Latest news with #DeepSeek-V3
Miami Herald
4 days ago
- Business
- Miami Herald
Atlas Cloud Launches High-Efficiency AI Inference Platform, Outperforming DeepSeek
Developed with SGLang, Atlas Inference surpasses leading AI companies in throughput and cost, running DeepSeek V3 & R1 faster than DeepSeek themselves. NEW YORK CITY, NEW YORK / ACCESS Newswire / May 28, 2025 / Atlas Cloud, the all-in-one AI competency center for training and deploying AI models, today announced the launch of Atlas Inference, an AI inference platform that dramatically reduces GPU and server requirements, enabling faster, more cost-effective deployment of large language models (LLMs). Atlas Inference, co-developed with SGLang, an AI inference engine, maximizes GPU efficiency by processing more tokens faster and with less hardware. When comparing DeepSeek's published performance results, Atlas Inference's 12-node H100 cluster outperformed DeepSeek's reference implementation of their DeepSeek-V3 model while using two-thirds of the servers. Atlas' platform reduces infrastructure requirements and operational costs while addressing hardware costs, which represent up to 80% of AI operational expenses. "We built Atlas Inference to fundamentally break down the economics of AI deployment," said Jerry Tang, Atlas CEO. "Our platform's ability to process 54,500 input tokens and 22,500 output tokens per second per node means businesses can finally make high-volume LLM services profitable instead of merely break-even. I believe this will have a significant ripple effect throughout the industry. Simply put, we're surpassing industry standards set by hyperscalers by delivering superior throughput with fewer resources." Atlas Inference's performance also exceeds major players like Amazon, NVIDIA and Microsoft, delivering up to 2.1 times greater throughput using 12 nodes compared to competitors' larger setups. It maintains sub-5-second first-token latency and 100-millisecond inter-token latency with more than 10,000 concurrent sessions, ensuring a scaled, superior experience. The platform's performance is driven by four key innovations: Prefill/Decode Disaggregation: Separates compute-intensive operations from memory-bound processes to optimize efficiencyDeepExpert (DeepEP) Parallelism with Load Balancers: Ensures over 90% GPU utilizationTwo-Batch OverlapTechnology: Increases throughput by enabling larger batches and utilization of both compute and communication phases simultaneouslyDisposableTensor Memory Models: Prevents crashes during long sequences for reliable operation "This platform represents a significant leap forward for AI inference," said Yineng Zhang, Core Developer at SGLang. "What we built here may become the new standard for GPU utilization and latency management. We believe this will unlock capabilities previously out of reach for the majority of the industry regarding throughput and efficiency." Combined with a lower cost per token, linear scaling behavior, and reduced emissions compared to leading vendors, Atlas Inference provides a cost-efficient and scalable AI deployment. Atlas Inference works with standard hardware and supports custom models, giving customers complete flexibility. Teams can upload fine-tuned models and keep them isolated on dedicated GPUs, making the platform ideal for organizations requiring brand-specific voice or domain expertise. The platform is available immediately for enterprise customers and early-stage startups. About Atlas Cloud Atlas Cloud is your all-in-one AI competency center, powering leading AI teams with safe, simple, and scalable infrastructure for training and deploying models. Atlas Cloud also offers an on-demand GPU platform that delivers fast, serverless compute. Backed by Dell, HPE, and Supermicro, Atlas delivers near instant access to up to 5,000 GPUs across a global SuperCloud fabric with 99% uptime and baked-in compliance. Learn more at SOURCE: Atlas Cloud press release
Hindustan Times
08-05-2025
- Business
- Hindustan Times
Silicon Valley leader's warning on AI: Tech's new bubble about to burst
