Anthropic

The US is accusing China of plagiarizing Anthropic models. We have three problems with that accusation.

by

Michael Kratsios, assistant to the president of the United States, did not bite his tongue this week when claimed that “We have information indicating that Moonshoot AI distilled Fable from Anthropic for the development of its K3 model.” Or in other words, they accuse China of plagiarizing the American company’s advanced AI model. There are several fundamental problems with that accusation. Let’s see them. Fear of Kimi K3. This model has just burst onto the AI ​​scene with overwhelming force. Benchmarks show that Kimi K3 is one of the best AI models in the worldwith performance close to Fable 5 and GPT-5.6, the best public models from Anthropic and OpenAI. This milestone has triggered its popularity among users and companies, and also the alarms in Washington. The immediate consequence has in fact been political, because since the Trump Administration there is now open talk of sanctions for “theft of technology” if the accusation is confirmed. The accusation. In his text, Kratsios explains the supposed mechanism used by Moonshot AI to copy his model. It states that the Chinese startup would have built an internal platform to make mass queries to US models, changing access methods to avoid detection and then transferring its capabilities to Kimi K3. He also adds that the company has managed to access servers with Nvidia GB300 chips in countries outside China to avoid export restrictions. Where is the evidence? The first problem we have with that speech is that a key piece of that speech is missing from the start: Kratsios does not support this accusation with technical evidence. It does not provide usage records, it does not provide examples of prompts and responses, nor query patterns. Nor does it provide independent analyzes that any researcher could review. What we have right now is an official story and accusation, not a proven “infringement.” The controversy over distillation. The key word in that speech is “distillation.” In AI, distill a model It means training a new one using the responses of another. The “master” model answers thousands or millions of questions, and the “student” model learns to imitate its reasoning patterns at lower cost and size. The industry itself describes distillation as a legitimate and standard technique when applied to permitted or proprietary models. The red line appears when it is applied to third-party models, on a large scale and without permission, which according to the US Government is theft of intellectual property. But crime, what is called crime, is not. The second problem is that the statement and the accusation are not supported by any clear and defined legal framework. Neither the US nor China currently have a specific law that says under what conditions distilling a rival model is an intellectual property crime. You may violate an API’s terms of service or even end up obtaining sensitive information from companies, but there is no jurisprudence about it. In fact, until recently the debate in the US was about regulating US frontier models like Mythos, which were considered too dangerous. Suddenly the debate is now on sanctioning the distillation of models. Everything is moving too quickly, once again, to adapt the legislation. Hypocrisy made in USA. The third big problem we have with that accusation is that the big American models—including those at Anthropic—have been trained on massive data sets that mix websites, books, code, newspaper articles, and other materials of which a notable portion is protected by copyright. own Anthropic reached an agreement with justice these days for that reason, Meta has already discovered it stealing terabytes of copyrighted books to train your models. Double yardstick. In the US, AI companies defend themselves by arguing that That falls under “fair use” of the content, but the authors neither gave their permission nor charged for it. That same industry and that same Government that have made it normal to train models on other people’s content complain that a Chinese startup uses the outputs of its models to train its own. They are not identical practices, but the double standard is difficult to ignore. AI, once again, as a weapon. The Trump administration is taking this issue to another scale. By accusing Moonshot AI of using restricted chips and developing Kimi K3 by distilling Fable 5, he places distillation in the same box as industrial espionage or the theft of military secrets. AI is no longer a software product and becomes a strategic technological weapon. And there it is not so important to provide evidence or be technically precise: it is important that what China has done is an attack on its (former?) technological hegemony. If this type of accusation sounds familiar to you, you have a good memory. In Xataka | An AI model did not have access to the internet. So he thought it was better to have it and decided to hack something along the way

