China's DeepSeek claims theoretical cost-profit ratio of 545% per day
BEIJING: Chinese AI startup DeepSeek on Saturday disclosed some cost and revenue data related to its hit V3 and R1 models, claiming a theoretical cost-profit ratio of up to 545% per day, though it cautioned that actual revenue would be significantly lower.
This marks the first time the Hangzhou-based company has revealed any information about its profit margins from less computationally intensive "inference" tasks, the stage after training that involves trained AI models making predictions or performing tasks, such as through chatbots.
The revelation could further rattle AI stocks outside China that plunged in January after web and app chatbots powered by its R1 and V3 models surged in popularity worldwide.
The sell-off was partly caused by DeepSeek's claims that it spent less than $6 million on chips used to train the model, much less than what U.S. rivals like OpenAI have spent.
The chips DeepSeek claims it used, Nvidia's H800, are also much less powerful than what OpenAI and other U.S. AI firms have access to, making investors question even further U.S. AI firms' pledges to spend billions of dollars on cutting-edge chips.
DeepSeek said in a GitHub post published on Saturday that assuming the cost of renting one H800 chip is $2 per hour, the total daily inference cost for its V3 and R1 models is $87,072. In contrast, the theoretical daily revenue generated by these models is $562,027, leading to a cost-profit ratio of 545%. In a year this would add up to just over $200 million in revenue.
However, the firm added that its "actual revenue is substantially lower" because the cost of using its V3 model is lower than the R1 model, only some services are monetized as web and app access remain free, and developers pay less during off-peak hours. (Reporting by Eduardo Baptista; Editing by Daren Butler)
Hashtags
Try Our AI Features
Explore what Daily8 AI can do for you:
Comments
No comments yet...
Related Articles
Sharjah 24
an hour ago
- Sharjah 24
Russia to build Kazakhstan's first nuclear power plant
Announcement from Kazakh authorities "Rosatom has been named as the leader of the international consortium for the construction of the first nuclear power plant in Kazakhstan," the former Soviet republic's nuclear power agency said in a statement. Kazakhstan's uranium production Kazakhstan, a vast resource-rich country in Central Asia, is currently the world's top uranium producer, providing 43 percent of supplies. However, it does not produce enough electricity to meet domestic consumption needs. Location of the new power plant The new power plant, whose construction was approved in a referendum in late 2024, will be built near the half-abandoned village of Ulken near Balkhash Lake. The lake, located in the southeast, is the country's second largest. Consortium bid participants China's National Nuclear Corporation, France's EDF, and South Korea's Hydro and Nuclear Power had all bid to build the plant. In their announcement on Saturday, the Kazakh authorities said that the three companies would be included in the consortium led by Rosatom but did not provide any details. Observers' perspectives Observers say the idea of the consortium is a way for authorities to maintain good relations with all the countries involved but are skeptical that it will come to fruition, believing Rosatom will end up building the plant by itself. Kazakhstan's diplomatic relations Kazakh President Kassym-Jomart Tokayev has sought to keep good relations both with former colonial power Russia and with China, which borders the country to the east and finances major infrastructure projects in the region. Project financing and next steps Rosatom has proposed financing the project, and work will now begin to thresh out the details, the statement said. The announcement comes a few days before Chinese leader Xi Jinping is due to visit Kazakhstan for a "China-Central Asia" summit. Historical context of nuclear power in Kazakhstan Kazakhstan had nuclear power plants when it was part of the Soviet Union, in addition to hosting Soviet nuclear weapons. It was also the site for Soviet nuclear testing. After the break-up of the USSR in 1991, the new country gave up its nuclear weapons, along with other ex-Soviet states Belarus and Ukraine; it decommissioned its nuclear power plants in the following years.