The launch of ChatGPT in November 2022 took the world by storm, marking a pivotal moment in the race for artificial intelligence and AGI. Within just five days of its release, ChatGPT reached over 1 million users, an unprecedented milestone for a tech product. Developed by OpenAI, this AI chatbot captivated users with its ability to write essays, generate code and images, answer questions and more. Since then, AI has become the most hyped buzzword in Silicon Valley and beyond. Thousands of startups have sprung up since OpenAI unveiled ChatGPT, with promises of using artificial intelligence for everything from writing code to medical and healthcare applications. But one Silicon Valley skeptic says the AI bubble is about to burst. 'I think the day of reckoning may have come – the signs are there,' Gary Marcus told The Telegraph. 'AI is reaching a point of diminishing results.' Marcus, formerly the head of AI for Uber, is well known for his critical stance on contemporary artificial intelligence methodologies, particularly deep learning and large language models (LLMs). In 2015, he co-founded Geometric Intelligence, an AI startup that was later acquired by Uber. Marcus has been a vocal critic of the prevailing AI paradigm, arguing that current models, such as ChatGPT, are fundamentally flawed. He contends that these systems excel at pattern recognition but lack genuine understanding, reasoning, and common sense. In a 2022 interview with Analytics India Magazine, Marcus had said that LLMs are not the way to get to Artificial General Intelligence (AGI). 'I was arguing that if the scaling of pure LLMs would run out, that some problems like 'hallucinations' would be very hard to solve,' Marcus told The Telegraph. LLMs like ChatGPT and DeepSeek-V3 are the best-known application of AI today. They are advanced artificial intelligence systems designed to understand and generate human-like text, capable of holding entire conversations with humans. However, complaints about LLMs 'hallucinating' or making up facts are not new. Marcus argues that real breakthroughs in AI are possible, but the industry needs to change the way it is trying to achieve them. The AI commentator believes that the era of diminishing returns has 'clearly been reached.'
Express Tribune
17-04-2025
- Business
- Express Tribune
Trump administration mulls ban on Chinese AI firm DeepSeek
Listen to article The Trump administration is reportedly considering new restrictions against Chinese artificial intelligence company DeepSeek, aiming to curb its access to American technology and consumers, The New York Times reported on Wednesday. The potential measures include banning US sales of Nvidia's advanced AI chips to DeepSeek and restricting American access to its AI services. The move comes as tensions escalate in the ongoing US-China tech rivalry, with Washington increasingly wary of Beijing's growing influence in AI. The White House on Tuesday strengthened rules limiting Nvidia's AI chip exports to China. Officials are particularly concerned about DeepSeek's rapid rise and affordability, which has disrupted the market and drawn attention from AI developers and investors globally. DeepSeek, which launched its open-source model DeepSeek-V3 for just $6 million in January, is suspected of using unlawfully obtained American intellectual property. OpenAI has alleged the Chinese firm violated its terms of use by distilling its models. A US congressional report also accuses DeepSeek of funneling American user data to the Chinese Communist Party and manipulating content to align with state narratives. According to the report, DeepSeek may have acquired up to 60,000 Nvidia chips, including 20,000 covered by US export controls, raising concerns about unauthorized transfers. Singaporean authorities recently arrested three individuals for illegally exporting Nvidia chips, allegedly on behalf of DeepSeek. Nvidia maintains it follows all US regulations and said it ships products according to legal instructions. However, the US Select Committee on the Chinese Communist Party is demanding more transparency from Nvidia about its sales across Asia. DeepSeek is also facing global scrutiny, with data regulators in Italy, France, South Korea, and Ireland investigating the company's compliance with privacy laws. The company has yet to publicly respond to the allegations.