Anthropic was accused of stealing books to train its AI models. Their solution has been simple: pay

by

A federal judge in San Francisco has given the final green light to the $1.5 billion agreement between Anthropic and a group of authors and publishers that They reported her for using her books without permission to train Claude. And after extracting millions of books from digital libraries without authorization, and from physical bookstores to scan and destroy en masse, everything has ended up being settled with money. What has happened? Judge Araceli Martinez-Olguin definitively approved this Monday the agreement that Anthropic reached with the plaintiffs, as collect Reuters. The figure was already known since last year, when Judge William Alsup gave his preliminary approval, but this last step was missing for the money to begin to be distributed. How we got here. A group of writers sued Anthropic in 2024 accusing her of using copies of her books to train Claude without her consent. Judge Alsup ruled in June 2025 that training AI with copyrighted books falls under “fair use,” a decision that set an important precedent for the entire industry. The judge also clarified that, although training with those books was legal, Anthropic had broken the law by downloading more than 7 million copies from digital portals without authorization and storing them in a kind of internal library that was not necessarily intended for training, according to explains Reuters. That specific point was going to be decided in a separate trial, with compensation that could skyrocket to hundreds of billions of dollars. So to avoid this, Anthropic preferred to make an agreement. Between the lines. As we explained some time ago, the court documents that came to light revealed the real scope of the project, baptized “Panama”, and with which the company came to physically buy and scan millions of bookscutting their loins to digitize them and then recycling them. Before going this route, Anthropic employees, including its co-founder Ben Mann, had downloaded books directly from unauthorized repositories such as LibGen. The company has always maintained that this content was never used to train a business model, but it was precisely that part of the process that took it to court. How much and to whom. The distribution is made at a rate of about 3,000 dollars per work, out of an estimated total of 500,000 titles between authors and publishers, according to details TechCrunch. Aparna Sridhar, deputy general counsel at Anthropic, said in a statement that the company reached this agreement in 2025 after the ruling that recognized training with books as fair use, and added that more than 91% of the affected authors and publishers have already claimed their share of the payment. For his part, Justin Nelson, lead attorney for the plaintiffs, has called the settlement “historic” and has stated that it represents the largest copyright recovery ever achieved, according to collect also Reuters. Not everyone is satisfied. Some authors have raised objections, arguing that the amount was insufficient, benefiting plaintiff’s lawyers too much or unfairly leaving out certain rights holders. Judge Martinez-Olguin has rejected these arguments, considering that they did not realistically reflect the risks of having gone to trial, and approved fees of more than $101 million for the lawyers, below the $187.5 million they had requested. according to Reuters. Some authors and publishers directly opted out of the agreement and have their own lawsuits against Anthropic that are still ongoing. Why it is important. This closure does not resolve the underlying legal debate for the rest of the sector. And since it is an agreement and not an appealed ruling, Alsup’s decision on “legitimate use” will never reach an appeal court, so it does not constitute binding jurisprudence, as underlines TechCrunch. Each judge is free to interpret similar cases in their own way, and that is just what is happening with the lawsuits opened against Google, Meta, Midjourney and OpenAI for similar reasons. Cover image | Anthropic, edited by Xataka In Xataka | It took 15 years for computers to improve the productivity of companies. AI may take longer for one reason: sabotage

Anthropic claims to have found Claude’s “consciousness”, although the data tells another story

