82 Comments

1. 

   @yessirri686

   July 10, 2026 at 12:40 pm

   he is spot on

   Reply
   • 

     @DeltaV64

     July 12, 2026 at 5:50 am

     I was about to write the same thing. He is right on every level.

     Reply
   • 

     @420th-b8q

     July 13, 2026 at 8:30 am

     Plus hugging face just announced scanning/image recognition hardware in the healthcare space. Stella move.

     Reply
2. 

   @JulesKruijtzer

   July 11, 2026 at 9:14 am

   How did he market this, how he became so rich doesnt make sense at all. If you would have made this website…

   Reply
   • 

     @rns10

     July 12, 2026 at 12:02 pm

     Many things, first – they track the actual usage of API and provide no free credit or subsidy for your api calls like chatgpt/anthropic. Its pay as per usage from start. Keeping the bills predictable and consistent.

     Why people/companies like it ? because hugging face has many open source models and you have full control over what kind of data the model should be trained on and you can deploy your own model instead of relying on credibility of openai to use the out of the box models.
     And the pricing is based on very strict consumption, so companies need to optimize heavily to reduce cost, which is more relying than just paying whatever subs price openai or anthropic sets and keeps increasing.

     Reply
3. 

   @richardleenknegt4749

   July 12, 2026 at 8:54 am

   Insightfull!

   Reply
4. 

   @tangobayus

   July 12, 2026 at 12:13 pm

   As long as I can download models, they are not in control.

   Reply
5. 

   @AbhishekSingh-xd1pf

   July 13, 2026 at 1:02 am

   last name rhymes 😄

   Reply
   • 

     @just_p3k

     July 13, 2026 at 5:12 pm

     yea, noticed too

     Reply
6. 

   @indiecrypto

   July 13, 2026 at 1:52 am

   Guys, the folks on Anthropic are very good people. They want safety of you and the world. Why can’t you understand open source is a threat for the world? Only Dario should control the world. Dario will overthrow the President and control the world for the safety of the world.

   Reply
   • 

     @ジャックマンジャック

     July 13, 2026 at 3:16 am

     😂😂😂

     Reply
7. 

   @issiewizzie

   July 13, 2026 at 3:43 am

   Question though what happens if the US government classes Hugging Face as a security risk?

   Reply
8. 

   @MMABeijing

   July 13, 2026 at 4:14 am

   I am not sure she likes his honest stance

   Reply
   • 

     @just_p3k

     July 13, 2026 at 5:13 pm

     exactly. reporters 😒

     Reply
9. 

   @LovableCutiess

   July 13, 2026 at 4:27 am

   There arent 17M people that can train agents. being a hugging face user does not make you an agent trainer.

   Reply
10. 

    @dustsucker4704

    July 13, 2026 at 4:31 am

    I only differ slightly from his opinion regarding openai and anthropic i think they expanded to much an will colapse either fully or almost dead. As he said the cheaper alternatives are used. So where is the demand for 50$/milion token models going to come from? Even 4,4$ per milion is quite expensive if you use an agentic workflow like 50k per developer/Month thats just insane. No company is willing to pay that amount of money its cheaper to buy hardware and use that it only costs about 10-20k/Month and cna have multiple devs work with it not just one. I dont see the demand for API prices models go up atleast not for companys that can choose.

    Reply
    • 

      @kooshanjazayeri

      July 13, 2026 at 10:42 am

      Anthropic is highly conservative in contrast to openai and they’ve laid their foundation for enterprises and universities and so on, they are in a much better position

      Reply
    • 

      @dustsucker4704

      July 13, 2026 at 11:00 am

      ​@kooshanjazayeriyes but no one who can use a cheaper model will pay for the more expensive one if the cheap one does the job. How many people exist, that need the best model? 100m people in the world? Maybe?

      Reply
    • 

      @kooshanjazayeri

      July 13, 2026 at 12:35 pm

      @dustsucker4704 well, open-source models still need a computer to operate, many businesses does not need endless token generation, they can do their job with least amount of token, and then it all depends on the revenue – costs balance to make it work, I’m just saying these companies wouldn’t just die, sure the hype will go away and also their prices will go down after sometime, but they offer convenient solutions for plug-and-play with “seemingly” secure systems, yes there’s a bubble and too much hype right now but i think they are here to stay

      Reply
11. 

