for their subreddits, now the subredditors are having their comments monetized by Reddit for free. And the upvoting ensures that high quality data is already reinforcement learning via meatbots.
          • fourside2 days ago
            How do you enforce it? Even if you try doing it through copyright laws a company like DeepSeek is not going to care.
      • mi_lk2 days ago
        And what do they have to compete with Google on content access?
        • better knowledge reach, the biggest search intent goal
      • 2 days ago
        undefined
      • xmprt2 days ago
        > locking up all content creators into exclusive deals

        This is the worrying part. The real content creators are people like you or me (not reddit or stackoverflow) and we aren't getting a piece of the exclusive deals. So if this is the end game, then it means either exploiting real people who are writing things or that people stop using it and the internet uses a lot of its utility as a place for communication and shared knowledge; probably a bit of both

      • bparsons2 days ago
        Since web 1.0 is our best analogue, I am thinking of it like this:

        Late 1990s internet infrastructure buildout means that most of the captured value was at the hardware level (HP Computers, Dell, Intel, Cisco etc.).

        2020s AI infrastructure buildout appears to be the same. NVIDA is capturing a lot of value, and experiencing explosive growth.

        Late 1990s/early 2000s was a time of tremendous development for web software , but no one really had a business model. These LLMs are like the web browsers of that era -- they are an interface to access other peoples IP, but no one really knows what the financial model looks like.

        It wasn't until the mid 2000s/early 2010s that the business models came into sharp focus for web services.

        Like web browsers, the LLMs might be a vital tool for enabling the capture of value for businesses, but it might not be the actual thing that they end up monetizing.

    • dmix2 days ago
      Oligopolies often exist in the same general categories by serving different submarkets (regions, business vs consumer, large vs small business, specialization, etc). Especially when it's the category is just a technology, like databases with Oracle vs the many db vendors/services. If LLM's technically plateau then it's all about who can find those high $$ markets (coding, consumer chat, phone integrations).
    • superfrank2 days ago
      I've been thinking about this for a little and it really feels we're on the path to LLMs becoming a commodity, similar to cloud computing.

      At this point, I would bet that for the average user the middle tier models for all the major players are more for their uses and any differences are indistinguishable. Even when major advancements are made by one party, it seems like everyone else can catch up pretty quick. I'd expect that to continue to happen, especially if companies start embracing open source as the future (like they're claiming they will).

      If that happens, the winner in this space won't be who has the best model, but who can build the better company. It's going to be able the deals they make, they products they build on top of their models, and who can bring their costs down the most.

    • TZubiri2 days ago
      >"live up to their valuations"

      IDK, I won't pretend to know the difference between $61.5B and $123B, we can safely assume both will have a similar value, OpenAI as first mover, and Anthropic like a Pepsi.

      >What's the scenario where both Anthropic and OpenAI both live

      Pretty much all of them, their risk of failure is pretty much independent. You said software usually has near monopolies, which I agree, there's usually 3 or 4 or 10 big entities. To my estimation because of the need for competition and market segmentation. So the survival of one only conditions the other as a dependent event in the order of 1/10.

    • ibejoeb2 days ago
      Anthropic is ahead of the game on the applied technology side. Claude code is pretty good. It feels like a product from the early PC days. There's so much further to go, and so many other verticals to get into.
    • dcdc1232 days ago
      I use agentic tooling when programming all the time and I can say for certain that Claude feels _several times better_ than any OpenAI model, despite what the benchmarks show.
      • dimitri-vs2 days ago
        Hot take: sonnet3.7 is as good if not better than OpenAIs $200/mon o1-pro model, were it were not for Deep Research I would have cancelled my ChatGPT subscription.
    • smugmaa day ago
      Counter-point: for “cloud”, there’s an oligopoly of AWS, GCP, and Azure that maybe make up 80%.

      And still there is plenty of space for those closely related, such as Cloudflare.

    • FloorEgg2 days ago
      The answer to this question that I almost never see come up in these threads has nothing to do with moat or typical startup dynamics, it has to do with macro policy and theoretical economics.

      Our economic system is set up to continuously increase the supply of money at a rate that results in the average price of things consumers buy increasing 2% a year.