by

Anthropic has published an article that promises to change the way we understand the inner workings of your business models. artificial intelligence (AI), like we explain to you last week. This company claims to have identified a kind of global workspace within Claudewhich he has named J-Spacewhere the model analyzes and manipulates concepts before these become part of the final answer. What is surprising is that, according to Anthropic, this space has not been deliberately programmed: it has emerged as a byproduct of training the model. To analyze it and understand how it works, a technique known as J-Lens which is capable of mapping the internal activations of the model onto words from its own output vocabulary. When asking Claude to solve a multi-step mathematical calculation the visible answer contained only the final result, but within the J-Space each intermediate step was represented separately. Without this technique no part of that process could have been observed from the outside. However, Anthropic itself recognizes that much of what a language model does does not even go through this internal space. Speaking fluently, remembering simple facts or applying basic grammatical rules almost completely dispenses with J-Spaceand when the company blocked its use Claude continued to function normally, although he lost some of his higher-order cognitive functions. Claude “knows” that he is being watched During one of his most striking experiments, Anthropic subjected Claude to injection attempts. prompts during data acquisition processes. He J-Space then showed words like “false”, “injection”, “erroneous”, “prompt“, “fraud” and “poison”, while the visible response of the model completely ignored the attempted manipulation. Interestingly, when pressed to comment on the matter, terms such as “transparency”, “honesty” and “truth” emerged in the internal space. The language used in the article sometimes borders on the idea of ​​an emerging consciousness The Anthropic researchers also observed that Claude seems to recognize when he is being evaluated, which leads him to behave differently when faced with tests than when faced with prompts everyday. In a particularly revealing test, Anthropic removed J-Space the vocabulary associated with that consciousness of being put to the test (terms such as “fake”, “fictitious” or “scenario”) and found that Claude became much more vulnerable to attempts at blackmail or provocation. Be that as it may, the researchers’ report admits important limits: the J-Space is restricted to a single-token vocabulary, so plans or concepts that cannot be expressed with a single word may not appear in the reading, even though the model is calculating them anyway. Anthropic also does not dare to claim that monitoring this space is sufficient to guarantee the alignment of a model. However, we must not overlook that the language used in the article sometimes borders on the idea of ​​an emerging consciousness. Neel Nanda, head of model interpretability at DeepMind, supports this finding as real evidence of a cognitive space within the modelsalthough he clarifies that the practical usefulness of J-Lens remains limited. This achievement, in any case, opens a promising avenue for auditing the honesty of models, although it is still far from being a complete window into a machine’s thinking. Image | Generated by Xataka with ChatGPT using a prompt created with Claude More information | Anthropic In Xataka | While most oppose AI data centers, there is one group enthusiastic about them: merchandise thieves

Anthropic has condemned him to “I would rather not do it”

by

The anonymous narrator of our story hires a clerk named Bartleby for his law firm. The employee starts working in an enviable way, but after three days, something happens. The narrator asks him to compare a document and Bartleby answers with a immortal phrase: “I preferred not to do it”. That famous quote is part of the story ‘Bartleby, the clerk’ that the American writer Herman Melville (author of ‘Moby Dick’) published at the end of 1853. The story was barely successful at the time, but over the years it gained more and more impact. That phrase became a kind of hymn to apathy, inaction and laziness, and these days we have been surprised to find a new Bartleby. one called Fables 5. Anthropic’s AI model was launched on June 9 as the most powerful, expensive and exclusive in historyand although its performance in benchmarks was spectacular, we barely had time to taste it: the US government decided that it was too good to be publicly available to non-US citizens. Then Anthropic decided to cut off access globallyand for three weeks Fable 5 was in limbo. This week Fables 5 has become available again, but from the beginning Anthropic warned that I did it with a few asterisks. The AI ​​model, they explained, brings new security filters to avoid misuse. In the original launch, the company already indicated that if the model detected problems in the conversation, it would not be Fable 5 that would answer. Instead, they said, “Queries on some topics will instead receive a response from our second most advanced model, Claude Opus 4.8.” With the “redeployment“Those filters would be even more severe. What we didn’t know was how much. Fable 5 prefers not to do things Multiple comments on networks like X or Reddit reveal that Fable 5’s performance has been absolutely artificially layered or, in industry jargon, “nerfed”. The model seems to behave worse, but the real problem is that it is a model that constantly resorts to Bartleby’s “I’d rather not”. Instead of doing what the user asks, it automatically makes the jump to Claude Opus 4.8, a model that is certainly capable, but is not what users would want because they intend to take advantage of the theoretical power of Fable 5. The worst thing is that Fable 5 resorts to that “I’d rather not” philosophy with practically everything. Dylan Patel, creator of the famous consulting firm SemiAnalysis, did the test with the traditional “how many R’s are there in ‘raspberry’” and Fable 5 warned him that the model could not answer that question. The performance of the model also seems to have been greatly affected by this redeployment, and independent tests They showed how in the BridgeBench benchmark the score of the original Fable 5 and the current Fable 5 was very different. That specific test It was too specific. as some analysts commented, and the problem is not so much that performance drops (it doesn’t do it too much, or even improve) like the one in Fable 5 just refuse to answerwhich also does very difficult Evaluate your current performance. Added to this factor is the fact that Fable 5 will be a particularly exclusive model: it can be used with Anthropic’s free and paid subscriptions until next July 7, but then this model will disappear from said plans and it will only be possible to use it with credits. That is, paying per use, which for intensive uses can be very, very expensive. We are therefore facing a particularly delicate moment: if companies like Anthropic begin to artificially limit the capabilities of their most powerful models or “route” them to inferior models at the first change… What is the point of developing these models? Commercially, the idea doesn’t make sense either: if business users are going to find that the model goes to “I’d rather not do that” and goes on to respond with an older and worse version, they won’t be particularly happy about paying for the “expensive” model. Anthropic is in a compromised situation, and what is happening with Fable 5 may end up marking a turning point in the way end users and professionals work with AI models. In Xataka | The future of geopolitics is played in AI models: Claude Fable 5’s veto indicates that Europe is offside