Arabian Post
5 hours ago
- Arabian Post
The High Stakes of the Latest U.S.-China Agreement
Dr Imran Khalid 'We made a great deal with China. We're very happy with it.' So declared President Donald Trump in his familiar tone of triumphant ambiguity on June 11, fresh off what was touted as a breakthrough agreement to restore a trade truce between the United States and China. But if history has taught us anything, it is that 'done deals' in the Trumpian lexicon tend to be either dangerously fragile or conveniently fungible. The latest accord, emerging from two days of intense talks in London, follows an alarming spiral in trade tensions that had once again threatened to upend global markets and rekindle the tit-for-tat tariff warfare that haunted the latter years of Trump's first term. According to Trump, China has committed to lifting its restrictions on the export of rare earths – materials critical to the global technology and defense sectors – while the U.S. has agreed to a calibrated rollback of punitive measures, including the threatened revocation of visas for Chinese students. ADVERTISEMENT As ever, the devil is not just in the details, but in their implementation. Much like the May Geneva agreement that this deal purports to reinforce, the London framework is conditional, tentative, and, crucially, subject to 'final approval' by both President Trump and President Xi Jinping. That qualifier alone renders the euphoria premature. Still, to be charitable, the very fact that Washington and Beijing are speaking the language of dialogue rather than confrontation is an encouraging sign. Following a phone call between the two leaders earlier this month, there appears to be a renewed willingness – albeit under duress – to keep diplomacy afloat. For a world economy battered by uncertainty, this resumption of talks is, if nothing else, a stabilizing force. Yet, Trump's boastful framing – that the U.S. walks away with a 55% tariff shield while China gets 10% – betrays a zero-sum worldview that continues to inform his trade doctrine. The truth, however, is far less tidy. Tariffs have proved to be a double-edged sword, inflicting damage on American consumers, industries, and allies as much as they have squeezed Chinese exports. The World Bank's recent downward revision of global growth forecasts points to tariffs and unpredictability as 'significant headwinds,' underlining the global costs of such brinkmanship. Beijing, for its part, has projected a more measured tone. Chinese Vice Premier He Lifeng, in remarks following the London consultations, emphasized mutual benefit, calling on the U.S. to 'honor their words with actions.' The Chinese side welcomed the 'principled consensus' as a foundation for predictability and stability in bilateral economic relations. While Beijing's rhetoric may be couched in diplomatic platitudes, it signals a strategic patience that stands in stark contrast to Trump's performative deal-making. Indeed, despite facing considerable pressure – both domestic and international -China has remained consistent in its emphasis on dialogue, reciprocity, and multilateralism. It is no secret that Beijing is playing a longer game. From its support for a multilateral trading system to its efforts in promoting South-South cooperation, China has positioned itself as a steady hand amid a turbulent global order. ADVERTISEMENT In this light, the reestablishment of a U.S.-China economic and trade consultation mechanism should be viewed as more than a temporary fix. It offers a framework through which recurring disputes can be ironed out, interests aligned, and trust slowly rebuilt. Importantly, it provides a venue for strategic communication -something sorely missing during the height of tariff wars in 2018–19. However, for this framework to bear fruit, both sides must resist the urge to revert to maximalist posturing. The United States must accept that unilateralism – whether in tariffs or technology controls – cannot substitute for a sustainable policy. Likewise, China must be prepared to meet the U.S. halfway, especially on issues of market access, intellectual property, and transparency. The elephant in the room, of course, is the technological cold war that continues to simmer beneath the surface. While rare earths and tariff percentages dominate headlines, it is the battle over semiconductors and AI supremacy that threatens to define the next phase of U.S.