      Technology is a deflationary force and if our system wasn't constantly adding more money the price of almost all consumption goods would continuously fall.

      These investors are banking on the idea that these companies will automate so much high value labour that there is a massive wave of deflationary pressure on the economy, resulting in massive increases to the supply of money to balance it out.

      So by crude example; 10 years from now anthropic could end up with 1% market share and be worth $10 trillion.

    • outside12342 days ago
      Also feels like we are just waiting for someone to swoop in with a open source model like DeepSeek that wipes both of them out.
      • whiplash4512 days ago
        The product is much more than the model.

        Our hacker community is not a good proxy for the overall userbase.

        • rowanseymour2 days ago
          Ok but where is the product? Aren't most people just using ChatGPT from a chat prompt like they can do with any other LLM?
          • 2 days ago
            undefined
      • AustinDev2 days ago
        QwQ 32B has replaced Sonnet 3.7 for me for quick hack apps I write like. 'Service that converts youtube tutorial videos into step by step instructions with timestamp references for each step.'

        And I can run it locally!

        • johnmaguire2 days ago
          For whatever reason, qwq on a Macbook under ollama & Open Web UI seems to often "stall out" for me. It just never reports it even started to answer. Have you experienced this at all?

          I don't have this issue with qwen2.5-coder, deepseek, etc.

        • dchuk2 days ago
          Are you chatting with it directly or using it with a coding tool or IDE?
      • ch33zer2 days ago
        Shouldn't deepseek have done that already?
        • keithwhor2 days ago
          DeepSeek V3 is easily the best cost / performance non-reasoning model on the market but accessing it via API is not straightforward -- at least for people who want their model hosted in the US. For the layperson OpenAI / Anthropic are much more accessible.

          Some of OpenRouter's [0][1] providers seem bunk; their version is corrupted. Occasionally just get mangled outputs. Fireworks [2] is my best experience so far: their serverless API, haven't messed with custom deployments. Fast, performant, better than other models in its size category and less expensive.

          [0] https://openrouter.ai/deepseek/deepseek-chat:free

          [1] https://openrouter.ai/deepseek/deepseek-chat

          [2] https://fireworks.ai/models/fireworks/deepseek-v3

        • rtkwe2 days ago
          There's a lot of money riding on their being a lot of headroom left on LLM performance and usefulness. I personally have my doubts you can get their with the Chinese Room model of LLMs but that's where all the money and hype is riding.
    • alex_abt2 days ago
      The profit margin will come from government contracts.

      It's the Azure/AWS business model.

    • xyzzy95632 days ago
      I think the way it works is that if the multiple providers get swamped by enough demand, it still drives up their profits and margins despite the existence of competitors. Similar things happen with cloud providers I think.
    • MangoCoffee2 days ago
      according to fans of Deepseek and some people on HN, isn't LLM going to become a commodity?
    • nerdponx2 days ago
      > What's the scenario look like where multiple LLM providers like Anthropic and OpenAI both live up to their valuations?

      Protectionism by the US federal government.

      • mkolodny2 days ago
        The same government where the head of the department of efficiency is also the owner of Anthropic’s and OpenAI’s competitor, xAI?
  • w10-16 days ago
  • benrutter2 days ago
    Some of these mega-evaluations make me nervous. Maybe I'm overly bearish but I don't think Anthropic is that likely to generate enough revenue for this, particularly in a world with so many, often free, competitors.

    Assuming I'm right, I'm not really looking forward to a point in 5 years when shareholders all suddenly decide they agree.

    • bhelkey2 days ago
      > Assuming I'm right, I'm not really looking forward to a point in 5 years when shareholders all suddenly decide they agree.

      Assuming everyone agrees Anthropic is worth very little, they still get to keep the 3.5 Billion investment.

      I imagine the consequence would be:

      1) Anthropic investors lose a lot of money.

      2) Anthropic probably struggles to raise more money

      3) Anthropic raises their prices to become profitable.

    • nradov2 days ago
      Why would you be nervous? If valuations collapse then what practical impact would that have? Are you a shareholder?
      • ncallaway2 days ago
        When the dotcom bubble burst, it affected far more people than just the shareholders and employees of those specific companies.

        The economy is interlinked, and a bubble bursting hurts many people.