Anthropic is already preparing just the only thing it was missing to reinforce its AI leadership: its own chip

by

Tech giants have learned something the hard way: depending on third parties is a weakness. Anthropic seems to have realized the same thing, and has been mulling over an important idea for months: making its own AI chip. As indicated in The Informationthe company is in negotiations with Samsung for a potential collaboration, although at the moment everything is unknown. Surprise, none. Already in April Reuters Indian that Anthropic was considering the idea of ​​developing its own AI chips in order to respond to the chip shortage. It seems that this proposal is really gaining traction, because what was once a possibility has now translated into concrete conversations with one of the largest semiconductor manufacturers in the world. The shadow of OpenAI. The move comes just a week after OpenAI unveiled its own custom inference chip, baptized as “Jalapeño” and developed in collaboration with Broadcom. According to company officials, that chip offers better performance per watt than other inference chips, and that could leave Anthropic behind in this race. The reaction of the firm led by Dario Amodei therefore seems logical. They want to diversify, not replace. The strategy, however, is not to completely eliminate its current partners, but rather to have a more diversified strategy for the future. In statements to TechCrunch Anthropic officials have indicated that its “diversified hardware stack, which includes chips from Google, Amazon and Nvidia, will continue to be crucial to its computing strategy.” Everyone wants their own chip. The announcements and news surrounding OpenAI and Anthropic are not, as we said, any surprise. In the last two years we have seen how more and more technology companies joined the trend of having their own AI chips, when before they delegated that aspect to specialized companies like Nvidia. Thus, we have: Samsung, a perfect partner. The South Korean company has been deeply involved in the AI ​​industry for some time. It is a key partner of Nvidia, as it manufactures some of the chips it needs to train and run AI models with its GPUs. Not only that: Samsung is improving your OPC lithography of chip manufacturing thanks in part to the “AI Factory“of both. In Xataka | Anthropic already had Claude writing code. Now he has put it in the laboratories