    @marx2000-y8v

    July 13, 2026 at 4:37 am

    More compute doesn’t mean better models, that’s the Dream Anthropic and OpenAI are selling you with more datacentres. They are about extraction. Build it on prem or private cloud.

    Reply
12. 

    @uselessoldman1964

    July 13, 2026 at 4:37 am

    The cartel of OpenAi Anthropic and NVidia would never normally get away with their shenanigans funding each other under a normal US Government but since they currently have the most corrupt President and his Oligarchs friends in crime running it and making themselves a personal fraudulent fortune things wont change. But like the current lawsuit against the memory monopoly cartel who knows what could happen in the future under a new US Government. All the US has proven to the world is they are “currently” totally unreliable, untrustworthy and only interested in their own personal interests.

    Reply
13. 

    @pete_shand

    July 13, 2026 at 5:01 am

    Great interview!

    Reply
14. 

    @EvertVorster

    July 13, 2026 at 5:24 am

    The argument that Chines models are only so good because they distill Western models will fall flat once Eastern models overtake Western models. Much more likely situation is that the Eastern models have a much larger research effort through open source and everybody working together, vs the West fragmented approach. In fact, quite a lot of Eastern tech has found its way into Western models.

    Reply
    • 

      @J3unG

      July 14, 2026 at 9:40 am

      that day has come and gone and Chinese are on top.

      Reply
15. 

    @excitedbox

    July 13, 2026 at 5:52 am

    I wish I could use huggingface but the website is so terrible I havn’t been able to find anything. The search sucks because no description field (or any useful data). No proper way to search for a use case. The model cards rarely tell you what it is for. Even when it lists a related research paper, the name is cut off and you have to click through to see an abstract.
    How are you supposed to find a pretrained model, or dataset when you can’t tell what you are looking at? I have made at least 10 attempts now including this week which is how I landed here.

    How many years now, and the website is worse than a minimum viable product.

    Reply
    • 

      @kooshanjazayeri

      July 13, 2026 at 10:40 am

      i agree that it’s seemingly made extensively for engineers who know where to look, although the search for models/datasets have a decent categorizations but there’s still too much clutter there, you’d better search elsewhere (maybe even with ai) to find what’s what and what’s the best model fit for your use case

      Reply
    • 

      @SS-hz4jo

      July 14, 2026 at 3:50 am

      Agreed. Evening the name “hugging face” is a goofy roadmap to a brilliant buffet of genius efforts.

      I use a YouTube channels that review and link directly to models I want to use.

      A search for “open source Model review” might help.

      Reply
    • 

      @joegaffney8006

      July 14, 2026 at 4:11 am

      it has an api and pythong client also for searching – site maybe could use some work and more stricter info requirements on model cards as is seems very much up to the author to update imporant feilds. Personanly would like a license to have to be allways required.

      Reply
16. 

    @officialmistel

    July 13, 2026 at 6:58 am

    Omelette du fromage

    Reply
17. 

    @BRuserOsaka

    July 13, 2026 at 7:33 am

    He’s thinks AI is going to work? 😂😂😂😂 he’s invested. Too bad

    Reply
18. 

    @BRuserOsaka

    July 13, 2026 at 7:40 am

    It’s just Nessus and Metasploit,, and a ping…..

    Reply
19. 

    @niklasm4298

    July 13, 2026 at 8:15 am

    Im so sry, but this french english holy moly Cooeeehmpooehniies xD
    Its like french peoples have less training data on saying “o” xD

    Reply
20. 

    @sidosphon4979

    July 13, 2026 at 11:13 am

    It’s impressive how insightful Clem is. Open weights beats renting black-box APIs, and his warnings on AI geopolitics are spot on. We desperately need more voices like his to keep the future of AI for all humanity and not a few.