        Part of it is realizing the pains of inefficient allocation of capital. A bubble "bursting" is often an indicator that the capital in within that bubble had been previously misallocated. So the broader pain when the bubble bursts, is partly society and the markets suddenly discovering that the $3.5B probably could've been put to better, more productive uses.

      • paxys2 days ago
        At this point the AI bubble is very tightly linked with the tech industry and the overall US economy. If/when it bursts we are all going to be affected, whether we are directly invested into it or not.
      • benrutter2 days ago
        A few other comments have already said this, but I was thinking that a "bubble burst" or "market correction" would really not be a good thing for the IT sector. It'd directly cause a bunch of unemployment, and additional spook investment out of other areas.
        • nradov2 days ago
          The investments just move on to the next hot area. It would be a temporary disruption at most.
    • throwaway7439502 days ago
      > when shareholders all suddenly decide they agree

      Sorry, what are you saying happens at that point?

      • benrutter2 days ago
        I'm thinking a market correction, like when the dotcom bubble burst and a lot of value disapeared from the stock market overnight. This is really not a good thing, and almost definitely would affect the IT industry as a whole.
  • bhouston2 days ago
    Makes a lot of sense. In writing my open-source agentic coder (https://mycoder.ai), I find Claude is head and shoulders above the rest. I wish I could invest.
    • pllbnk19 hours ago
      There's no reason not to think that within their current model architecture they accidentally discovered lucky configuration which helps them achieve better results than competitors. While companies still need tens of thousands of GPUs to train the models, we no longer need that for inference; with DeepSeek you can infer efficiently at a fairly modest price with comparable quality. As the model training is understood better, we might hit the ceiling where additional thousands of GPU-hours won't give any noticeable quality benefit and good results from training can be achieved even without investing hundreds of millions into hardware.
    • justinc86872 days ago
      FYI, the dark blue text on a black background for your examples is almost unreadable without highlighting the text.
      • bhouston2 days ago
        (More examples are here and I think the styling bug isn't affecting this page: https://docs.mycoder.ai/docs/examples/ I will look into the bug you ran into.)
        • nijaru2 days ago
          They're likely running the Dark Reader extension and turned it off for the site, but didn't refresh the page. The extension doesn't always fully reset when you disable it for a specific site. Reloading the page with the extension turned off fixes the issue, so it is not your site.
          • bhouston2 days ago
            Thanks for the explanation
    • rezistik2 days ago
      Fundrise innovation fund is an investor and so you could through that fund.
      • 2 days ago
        undefined
    • tsycho2 days ago
      Anthropic has been on-and-off available on EquityZen. I personally find the valuation expensive, but you do have the opportunity to invest if you want.
      • JumpCrisscross2 days ago
        > Anthropic has been on-and-off available on EquityZen

        Please note that EquityZen is notorious for being a premium payer in the market, to the tune of 20 to 500%.

        • pergadad2 days ago
          May I ask what you mean by this? They pay 20-500% more for the same #/% of ownership as other investors buying at the same time? Or do you mean they charge their customers such a high premium? Is there any data on this available?
    • tristor2 days ago
      Hiive has listings where folks are selling their employee shares as secondaries, if you really want to get exposure that much.
    • henning2 days ago
      The page does not scroll smoothly. https://simonhearne.com/2015/jank-meter/
    • 2 days ago
      undefined
  • strangescript2 days ago
    I have been using Claude Code, and its a bit pricey, but has exceeded my expectations so far.
    • jstummbillig2 days ago
      Do tell. How is it better than Cursor, Windsurf or (probably closest) Aider? Does it do anything the former don't, that was not possible before? I did not really understand the value prop, but am super interested to learn more.
      • rolisz2 days ago
        I don't like Cursor, because I like PyCharm as my very awesome IDE. Claude Code is a terminal app, it's something that I just issue commands to and it mostly works, and then I check everything in PyCharm.

        I tried Aider a couple of months ago and it gave me crappy code. Maybe it was the model I used back then, but I didn't like it. I had much better experience with Claude Code.