Anthropic already had Claude writing code. Now he has put it in the laboratories

by

Anthropic had already placed Claude in one of the most everyday and valuable tasks in the technology industry: writing code. Now he wants to take it to more delicate terrain and with potentially much greater consequences: scientific work within laboratories. The company has introduced Claude Sciencea product designed to help researchers move between literature, data, specialized tools and computing resources. Claude to science. The key to Claude Science is not only that Anthropic has added more tools to Claudebut in the type of problem it is trying to solve. In science, a huge part of the work involves jumping between databases, files, code, figures, citations, and computing resources that rarely talk to each other comfortably. The company wants to integrate all this into a specific application, available from June 30, 2026 in beta for Pro users, MaxTeam and Enterprise on macOS and Linux. A category jump. Anthropic had already begun to bring Claude closer to scientific work last fall, when it launched connectors and functions under the umbrella of Claude for Life Sciences. This helped the model to relate better to software and scientific databases, but it still had a more limited scope. What is happening now goes one step further. Anthropic seems to want science to stop being just a use case and become a product line. Verifiable work. The promise of Claude Science is not limited to helping you write or summarize. Anthropic claims it can analyze scientific literature, execute multi-step investigations, generate figures and manuscripts, and allow the researcher to refine them iteratively. The most important part is how it leaves a trace: each result includes the code, the environment, and the message history that produced it. In addition, a review agent checks quotes and calculations, and can point out untraceable numbers or figures that do not match the code that generated them. Claude Science’s ambition might sound very broad, but his first steps have a fairly recognizable accent. Anthropic has prepared it with more than 60 capabilities and connectors targeting areas such as genomics, proteomics, structural biology, computational chemistry, and single-cell analysis. The computation, within the flow. Many investigations do not stop at reading articles or generating figures: they also require carrying out heavy work on machines prepared for it. Anthropic says Claude Science can help prepare those processes on the researcher’s laptop, on a Linux machine, on an HPC access node via SSH, or with on-demand computing in Modal. The company clarifies that the system writes a plan and asks permission before accessing new resources, so that the researcher can review or revoke decisions. It also states that large or sensitive data can remain in the lab infrastructure, sending Claude only the context necessary for each step of the analysis. Anthropic accompanies the launch with examples. Manifold Bio, dedicated to the design of drugs aimed at specific tissues, used Claude Science to propose targets in its experiments, evaluating surface expression, cell trafficking and safety according to the company’s own criteria. The Allen Institute used it to build a computational review template with about 20 custom skills, capable of reading thousands of articles and organizing findings into an evidence base. And at UCSF, epidemiologist Stephen Francis says the tool sped up glioma analysis to about one-tenth the time before, with results independently validated by his group. Images | Anthropic In Xataka | South Korea has a plan to dominate in memory chips and robotics. One of a billion dollars

Anthropic CEO repeats what Ballmer said 25 years ago when calling Linux “a cancer”