    Reply
    • 

      @ComeAlongKay

      July 15, 2026 at 4:20 am

      People want there to be safety regulations on AI, then those are introduced and everyone loses their mind. Everyone wants to be negative it doesn’t matter what happens people will find a way to doom freak out be apathetic spread fear and take whatever view let’s them be more negative and risk less hope. People confused cynicism for intelligence.

      Reply
21. 

    @MAbed

    July 14, 2026 at 12:00 am

    34:29 it’s a great argument for why open source models can be better, and 34:53 is even more compelling why others should go towards open 🙂

    Reply
    • 

      @CuratedLearnings

      July 14, 2026 at 6:56 am

      can we trust chinese?

      Reply
22. 

    @jadanpoll4137

    July 14, 2026 at 1:58 am

    (Référence personnelle pour les personnes sourdes)

    Principale: Historical technological dominance in the machine learning ecosystem is a direct byproduct of open, collaborative research sharing rather than proprietary secrecy.
    Enrichissement: Innovation economics: The development of the Transformer architecture at Google and its subsequent open release catalyzed the entire modern LLM industry, proving that cross-institutional collaboration and public knowledge spillovers generate greater macroeconomic returns than closed intellectual property regimes.
    (14:06) “open source creates in a way the conditions for your AI leadership… the reason why the US is ahead now is because from 2016 to 2020 to 2023, the US was super open with open research, open source AI… The famous example is the T in ChatGPT came out from Google sharing an open source transformers”
    Aperçu: Network effects generated by open technical ecosystems outpace the developmental velocity of isolated, proprietary laboratories.

    Principale: Restricting artificial intelligence models behind closed commercial walls increases global security risks by engineering critical power asymmetries between select monopolists and the defenseless public.
    Enrichissement: Security architecture: Cryptographic safety principles, such as Kerckhoffs’s principle, state that a system should be secure even if everything about it except the key is public knowledge; obscure, proprietary barriers provide a false sense of security while delaying the collective development of robust mitigation patches.
    (19:07) “you don’t really make it safe by keeping it behind closed door for just a few players. You actually make it more dangerous because you create asymmetry of power and asymmetry of capabilities between some actors… and other people who can’t defend themselves. The way you make the world safer… is by… leveling up the playing fields.”
    Aperçu: True systemic resilience is achieved through distributed transparency and defensive parity rather than centralized informational hoarding.

    Principale: Elite frontier AI labs have accumulated unprecedented geopolitical leverage, establishing an unprecedented baseline of operational power relative to sovereign defense institutions.
    Enrichissement: Political sociology: The monopolization of critical national security infrastructure by private tech conglomerates alters classic state authority structures, forcing military institutions into vendor dependencies where sovereign defensive readiness is governed by commercial corporate executives.
    (✨)(20:37) “one of the… biggest risks in AI is concentration of power… If you would have told me a few months ago that an AI company could be in the situation of power compared to the American Department of War I would… have told you that’s crazy, but it’s actually what’s… happening.”
    Aperçu: When core national defense capabilities become downstream functions of private proprietary computation, sovereign state authority is subverted by capital concentration.

    Principale: The American regulatory discourse has shifted toward weaponizing safety concerns to stigmatize open-source development, creating structural counter-incentives for public scientific contributions.
    Enrichissement: Public policy: Regulatory capture frequently occurs when dominant market incumbents advocate for complex, safety-oriented compliance frameworks that small open-source entities cannot afford to navigate, effectively criminalizing decentralized open innovation under the guise of public protection.
    (22:22) “every time we talk about open source, now we… talk about how it’s unsafe, which is really… weird… and it creates this weird counter-incentive for anyone to actually do open source versus what it used to be where open source… was really celebrated”
    Aperçu: When altruistic knowledge sharing is culturally repositioned as an existential hazard, market incumbents successfully insulate themselves from decentralized competitive disruption.

    Principale: Open-source data sharing satisfies historical fair-use frameworks by facilitating non-profit education and public utility, contrasting with closed labs that exploit data for exclusive capital extraction.
    Enrichissement: Intellectual property law: The transformation vector of fair-use doctrine distinguishes between commercial exploitation that directly replaces the market value of the original content and public-interest usage that enhances baseline technological capabilities for the global commons.
    (26:37) “if some data and some data sets are shared in open source for everyone to use for free… not for profit… this is very different than… a lab using that to make billions of dollars of revenue without any public contributions.”
    Aperçu: Ethical data utilization is determined by its distributive destination; public enrichment legitimizes shared inputs, whereas private exploitation demands strict transactional accountability.