        • MyOutfitIsVague2 days ago
          There are valid criticisms of Aider (I feel like it might use more tokens than it needs to, but I haven't done a lot of comparisons), but output quality is certainly the model.
        • bredren2 days ago
          Are you running it from the built-in terminal in PyCharm? I also use that IDE and want to give this a swing.

          Can you elaborate on "check everything in pycharm?"

      • paxys2 days ago
        Cursor's in-house models are vastly inferior to GPT or Claude. I bet the majority of Cursor users are still using one of these two under the hood.
        • Kiro2 days ago
          Yes, but the question was about Claude Code and not the underlying model.
      • voidsnax2 days ago
        Before Claude Code, out of the alternatives you mentioned, I'd only tried Windsurf. I find Claude Code to be roughly comparable to Windsurf in terms of capabilities, but I can use it from my terminal. I'm much more at home in a terminal than I am in VSCode (which Windsurf is a fork of).
        • dhc022 days ago
          Oh wow you owe it to yourself to try aider then. Architect mode especially, where you use one (usually reasoning) model to plan and one (usually Claude) to write code, is pretty awesome.
      • ranscendental502 days ago
        Nope -- he's giving us his personal feedback and experience. I agree with him. I dislike Cursor and Windsurf and enjoy Claude Code alot.

        Bother to try it for yourself, and make your own decision. I know it make actually be a bit of work...but you'll probably find everyone has their own style from the web API, to some IDE. No need to be so hostile.

        • nickthegreek2 days ago
          I do not read hostility there, just directness. He has a valid inquiry and states he is interested to learn more. Assume good faith.
          • j_bum2 days ago
            Agree. Not sure how, “I’m super interested to learn more,” indicates hostility.
            • sundarurfriend2 days ago
              I think ranscendental50's reply itself was unnecessarily hostile and don't agree with the way they responded, but I can see why they assumed hostility. "Do tell" is unfortunately to a large extent used passive aggressively on the Internet, and generally actually means "Don't bother telling me, you're obviously ignorant".

              This seems to be a case where it's meant genuinely, but it's not surprising that not everyone picked up on that.

    • BeetleB2 days ago
      Does Claude Code charge based on API usage?

      And does it let me limit the files it's sending?

      • hectormalot2 days ago
        Yes, by API usage. I think you can’t limit which files it sends.
        • danso2 days ago
          They’ve made a recent change to Claude Code that apparently makes it respect your gitignore

          https://github.com/anthropics/claude-code/issues/79

        • BeetleB2 days ago
          In that sense, Aider is definitely better.

          A coworker was trying both out on a project, and was finding Claude Code to be 4x more expensive (both using the Sonnet-3.7 mode) - because he couldn't limit the context Code would send. As the project grows, that multiplication factor could be a lot larger.

    • 2 days ago
      undefined
    • xyzzy95632 days ago
      Claude Code is amazing. Typically it has costed me about $25/day of using it but it's well worth it. I ran into the $100/month limit and now waiting for my account upgrade so I can use it some more.
      • kashnote2 days ago
        Wow, how are you using so much? Are you basically prompting it to build out a whole application?
        • xyzzy956320 hours ago
          I built a large front-end component with it.
  • buyucu2 days ago
    How does one justify this valuation in a world DeepSeek and Qwen are free?
    • elicksaur2 days ago
      How does Microsoft justify a $3T valuation in a world where Linux is free?
      • stackskipton2 days ago
        Microsoft valuation is high but they have clear profitable products.

        Most people going, “these valuation are insane” because these AI companies are not profitable and doesn’t seem to be a clear path to profitability.

        • vdfs2 days ago
          I think they need a lot of money to build AGI, then ask it how to be profitable.

          That's what OpenAI head said in an intreview

      • caseyy2 days ago
        It doesn’t. It’s just speculative markets.

        To recreate all Microsoft’s meaningful profit-generating offerings would cost a very small fraction of $3T. In SaaS space startups are competing with MS products for much cheaper. In operating systems, commercial Linux (RHEL, SteamOS, etc) is competing for much cheaper. In gaming, Microsoft has contributed much to game cost inflation, meaning others compete for cheaper. In hardware others compete for cheaper. In home office, even Google and Apple likely spend less to compete. Maybe in web services Azure leads in costs and has some moat, but I’m only saying this because I don’t know it well.