by

In June 2001, Steve Ballmer, who had barely been CEO of Microsoft for a year and a half, granted an interview to the Chicago Sun-Times newspaper. During the course of it, he would make a historic statement by saying that “Linux is a cancer”. The curious thing is that 25 years later the CEO of Anthropic, Dario Amodei, made very similar statements when talking about how “Open Source AI is becoming a danger.” Both then and now, the reason that provoked these statements was none other than the fear that the Open Source philosophy would end up triumphing in the world. And if history teaches us anything—and perhaps Amodei should have foreseen it—it is that precisely what Ballmer did was not weaken Linux, but rather make it stronger than ever. That may also be what Dario Amodei ends up achieving. Amodei’s statements They were actually produced three years ago.. He made them in a speech before the US Senate Judiciary Committee in July 2023, but at that time they went somewhat unnoticed because at that time the most advanced AI models in the world were still very limited, and the situation for open models like Llama 3 was even worse. Linux was not dangerous per se. AI models of open weights, neither Three years later things have changed radically. The open models that several Chinese startups and technology companies have managed to develop are already very close to the impressive frontier models of Anthropic, OpenAI or Google, and Amodei’s prediction now becomes much more relevant. But it does so at a time when its Mythos and Fable 5 models have had a lot of problems precisely for being “dangerous.” Mythos Preview first and Mythos 5 now are only available for a small group of entities and companies due to its potential to find cybersecurity vulnerabilities. AND Fables 5which was a “layered” version of Mythos ended up being vetoed by the US government three days after going on the market. only yesterday its deployment was reinstatedbut it has done so with more restrictions to use it: if the model detects any dangerous intention, it is deactivated so that the user switches to using Opus 4.8. And while the US tries to put doors in the field with the excuse of national security, China does not even bat an eyelid. Chinese companies have not stopped launching more and better models of open weights, and We have the last and most splendid example in GLM-5.2the Zhipu.ai (Z.ai) model that is surprising everyone and everyone. Its creators already warned when launching it that its performance in various benchmarks is at the level of Claude Opus 5.5 or GPT-5.5. But independent analyzes in the field of cybersecurity they claim that GLM-5.2 is “as dangerous” as Opus 4.8 also in terms of cybersecurity. This points to a disturbing future for the US: that China will have models as powerful as Mythos in the short term. Jie Tang, CEO of Z.ai, agreed with that perspective: Elon Musk anticipated that these models would arrive in the first quarter of 2027, and Tang indicated that “it won’t take that long.” The real problem is not that Chinese companies develop open weight models with capabilities similar to those of Mythos. That will inevitably come, but as happened with Linux and Open Source software, The danger is that these models displace commercial software and threaten the dominant position of Anthropic and OpenAI. That’s what Ballmer feared 25 years ago, but what he seemed to point to with that FUD statement never happened. What happened was precisely what he would never have imagined: that Microsoft ended up “appropriating” Linux and Open Source solutions by integrating them into its cloud infrastructure, Azure. Today Linux virtual machines represent 61.8% of all those in Azure: this operating system has become an even more important option than Windows on that platform. It is no coincidence: the presence of Linux and Open Source platforms in the server market is absolutely dominant (about 90% globally), and the adoption of these solutions by Microsoft has been total. Not only in server environments, be careful: the Windows Subsystem for Linux (WSL) layer of Windows operating systems has been a crucial attraction for users and especially developers for years. The company made its definitive move in that section when he bought GitHub in 2018and he hasn’t looked back since. The analogy with the Anthropic (or OpenAI) situation is inevitable. Linux then threatened Microsoft’s position, and open AI models threaten that of Anthropic or OpenAI. The question here is not whether those AI models developed by Chinese companies can be dangerous: Mythos, Fable 5 and GPT-5.5/5.6 have already shown that they can be. The question is who they are for. For the world… or for the companies trying to become the de facto monopolies of this industry? Linux, after all, wasn’t a cancer. Ballmer was not right. It’s very likely that Amodei doesn’t have it either. Image | World Bank Photo Collection | Wikimedia Commons In Xataka | For decades, Linux has earned a reputation as a “shielded” operating system. Until now

Anthropic has moved ahead of OpenAI in its race to go public. This is very bad news for Sam Altman