    Principale: Building open platform infrastructure requires severe capital efficiency and long-term sustainability to protect the developer ecosystem from the volatile pressures of venture fundraising cycles.
    Enrichissement: Corporate finance: Platforms that scale through network effects require low operational overhead relative to capital intensive foundation labs, insulating their corporate governance from investor demands for immediate monetization, which often breaks user trust.
    (29:22) “we’ve always taken kind of like a bit of conservative approach to things and not necessarily… maximizing short-term revenue, but instead… focusing on… long-term sustainability… we’re quite capital efficient… we optimize more for… long-term sustainability of the company than… fundraising… maximization.”
    Aperçu: Preserving the alignment between infrastructure providers and decentralized user ecosystems requires insulating the enterprise from short-term financial engineering.

    Principale: Macroeconomic over-investment in lookalike text LLM APIs has created a localized market bubble while systematically starving foundational deep-tech vectors like local computing, biology, and chemistry.
    Enrichissement: Behavioral finance: Herd behavior among venture capital funds often leads to massive asset bubbles in highly visible consumer-facing applications, resulting in redundant capital deployment that overlooks capital intensive industrial applications requiring deep scientific validation.
    (31:47) “we were probably in a LLM API bubble, but definitely not in an AI bubble because there are a lot of domains… that are under… invested. For example… local AI… biology, chemistry… all these domains have seen very very little investment compared to… text LLM APIs”
    Aperçu: Retail-facing commoditization distorts investor judgment, hiding highly generative scientific infrastructure opportunities behind superficial software hypes.

    Principale: Physical robotics computing exponentially multiplies data storage requirements and elevates the absolute necessity for transparent open-source code to mitigate acute safety risks in private environments.
    Enrichissement: Risk engineering: Operating autonomous physical hardware within private human domiciles introduces physical liability hazards and data leaks that cannot be reliably mitigated by closed, proprietary software layers without severe trust violations.
    (✨)(33:41) “when we look at the robotics data sets on Hugging Face, they are huge… petabytes sized… when I think about kind of like having a robot at home that interacts with my environment… It’s even scarier… to have like a black box system just controlled by a few organizations.”
    Aperçu: The physicalization of artificial systems alters the risk calculus, making proprietary opacity intolerable when algorithmic failures carry real-world kinetic consequences.

    Reply
23. 

    @jadanpoll4137

    July 14, 2026 at 1:58 am

    (Référence personnelle pour les personnes sourdes)

    Principale: Enterprises follow a structural migration pattern where they prototype using frontier APIs but transition to open-source models at scale to eliminate unsustainable token costs.
    Enrichissement: Cloud economics: This mirrors the classic cloud repatriation trend where early-stage startups leverage AWS or Azure for rapid feature deployment, but mature enterprises transition to private infrastructure or hybrid environments once compute costs scale non-linearly with user growth.
    (02:48) “the typical kind of like flow is that companies starting by using frontier APIs maybe at the beginning, you know, to experiment to launch the new feature. And then when they really hit production and they hit scale, the cost is starting to be too big really with with frontier models. And so, that’s usually when they switch to… open source models to… power their… workload.”
    Aperçu: API-driven architectures are transitionary tools for rapid validation, whereas open-source systems represent the long-term operational equilibrium for scalable enterprise computing.

    Principale: Enterprises refuse to permanently outsource their core technical competencies to third-party black-box APIs, treating AI ownership as equivalent to controlling a proprietary software stack.
    Enrichissement: Corporate strategy: In the software industry, outsourcing core runtime dependencies creates severe vendor lock-in and operational vulnerability, a lesson formalized by the widespread enterprise adoption of Linux and open-source database engines over proprietary alternatives in the early 2000s.
    (05:27) “companies want more control basically and and transparency in the systems they use… you don’t want to outsource your core capabilities, AI, to another company the other kind of like a black box API that that you don’t control… This kind of like idea that companies need to own AI and own models instead of renting them… makes a lot of sense”
    Aperçu: Technological sovereignty demands asset ownership over utility rental to preserve strategic differentiation and long-term viability.