        This is not to say MS is not an impressive company, but $3T is not its real value. From the stock market side, PE of 32-ish suggests it’s meaningfully over-valued too.

        So I know you kind of meant that question in a half joking way, but it is a good question. You can’t explain $3T with real value, it has to be speculative or bubble-ish.

        • ghc2 days ago
          Those are just products. Microsoft has distribution at a level that is orders of magnitude harder to build and worth far more.

          Comparing how much money it would take to build a competitive portfolio of products to compete with Microsoft is like comparing how long it would take to build a Twitter or Facebook clone. Sure, it wouldn't take that long or that much money, but product is the easy part.

          • caseyya day ago
            Companies ultimately participate in the economy with their products and services. Infrastructure is an asset used to provide these services and goods, which can (through economies of scale and similar effects) make the company more competitive. However, Microsoft's services and products wouldn't be competitive in the natural, free market as exceedingly similar products can be provided cheaper. This is just general microeconomics.

            We can also ask whether Microsoft's market is free, from a macroeconomic perspective. I would say it is not. There is still a very significant consumer lock-in from Microsoft's dominant position. Some would even argue that this lock-in is monopolistic as Microsoft created many markets it inhabits (operating systems, home publishing, small business software, PC gaming, and similar). I don't mean to criticize here - monopolistic is a very loaded word politically. Objectively, It's natural that businesses are monopolistic for some time when they create markets. But not forever.

            My point is that Microsoft's monopolistic position is not sustainable unless it provides services that are either significantly differentiated or cheaper than the competition. As such, it should not command its market cap. Microsoft understands this, as its strategy is to establish itself as a monopoly in emerging markets (like AI) by using "embrace and extend," a business tactic they are known for. But I don't believe monopolies in moat-less markets are easy to sustain. This, once again, brings the market cap into question.

            Perhaps I am missing something; this is just my opinion, anyway. But I am trying to look at it objectively from a corporate finance perspective. It's easy to get lost in marketing speak. Microsoft does a lot of "upward management" aimed at its public shareholders, so it's important to look at the fundamental facts. Microsoft is an impressive company, and it has many things to brag about, but some matter more in its actual business economics, and some matter less.

            I don't believe infrastructure matters as much as the monopolistic lock-in they have (for now) and the competitive edge they don't. Perhaps they will find a more sustainable business model as it's clear from how their products and offerings have been developing that they are urgently trying to.

            • ghc14 hours ago
              > Companies ultimately participate in the economy with their products and services

              This is the important point. A SaaS company with a competitive offering isn't participating in the economy. Participation requires distribution. Distribution consists of on the ground sales, channel partners, post sales support, marketing, education, a developer ecosystem, trust, security and compliance buy in, workforce training, and a host of other components that a new company doesn't have. The vast majority of startups with great products perish because they don't have this and there's few shortcuts.

              Monopolistic practices are really neither here nor there. You can see distribution win over product quality in every industry. Out of product, brand, and distribution, product contributes the least to market cap.

        • pergadad2 days ago
          I would argue their main valuation comes from lock-in. They managed to basically grab the entire public sector and big corp OS and office software space, and continuously expand from this into communication, process management, etc.

          No or government in the world currently functions without Microsoft...

      • asah2 days ago
        Once a government or company adopts Microsoft (and 95+% have), it's *very* difficult to migrate, allowing Microsoft to charge whatever they want and force you to buy other products, including Azure.
        • jjcob2 days ago
          At a non-profit that I contribute to we recently switched everything to Microsoft because it was the easiest way for ~10 people with varying backgrounds to collaborate, when you have only a part time technical person who has limited time to set everything up.

          Costs a bit of money, but dealing with Google Docs, Synology, Dropbox, email and web hosting providers took so much effort. Setting up Sharepoint etc. isn't trivial either but at least it's a standard setup that you'll always find someone who can manage it.

      • Sevii2 days ago
        Windows isn't even Microsoft's main product anymore.
      • rendang2 days ago
        Windows can't be even 1 of those 3T by itself right?
      • jjcob2 days ago
        Because they are the ones who have Office, which 90% of non-tech businesses and governments run on.

        Word, Excel and Powerpoint are the standard tools everyone uses to get their job done.