by

Anthropic confirmed on Monday which has formally registered its application for its long-awaited IPO. The operation may become the largest in the history of its type, and reminds us of another singular moment. In August 1995, Netscape went public and marked the beginning of the era of the Internet and dotcom fever. That turned out to be a bubble, but “good”. The question is if it will be repeated what happened then. The original Netscape moment. When Netscape went public, the company had only been on the market for 16 months and had not made a profit in all that time. It didn’t matter. The shares went on the market on August 9, 1995 with an initial price of $28. On its first day of trading, the value skyrocketed quicklyreaching a high of $75 before closing at $58.25. In December of that year it would reach its maximum value, $171 per share. The rest, as they say, it’s history. Netscape’s IPO sent the Nasdaq technology index soaring… until the dot-com bubble hit in 2000. Source: Reuters. Anthropic could break all records. Anthropic’s spectacular growth in recent months has made the company in the pretty girl of the AI ​​sector. The recent investment round has raised its valuation to $965 billionan incredible figure considering that the company is barely five years old. It has also overtaken OpenAI, whose valuation It is currently around $850 billion.. Both were moving to go public this year, but Anthropic has gone ahead again, something that at first glance seems like another victory against its main rival. What Netscape taught us. The explosion of Netscape in 1995 gave rise to fierce competition: companies promising gold and moro did not stop appearing, and the dotcom bubble grew. Too many companies managed to attract investment without a clear business plan and the situation ended up leading to the bursting of the bubble. A few companies survived and managed to become the great giants of today’s technology. good bubbles. That bubble could be described as “good” because although many companies failed, those that remained and those that were created later ended up leading this revolution called the internet. For many, the AI ​​bubble exists, but it is similar to the dotcom bubble in that: many companies could disappear if it bursts, but the final result, they say, will be positive for the evolution of our planet. But Anthropic is very different from Netscape. Although these IPOs present certain analogies, the situation of these companies is very different. Netscape suffered greatly to monetize its software and would end up in the hands of AOL in 1999 when its stage was closing. Anthropic has shown that its approach to businesses works, and in fact this past quarter it surprised by achieving profits (with small print) when everyone expected losses. And still, total uncertainty. Anthropic’s projection—like that of OpenAI—is spectacular on paper, but we are talking about companies that in recent years have not stopped burning money to achieve the most powerful models on the market. All technology companies have been devoured by the AI ​​fever, but today the only ones who win (a lot) money are those that provide components for AI infrastructure. Milestone. The bet is that this infrastructure will be necessary because we will all use AI models on a massive scale, but it is not at all clear that this expectation will be met. It may not, but Anthropic’s IPO will certainly mark a milestone in the dizzying growth of this segment. And victory for Amodei. This year we will likely see three historic IPOs. SpaceX seems to be the first in breaking records, but both Anthropic and OpenAI follow in their footsteps. That the company led by Dario Amodei has formally confirmed its preparation for that exit is a symbolic victory against its great rival, Sam Altman, who is also planning the IPO of OpenAI. In recent months Anthropic has managed to turn the tables, and has gone from being the pursuer to the leader of a race that certainly is not over yet. Image | Wikimedia In Xataka | Anthropic is one step away from being worth as much as Samsung. And what the market is buying is not Claude

Anthropic just surpassed OpenAI as the world’s most valuable AI startup

by

Anthropic is no longer the eternal second fiddle. The company that was always in the shadow of OpenAI has become the main protagonist of this segment in recent months. Its growth is so spectacular that in its latest round of financing it has managed to surpass OpenAI’s valuation. It is an extraordinary milestone, especially for one reason: both hope to go public before the end of the year, and here Anthropic has the upper hand (again). Overtaking on the right. The company founded by the Amodei brothers has raised a colossal financing round of 65 billion dollarsand with it Anthropic’s valuation becomes 965,000 million post money. It is a financial achievement that suddenly destroys OpenAI’s valuation, which is currently stuck at $730 billion. This latest round comes just three months after Anthropic will raise 30,000 million of dollars, cccadadasdsas in an agreement that placed its valuation at 350,000 million dollars. The growth is simply amazing. Anthropic is the coolest company. The valuation reflects a compelling reality: Anthropic is (much) more fashionable than OpenAI. The company has taken great advantage of recent controversies to increase its popularity, and its brand image has been greatly reinforced because it is the company that everyone is talking about. What happened to the Pentagon first and what has happened with the encyclical Magnificent Humanitas of the Pope then they show it. And the one with the best models (seems) to have. OpenAI seemed to be ahead in the AI ​​race with models leading the way. That changed with the arrival of Claude Code and Claude Opus 4.5. Since then, Anthropic’s advances have been striking, and although the differences are small, the popular perception is that Claude Opus is now the model that leads in performance. This has just been confirmed in benchmarks with the recent release of Claude Opus 4.8but above all with Claude Mythos Previewthe model that has been put the world of cybersecurity upside down. They already make money. A few days ago, surprising news leaked: Anthropic could close the second quarter of the year with an operating profit of 559 million dollars. He would make money when the rest of his rivals lose a lot. The projected annual turnover has managed to exceed $47 billion this month, five times more than the amount estimated at the beginning of the year. The reason: the overwhelming success of Anthropic models in companies. That’s where the money isand the company has known how to 1) detect and 2) take advantage of it before anyone else. Memory manufacturers enter the round. The financing round is led by venture capital firms such as Greenoaks, Sequoia, Altimeter and Dragoneer, but this time there are other protagonists. These are the semiconductor firms Samsung, Micron and SK Hynixwho have also participated and who have taken advantage of their current privileged position to also bet on the success of Anthropic. It’s a win-win: they bet on the current winning horse, and Anthropic manages to strengthen relationships with the companies that right now they control one of the big bottlenecks of the AI ​​industry: memory chips. The IPO is imminent. This surprise meteoric intensifies the pressure on OpenAI and further encourages (if that was possible) that other race, which is the IPO of both these two companies and SpaceX. We are in a year that will be remembered for three stratospheric IPOs, but these latest achievements by Anthropic have made the company led by Dario Amodei now the main protagonist in the technology segment. Image | Fortune Brainstorm Tech In Xataka | The surprise of the new Claude Opus 4.8 is not that it is (a little) better. The surprise is the “I only know that I know nothing”