    Principale: The rise of autonomous AI agents is lowering the technical barrier to machine learning, transforming traditional software engineers into self-sufficient AI model builders.
    Enrichissement: Software engineering history: The democratization of model optimization via abstraction layers functions similarly to the introduction of high-level programming languages like Python or Ruby, which bypassed assembly code constraints and exponentially multiplied the global developer base.
    (07:50) “with… agents we’re seeing that it’s becoming easier and easier for software engineers to… run their own models, optimize their own models, train their own models. And we’re seeing that like across the board… not only from… smaller startups… but also… in enterprises.”
    Aperçu: Tooling abstractions convert hyper-specialized scientific workflows into standardized, accessible engineering practices, accelerating systemic adoption.

    Principale: Enterprise software differentiation is achieved through progressive optimization and post-training of open weights rather than consuming identical off-the-shelf API solutions.
    Enrichissement: Competitive strategy: Michael Porter’s frameworks indicate that reliance on standardized, externally managed inputs commoditizes competitive execution, whereas embedding proprietary business logic directly into the software infrastructure creates a defensive, non-replicable operational advantage.
    (10:07) “You start from really off-the-shelf solutions and then you end up by really controlling a lot of the workload yourself and building a lot of the systems yourself, which creates actually your differentiation from other companies and other organizations, right? Like you… want to build these skills of like building AI systems better than your competitor, and that’s what’s going to differentiate you in the long… run.”
    Aperçu: Strategic alpha shifts from accessing data processing capabilities to natively customizing and owning the underlying algorithmic weights.

    Principale: Chinese open-source machine learning models are systematically eclipsing American alternatives in developer mindshare, capturing the absolute majority of global model downloads.
    Enrichissement: Global trade dynamics: The rapid proliferation of Chinese open weights across Western platforms bypasses geographical trade restrictions, establishing an infrastructural dominance that embeds specific algorithmic paradigms into global commercial systems outside state oversight.
    (✨)(10:58) “Chinese models accounted for most of the downloads, right? Like 41%. So China’s surpassing the US monthly and in overall downloads.”
    Aperçu: Open-source distribution channels allow secondary nations to capture global developer ecosystems, circumventing geopolitical containment strategies.

    Principale: The Western research community, elite academia, and commercial scaleups are structurally dependent on Chinese open weights due to the un-examinable nature of closed American APIs.
    Enrichissement: Epistemology of science: Validated academic research requires reproducible empirical conditions, a constraint that renders proprietary black-box APIs mathematically useless for peer-reviewed validation, forcing institutions to default to transparent open weights regardless of national origin.
    (12:31) “majority of the scale ups in the US that are using open source are now using open… weights from… China… all academia… because the only way to really learn, study, do research on AI is to have open source and open weights, right? You can’t really study an API because it’s a complete black box.”
    Aperçu: Scientific progress enforces systemic transparency; proprietary hoarding inherently alienates the global research ecosystem.

    Reply
24. 

    @SS-hz4jo

    July 14, 2026 at 3:30 am

    RAM prices are artificially high. All companies buying compute or selling compute are knowing putting RAM and GPUs on their balance sheet to make it more expensive for consumers to buy a home computer and freely run open source models. Corporations that use commercial buoying the circular Ai financial bubble. If you want RAM and GPUs prices to come down, don’t support commercial Ai companies. No SpaceX, Meta, Microsoft, Nvidia, Amazon, OpenAi…

    Reply
25. 

    @joegaffney8006

    July 14, 2026 at 4:06 am

    Still people are having to rent compute (pretty much impossible to buy enough of your own hardware ATM) and its not allways cheaper than using an API provider depending on your use case as there is lots of work to do on efficency particuallry with LLM batching, throughput etc.

    Local for privacy and full ownership is a very valid reason, but the case that its allways cheaper is not true.

    Reply
26. 