-China relations. Washington's decision to maintain restrictions on high-end AI chips – particularly those from Nvidia- while easing others, reveals both the complexity and the stakes involved. Beijing, not surprisingly, has responded with innovation. The resurgence of Huawei, once a poster child of American sanctions, stands as testament to China's determination to chart its own technological path. As Huawei's founder Ren Zhenfei put it bluntly this week, China may still be a step behind, but it is catching up – by stacking and clustering if necessary. In the short term, these dynamics will continue to fuel friction. But in the long term, they offer a compelling reason for structured cooperation. For neither side can afford the costs of sustained decoupling. The global economy – still reeling from inflationary shocks, supply chain disruptions, and climate-induced volatility – desperately needs the world's two largest economies to find common ground. To that end, the inclusion of Chinese students in American universities, affirmed in this deal, is more than a diplomatic gesture. It is a recognition that people-to-people ties remain a cornerstone of bilateral engagement. Academic exchanges, research collaboration, and cross-cultural education build bridges that tariffs and bans cannot destroy. They plant the seeds of mutual understanding in a landscape too often scorched by suspicion. The road ahead remains bumpy. Structural trade conflicts persist, strategic mistrust abounds, and electoral politics – particularly in the U.S. – can derail even the most promising of frameworks. But the London agreement offers a glimpse of what is possible when mutual interest outweighs mutual animosity. This development not only helps stabilize U.S.-China relations but also injects much-needed momentum into the global economy. It serves as a reminder that even amid intensifying geopolitical rivalry, there is still space – indeed, an urgent need – for pragmatic cooperation. Trump may brand it a win, but real victory lies not in tariffs or trophies, but in the hard, unglamorous work of sustained diplomacy. For now, both sides have stepped back from the precipice. The challenge will be to keep walking forward – together. Also published on Medium. Notice an issue? Arabian Post strives to deliver the most accurate and reliable information to its readers. If you believe you have identified an error or inconsistency in this article, please don't hesitate to contact our editorial team at editor[at]thearabianpost[dot]com. We are committed to promptly addressing any concerns and ensuring the highest level of journalistic integrity.
Arabian Post
5 hours ago
- Arabian Post
AI Browser Agents Mark New Era with H Company Launch
Runner H, Surfer H and Tester H, three autonomous AI agents developed by Paris-based H Company, make web-native task automation available across consumer and enterprise settings. The framework integrates advanced vision–language models that perceive browser interfaces and predict actions like clicking buttons or filling text fields, executing them via a headless browser. Runner H acts as the orchestrator. It receives natural‑language instructions, composes workflows using specialised sub‑agents and interacts with platforms such as Google Drive, Slack and Notion. In corporate pilots across France and abroad, Runner H has already begun generating revenue, even as the company offers a limited-edge free version for individual users. Surfer H is a browsing‑focused agent powered by an open‑sourced model called Holo‑1. Built on Alibaba Cloud's Qwen with H Company's enhancements, Holo‑1 records a 92.2 % success rate on the WebVoyager benchmarking suite—surpassing competitors including Google's Project Mariner and OpenAI's Operator—while reportedly cutting per‑query costs by over five‑fold. ADVERTISEMENT Tester H is tailored for web automation and testing, converting English-language prompts into scripted interactions like form‑filling or button‑clicking. The tool rides on the back of the company's 2024 acquisition of Mithril Security, integrating security‑focused mechanisms while aiming to simplify QA workflows for engineers. Benchmarks show Runner H outperforming market alternatives. On WebVoyager, its VLM–LLM pipeline yielded a 67 % task‑completion rate—exceeding Emergence AgentE at 61 % and Anthropic's Computer Use agent at 52 %. The backbone of Runner H comprises a 3‑billion‑parameter H‑VLM for interpreting GUI elements and a family of internal LLMs optimised for decision‑making and code generation. H Company introduced these agents alongside its Studio platform in March. The Studio provides a unified interface for designing, editing and running web automations, with self‑healing UI selectors and workflow version control. The platform currently operates in a private beta, yet H Company envisions expanding access to developers and broader audiences soon. Founded by former DeepMind researchers and headquartered in Paris, H Company has grown to a team of around 70, with an office in London and plans to open in the United States. Its vision is to redefine AI—from dialogue assistants to action‑driven agents capable of autonomously executing tasks across software and services. AI‑powered automation is gaining momentum. Open-source browser agents such as OpenAI's Operator, Anthropic's Claude Computer Use, and various community initiatives are rapidly evolving. Notably, H Company is contributing to that momentum by publishing research on its Vision‑Language and Large Language Models, and by releasing Holo‑1 under an open‑source licence, accelerating accessibility for developers. H Company is refining capabilities via reinforcement learning, agent debugging tools, memory, planning modules and community support. Their research emphasises modularity and cost‑efficiency—showing VLMs can outperform large generalist models in task‑grounding while operating at a fraction of the size and serve‑time cost.