      • outside12342 days ago
        By running Linux in their cloud
      • rvz2 days ago
        You are comparing two different things.
      • buyucua day ago
        I doubt Microsoft makes much money from Windows anymore. Microsoft's valuation primarily comes from Azure and other cloud products.
    • ChadNauseam2 days ago
      Claude is simply much better. Additionally, since it’s possible to train on synthetic data from other LLMs, it’s easier to catch up to competitors than it is to surpass them
      • rvz2 days ago
        > "Claude is simply much better."

        Until lots of free AI models close the gap again. It is simply not good enough for such valuations like this for Anthropic.

        Any other valuation higher than 70B for Anthropic is really overvalued and such higher valuations are always defended by those who invested in it already.

        • Karrot_Kream2 days ago
          A few friends and I run Deepseek (and other LLMs) on some hardware we jointly own and offer endpoints to each other via VPN. Deepseek is a big step forward for open LLMs and is something I use when writing scripts that process lots of data because we've all split the capital to build the machine, but Claude was and still is a lot better than Deepseek in the kinds of day-to-day usage that I'm after. And with how cheap Gemini 2.0 is, I find myself considering whether to use my local LLMs even for data analysis.

          I realize that on benchmarks it appears R1 is very close to Sonnet 3.7, but in practice I use Sonnet 3.7 much more.

      • bugglebeetle2 days ago
        I would say “for now” and that Anthropic’s inability to execute (3.7 was only released in response to competition from DeepSeek) raises a lot of questions about their future. While I agree that DeepSeek is still behind, Anthropic also does not have the same caliber of ambitious talent that DeepSeek has and has become a magnet for safety-obsessed cranks, whom (in total) act as a drag on the company. That Dario has also been engaging in dull-minded saber rattling about China does not speak well about their ability to deliver (vs. posture) in the future.
        • eagleislandsong2 days ago
          > has become a magnet for safety-obsessed cranks

          I'm out of the loop. To whom are you referring?

        • 2 days ago
          undefined
    • nilkn2 days ago
      Company valuations don’t just take into account the current state, but also projections of the future state. Anthropic has a reputation of consistently being right on the cutting edge. Their products continue to get better and better and gain more and more traction across more and more use cases, even in the face of competition. This valuation simply reflects that investors believe these trends will continue. Another way of thinking about this is that this is increasingly an industry where creating one good model just isn’t enough. Just as important, if not more important, is the ability of the company to repeatedly and consistently match or exceed everyone else while responding in a very smart and nimble manner to constantly evolving market pressures. Regarding DeepSeek specifically, it’s a fantastic open model, but the incumbents have already surpassed it in almost every way, and are continuing to blaze ahead. Frankly, it’s old news already as far as the absolute frontier and pinnacle are concerned. And that’s where all the money is.
    • mechagodzilla2 days ago
      And the Llama series, and they have half a dozen closed competitors with similar performance/capability.

      It really seems impossible to justify these evaluations from any kind of economic perspective.

      • buyucua day ago
        Llama was great because it showed that open LMs can be competitive. But today Llama is simply not in the top league anymore.
    • somanyphotons2 days ago
      At any point the open models could fall behind new closed offerings
      • mechagodzilla2 days ago
        But the open models create a (rapidly rising!) 'floor' - models that are worse/less capable than the best open-weight models effectively have zero economic value to their creators, even if they just spent $10B to create them. What fraction of queries require a better-than-R1 answer? Every time that floor rises, it puts even more pressure on whatever narrow, temporary advantage these closed companies can achieve to actually justify the valuations.
        • lanstin2 days ago
          All the interesting benefits of LLM/generative AI grow exponentially with the intelligence of the model. There's no point in a 90% capable model, except something like replacing phone trees or extremely bad customer service.

          A model that is 50% less likely to try to prove false theorems tho is much more useful than than one that happily tries.

          I'm not sure that HN folks appreciate this, but making commercial software is way easier than math and science (all the evidence you need, except the experience of understand QFT or advanced mathematics) is found in HN itself.