    @palpalps

    July 14, 2026 at 5:02 am

    Its not just about downloading and running models on your own hardware; the fact that model weights are out in the open also means 3rd party providers are incentivized to serve them as close as possible to the cost to run them, without unnecessary premiums on top of that. And you can choose who you trust for the dependency, without vendor lock-in.

    Reply
    • 

      @TheDoomerBlox

      July 16, 2026 at 7:56 am

      I would not say “incentivized” to serve them close to the electricity cost plus a reasonable return on investment time for the hardware, but they certainly can’t get away with a rugpull like what is extremely fashionable these days;
      where suddenly, one day, prices jack up 10x ‘take it or leave it’ (what a shame we’re the only game in town)

      Because if the software isn’t locked down, someone else could now spin up an alternative at less than 10x the original cost, and steal the market.

      Reply
27. 

    @Alex-kg1xh

    July 14, 2026 at 5:52 am

    The question is how Chinese open-weight companies make money

    Reply
    • 

      @J3unG

      July 14, 2026 at 9:50 am

      That was never a core concern for the chinese. The Chinese want AI to be only a certain aspect of their industry. It’s not the core of their industry. Manufacturing who are the Chinese at their best and they are leading and won’t definition. AI only assists that. And again at this point is moved because the Chinese give no f**** about the profitability of ai. It seems that America is the only one who cares about it and frankly it’s because they have nothing else. Wall Street and the current Administration seems to think AI is going to somehow save this country and that’s why the crazy investment. This will lead to a bubble which will happen shortly. As a matter of fact as SpaceX stocks go down and open AI delay their ipo, the bubble will burst very soon perhaps by the end of the summer of 2026.

      Reply
    • 

      @legacyexploits8501

      July 15, 2026 at 8:37 pm

      Because the Value isnt it the AI but the productivity it enables. America focuses as selling AI as a product/service. China is distributing it as a tool to increase productivity gains and RND of their already existing market.

      Reply
28. 

    @Tamos40000

    July 14, 2026 at 6:08 am

    The business model of OpenAI and Anthropic is idiotic. Instead of billing cloud resources with which the company is running their models, they’re billing per token instead. Your clients being unable to stop a costly computation once it launched and having very little control on how much they’re spending or how the resources are distributed is shooting yourself in the foot. Those are easily solvable problems, but I won’t complain about people turning to open source because those companies are unable to fix their services.

    Reply
29. 

    @dzlllz

    July 14, 2026 at 6:25 am

    what an absolute legend!

    Reply
30. 

    @dwiss2556

    July 14, 2026 at 6:47 am

    The easiest way to think about the AI issue is in terms of ‘thoughts’. If a child is kept from certain aspects it will never develop certain abilities. This works for both ways, the good and the bad. That is why there is guidance by parents, but control fails at the same time. AI is not much different to this. Keep it locked away and face missing out on a large part of understanding and ability. Control. Put up decent and internationally accepted restrictions, regulations and guidance and humanity is at the heart of it. Guidance.

    Reply
31. 

    @albin1816

    July 14, 2026 at 6:50 am

    It makes no sense to have self hosted AI on your company, unless you’re a global company with hundreds or thousands of developers who can utilize the hardware at least 70%+ at all times.
    In the same way that it makes sense to share electricity – it also makes sense to share hardware from a resource utilization and cost effectiveness perspective.

    Reply
    • 

      @funtechu

      July 14, 2026 at 8:49 am

      That’s not even remotely correct. It makes sense to have self-hosted models even as an individual user. There are several lightweight models that are excellent for most common daily tasks and can run locally.

      Reply
    • 

      @albin1816

      July 14, 2026 at 10:41 am

      ​@funtechu Go ahead and make it make sense. In what world do you need sub 10ms latency on a response from an LLM?
      When you’re shooting off even a huge 30-40min implementation of a new feature, it doesn’t matter to me if that is done at minute 40 or at minute 60, i’m usually away on lunch or some other task at that time.
      Fine, you can have a shitty model run code reviews locally e.g. using your already existing GPU on your PC to run code reviews over night, that’s the only thing that would make sense, but not even then does it make sense. The hardware they use in big data centers are so much more efficient for these AI workloads than my dev work PC running a 5060 ti.
      From an ecological footprint perspective it makes no sense. Please make it make sense, I’m open to changing my mind.