Anthropic is about to achieve something that seemed impossible for a large AI company: make money

by

In a data leak published by The Wall Street Journalthe artificial intelligence laboratory founded by the Amodei brothers has informed its investors that it will close the second quarter of 2026 with revenues 130% higher than those of the first quarter of the year. It is a colossal achievement that also achieves something unusual for these companies: they will have an operating profit of 559 million dollars. They earn more than they spend. According to these data, the company will reach $10.9 billion compared to $4.8 billion in the first quarter. Its quarterly growth rate already exceeds Zoom during the pandemic or those that Google and Facebook had before their stock market increases. It is quite a breath of fresh air for an industry accused of being a gigantic bubble. The rivals, fatal. While Anthropic gives the big surprise, the rest of the competitors are still in a good financial situation. For example, OpenAI confessed to its investors that does not expect to see benefits until 2030. It didn’t work out well for xAI either, which carries losses of 6.5 billion due to investments in data centers. How did they achieve it?. To achieve this milestone, Anthropic has differentiated its strategy from the beginning. It has focused mainly on companies that pay for the intensive use of its agentic tools (Claude Code) and its APIs (Claude Opus/Sonnet 4.7). It also uses chips from manufacturers such as Google and Amazon, and has managed to optimize its spending in the cloud. It is therefore more focused and it is more efficient than its rivals, and that has had a clear effect on its balance sheet. Mythos as reputational success. In recent months Anthropic has fought several political and media battles and seems to have emerged victorious from all of them. Have Pentagon attempt rejected By controlling how its AI models were used was a clear boost to that brand image. But also the launch of its Mythos model It has been especially striking because although it is not publicly accessible, it does not stop giving headlines that seem to confirm that what Anthropic said (“it is so good that we better not release it”) was true. But. Although the figures are promising, there are nuances in these estimates. Not being a public company, Anthropic uses accounting methods that benefit it in this forecast. For example, it includes as direct revenue the sales of its models through its partners, such as AWS or Google Cloud, something that OpenAI does not do. In addition, it excludes stock compensation for its employees and these results do not guarantee that this profitability will be maintained throughout the year. We will see more quarters in red. The profit achieved would be extraordinary for many companies, but it is pocket change for Anhtropic. The company recently committed to spending $15 billion in SpaceX computing capacity using Colossus clusters. At the moment everything indicates that these benefits will be temporary and the company will return to red numbers. And yet, its evolution is currently more positive than that of OpenAI, against which it has not stopped winning battles for some time. In Xataka | Nvidia’s financial results are simply dizzying. And it still hasn’t sold a single chip in China

Log In

Forgot password?

Forgot password?

Enter your account data and we will send you a link to reset your password.

Your password reset link appears to be invalid or expired.

Log in

Privacy Policy

Add to Collection

No Collections

Here you'll find all collections you've created before.