        • nilkn2 days ago
          I think for at least the next several years at minimum we’re going to see people continuing to integrate large language models more deeply into more aspects of their workflow. Many people will become used to what’s only possible with the absolute best and will simply lose interest in second rate models. I have no doubt that at some point this process will stabilize, but we’re definitely not close to that point yet.
          • ndiddy2 days ago
            The problem I see is that all the current AI leaders are spending way more than their revenue in training costs to stay ahead of the open source models. As an example, in 2024 Anthropic made $908 million in revenue, but spent $5.6 billion on training. If they cut down on training costs to become profitable, it would only take a year or two before open source models outperformed Anthropic's models, essentially making them worthless. With this in mind, I don't see how it's possible for Anthropic to generate a return on investment.
            • somanyphotons2 days ago
              I don't see how it's possible for the open models to continue to get better without similar levels of extreme money expenditure on training hw. Someone is paying
      • tylersmith2 days ago
        They've always been behind. They're closer than ever but still miles away from Claude's code abilities.
      • buyucua day ago
        likewise closed offerings could also fall behind open models.

        No matter who you are, the rest of the world combined will always be bigger. That is why Linux won the cloud, and why Windows server simply could not compete.

    • ttoinou2 days ago
      If it’s 20% better it can still be worth 10x more
    • psychlops2 days ago
      I'd invest large amounts of other people's money into it too.
    • tylersmith2 days ago
      No free models are even in the same class as Claude when it comes to code. Investors are clearly confident that moat won't be crossed soon.
      • BeetleB2 days ago
        According to Aider's leaderboards, R1 is not that far behind Sonnet (without thinking): 56.9% vs 60.4%

        Sonnet with thinking is at 64.9% (the top), but it's pricey - about double the cost without thinking.

        https://aider.chat/docs/leaderboards/

      • buyucua day ago
        In my experience DeepSeek is far better than Claude for coding.
    • deadbabe2 days ago
      Not investing and being wrong will hurt a lot more.
    • ein0p2 days ago
      Deepseek and Qwen are in China, that's how. If you look at non-US asset prices, the only conclusion is either they are massively underpriced, or US assets are massively overpriced. You have industrial giants (with revenues and profits to match) with valuations smaller than a permanent negative profit silicon valley startup.
      • buyucu18 hours ago
        I have DeepSeek and Qwen running on my laptop and home server. I'm pretty sure neither my laptop nor my home server are in China.
    • tsujamin2 days ago
      By lobbying for government-bans on foreign models? /s
  • hnarayanan2 days ago
    Claude is my favourite.
  • ktusznio2 days ago
    Can someone explain what is driving the growth of xAI's valuation?
    • nradov2 days ago
      A lot of these large AI investments at huge valuations are essentially just buying lottery tickets. There are very few other opportunities available to large investment funds targeting high risk and high returns. They have to put their money somewhere so where else can it go? This gives AI company management huge negotiating leverage in setting valuations for late stage funding rounds.
    • margalabargala2 days ago
      Having an owner who also has control over procurement for the entire US Government is very, very valuable. The quality of the AI tech is irrelevant.
    • 2 days ago
      undefined
    • laluser2 days ago
      It's integrated into X, but they are separate companies.
  • rvz2 days ago
    Raising money like it is the year 1999.
  • logicallee2 days ago
    $61.5B market cap is a steal for Anthropic. I pay $200/month for ChatGPT pro and the latest model o1 sometimes gets stuck on something in code that Sonnet 3.7 gets instantly. You can see a long list of things I built with Claude AI right here:

    https://news.ycombinator.com/item?id=43164257

    EDIT: since I posted this 54 minutes ago I started this MVP of a valuation calculator: https://claude.site/artifacts/4e7c0461-d62f-4c47-b285-971af2...

    • mi_lk2 days ago
      Any reason to keep pro subscription if you're satisfied with Claude?
      • logicallee2 days ago
        They are good at different things, they don't have the same skillset and when one gets the stuck the other can do it. It's kind of like asking "Any reason to ever hire more than 1 developer if 1 developer is full stack and can do it all?"
  • blackeyeblitzar2 days ago
    Is Anthropic really relevant in this race? It feels to me like ChatGPT has the users, and enterprises are willing to switch between models, so what can Anthropic really offer in a defensible way?
  • Layerman2 days ago
    [dead]
  • 2 days ago
    undefined