      Reply
    • 

      @funtechu

      July 14, 2026 at 11:13 am

      ​@albin18161) The cost is lower in many cases. 2) You own your own implementation so it isn’t subject to the risk of changing prices, billing practices, etc. 3) From a data security perspective, it’s safer and can be run on air gapped systems.

      Reply
    • 

      @NicholasIstre

      July 14, 2026 at 5:36 pm

      For pure resource utilization, sure — but that’s not the only metric that matters. Data privacy, legal compliance, vendor lock-in, and geopolitical risk are all real concerns. You’re at the mercy of frontier AI companies’ pricing, availability, and government-mandated restrictions. Plus, self-hosted open models let you customise and fine-tune for your specific use case — something you can’t do with a black-box API. Yes, it doesn’t make sense for every company to run its own AI cluster. But there are many valid reasons to keep some parts of your stack self-hosted.

      P.S. This was drafted with my locally run Qwen 3.6 35B model. The electricity this took is basically a drop in the bucket compared to the total AI power consumption.

      Reply
    • 

      @albin1816

      July 15, 2026 at 2:52 pm

      @NicholasIstre For data privacy, and custom / fine tune use cases sure, that is a valid point.
      My perspective is generic software development and that’s very similar cases everywhere across the world in how code gets written whether its for banks, games, etc. From that perspective (the 80% of people use case), to me it doesn’t make sense if we want to think of the planetary resources. Hopefully we can have some competition with privacy focus hosting these data centers with open models, similar to what we have seen with the cloud (e.g. Nordlo in sweden), so its not a monopoly.

      Reply
32. 

    @WEB3GISEL

    July 14, 2026 at 7:52 am

    It is most important for the core intelligence of a business to be CONSISTENT.

    Reply
33. 

    @DavidBoura

    July 14, 2026 at 8:04 am

    VOCAL FRY FROM HELL NO THANKS

    Reply
34. 

    @chiguireespacialespecial

    July 14, 2026 at 9:05 am

    🤗

    Reply
35. 

    @J3unG

    July 14, 2026 at 9:37 am

    Current AI tech started from Open Source. there is no good reason for these so-called Frontier or for-profit AI systems (well…besides greed). Hugging face and the Chinese models were always set to be the prominent platforms for AI open source in anthropic and even to an extent Gemini and Microsoft co pilot we’re never going to be profitable. If anything, it’s an attempt by the check oligarchs to make some money when no one was looking. AI will always be open source as it was in the beginning.

    Reply
36. 

    @J3unG

    July 14, 2026 at 9:47 am

    The dude is not correct when he says open Ai and anthropic are going to be fine. They’re not. I guess he just dismissed the fact that both of these companies are actively seeking as much financing cuz they can just stay alive. I think he also ignored that these companies are not profitable at all. The idea that AI can be profitable when it came from open source is ridiculous. AI what continue to be open source. And the Chinese will have a lot to do with this. Hugging face is a good company but but as it compares to the Chinese enterprise, is just a hobbyist website. His position also makes me wonder if this guy has investment in openai or anthropic. In any case he’s wrong.

    Reply
    • 

      @ArnaudMEURET

      July 17, 2026 at 11:18 pm

      You sound like you don’t understand why OpenAI and Anthropic are in high-burn phase. It’s a development strategy, not a business model. The profitable business model is already known. They’re just spending cash to acquire and lock customers in. Customers who, rightly so, prefer AWS to massive CapEx, will always prefer SaaS SLA-backed AI. As soon as their critical customer base is established, free and cheap tiers will vanish, spooking away the millions of leeches (I am one) that are unprofitable.

      Reply
    • 

      @liberteus

      July 19, 2026 at 4:29 pm

      “the profitable model is known”
      Oooh so many people would like to know! Because so far, none of the llm companies is profitable.

      Reply
37. 

    @truefacts404

    July 14, 2026 at 10:12 am

    They are not “ taking an opportunity “ to share collaborate lol they are forced to lol. As far as not bowing to the “ Department of Defense” it’s called the constitution that was their right sir

    Reply
38. 

    @lawoei

    July 14, 2026 at 11:50 pm

    Open source AI has matured to the point the average Joe can run a decent 32B model on their laptops today! Only government enterprises rent NPU/ML for deep frontier simulation.

    Reply
    • 

      @Sam-o7n7v

      July 15, 2026 at 7:24 am

      Careful being too confident in what you believe to know.

      Reply
39. 

    @LeonVanDyk

    July 15, 2026 at 5:01 am

    Memory costs still bedeviling local AI

    Reply
    • 

      @MikeStoneJapan

      July 15, 2026 at 6:46 pm

      That part. Especially with anything after q1 2026. I’m hoping and praying so hard for the unused data center gpu selloff

      Reply
40. 

    @DeannaSharing

    July 15, 2026 at 12:31 pm

    Shoot. I worked for a Fortune 500, and if they can get it free, that’s the way to go. If they pay for something they rachet it down to as few users as possible.

    Reply
41. 

    @jasonscala5834

    July 15, 2026 at 11:07 pm

    Did you hug his face?

    Reply
42. 

    @EnjoyurbleThings

    July 16, 2026 at 12:04 am

    Wait. This isn’t Clem Fandango.

    Reply
43. 

    @dehilster

    July 16, 2026 at 5:35 am

    Can people make money using these models. That is the question.

    Reply
44. 

    @dehilster

    July 16, 2026 at 5:36 am

    Open source models still hallucinate. Extraction is still not there.

    Reply
45. 

    @dehilster

    July 16, 2026 at 5:38 am

    Another take he is wrong. Frontier models will not exist. They are not maintainable financially. I hate all this vibing about technology. Too much simplistic thinking. People like hugging face are seeing experimenting with these models but do not check to see if anyone is making money with them.

    Reply
46. 

    @ВЛАДИМИРВОРОБЬЕВ-ш5й

    July 16, 2026 at 6:03 am

    OKAY this book actually wrecked my week in the best way. Smart Broke Dumb Rich by Zor Veyl. always thought my family was just bad with money but nah, nobody ever taught them anything. work, pay bills, stay humble, survive. thats not money education thats just being tired on schedule.

    Reply
47. 

    @NomaseGiu

    July 16, 2026 at 6:08 am

    i NEVER push finance books on people, they all make you feel broke and dumb at once. this ones different. Smart Broke Dumb Rich by Zor Veyl actually doesnt talk down to you. read it and it just clicks that most of us got handed survival advice, did everything right, then got blamed when it didnt turn into freedom. been stuck on that for days.

    Reply
48. 

    @МишаКисляк-г3м

    July 16, 2026 at 6:13 am

    Очень информативно и без лишней воды. Спасибо!

    Reply
49. 

    @UserName-q8r-m9h

    July 16, 2026 at 7:10 am

    The famous business phrase “Nobody ever got fired for buying IBM” (or “choosing IBM”) means that employees prefer choosing large, established, and safe vendors to protect their own careers.

    Sure, some open sources AI models are cheap or free, but if your businesses are MISSION CRITICAL, choosing cheap or free AI models could easily get you being fired when the systems break down and there are no support.

    Reply
50. 

    @priapushk996

    July 16, 2026 at 7:10 am

    Hugging Face? More like Mogging Face.

    Reply
51. 

    @FindingNemotron

    July 17, 2026 at 3:08 am

    Generalization is not a business, it’s a path to mediocrity. AI should be tailored to a person or a business like a suit.

    Reply
52. 

    @pavelkravchenko2810

    July 17, 2026 at 7:33 am

    bullshit

    Reply
53. 

    @steveclark9934

    July 17, 2026 at 4:34 pm

    Foolish to trust somebody else with your data definitely go with home based local AI at least that way you got a chance

    Reply
54. 

    @sevenismy

    July 18, 2026 at 8:13 am

    Open weights models main point is non-USA models,

    Reply
55. 

    @timsonner

    July 18, 2026 at 7:06 pm

    Amazing, thank you!

    Reply