New York Times
16-04-2025
- Business
- New York Times
Washington Takes Aim at DeepSeek and Its American Chip Supplier, Nvidia
Two months after DeepSeek, China's artificial intelligence star, rattled Washington and shook Wall Street, U.S. officials are taking steps to crack down on the Chinese start-up and its support from America's leading chip maker, Nvidia. The Trump administration this week moved to restrict Nvidia's sale of A.I. chips to China. It also is weighing penalties that would block DeepSeek from buying U.S. technology and debating barring Americans' access to its services, said three people with knowledge of the actions who spoke on the condition of anonymity. Congressional leaders are also putting pressure on Nvidia. On Wednesday, the House Select Committee on the Chinese Communist Party, which focuses on national security threats from China, opened an investigation into Nvidia's sale of chips across Asia. It is trying to assess whether the U.S. chipmaker knowingly provided DeepSeek with critical technology to develop A.I., potentially in violation of U.S. rules. It is the first Congressional investigation into Nvidia's business. It comes as the Trump administration wrestles with how to implement a Biden-era rule that limits the number of A.I. chips that companies can send to different countries. The attacks on DeepSeek and Nvidia are an outgrowth of fear in Washington that China could leapfrog the United States in A.I., which would have wide-ranging implications for national security and geopolitics. If China took the lead, it could more quickly use A.I. systems to design next-generation weapons like autonomous missiles and drones. It also could persuade other countries to use its technology for their network of A.I. systems and infrastructure, weakening U.S. influence across the world. Klon Kitchen, a senior fellow at the American Enterprise Institute who focuses on national security and technology, said that the U.S. strategy with A.I. has been to use its current lead in making A.I. chips and systems to persuade countries to ally with the United States. The competition between Washington and Beijing for technological supremacy has scrambled the semiconductor business. In the wake of the Trump administration cracking down on chip sales to China this week, Nvidia and another chip maker, Advanced Micro Devices, said they would lose billions of dollars in sales. ASML Holding, the Dutch company whose machines are essential for manufacturing the most advanced semiconductors, said on Wednesday that orders for its equipment had fallen short of expectations. On Wednesday, shares of Nvidia and A.M.D. fell more than 7 percent. ASML shares were down 5.4 percent. Nvidia didn't immediately respond to a request for comment. Speaking in March at Nvidia's annual conference, Jensen Huang, the company's chief executive, said, 'We have a fundamental obligation to run our business, obeying laws and doing our best to compete and serve customers.' In January, DeepSeek jolted the tech industry after the release of a new A.I. system called DeepSeek-V3 that it claimed to have trained for $6 million, roughly a tenth of what U.S. companies were spending. The low-cost system challenged the industry's consensus that bigger and better A.I. systems would be built by companies that spent the most on chips and data centers. Shares of Nvidia, which controls the A.I. chip market, plummeted 17 percent in a single day, reducing its market valuation by $600 billion. 'DeepSeek's claims sent everyone through the roof because it showed we're one moment away from the centerpiece of American policy being undermined,' Mr. Kitchen said. At a conference with industry executives in Washington in March, Commerce Secretary Howard Lutnick, who oversees U.S. technology controls, said the Trump administration would pursue 'a dramatic increase' in enforcement of U.S. tech restrictions on China, including imposing fines on companies. 'We have had enough of people trying to make a dollar supporting the people who seek to destroy our way of life,' Mr. Lutnick said, referring to China. Last week, Mr. Lutnick made good on his promise. The Commerce Department notified Nvidia of new restrictions on its sale of A.I. chips to China. The limits mean Nvidia will need a license to sell a chip they specially made for China, the H20. The company had developed that chip by modifying the abilities of a leading A.I. chip, the H100, to comply with the Biden administration's performance thresholds for sales to China. The crackdown on Nvidia came after White House and Commerce Department officials received briefings in recent weeks about the national security risk posed by DeepSeek. The briefings included details about the ties that some of its researchers have to the People's Liberation Army and other institutions in China sanctioned for aiding the Chinese military, a person familiar with the conversations said. The findings, which were compiled by Exiger, a data analytics firm, and reviewed by The New York Times, showed that dozens of DeepSeek's researchers have or have previously had affiliations with P.L.A. laboratories or defense-research institutes in China. They had worked with groups that included an institute that develops and tests China's nuclear weapons, universities known as the 'seven sons of national defense' and organizations barred from buying American technology by the U.S. government. It is common for private companies in China to have close ties to state-backed industries and create technology for state-funded projects. But the research indicates that DeepSeek, which claims to be a private company not linked with the state, has closer ties with the Chinese military and government than previously understood, said Kit Conklin, a senior vice president at Exiger, who wrote the report. DeepSeek did not respond to a request for comment. In February, the congressional committee on China began work on a separate report seeking to explain how DeepSeek made its technological leap. It cited estimates by SemiAnalysis, a semiconductor research firm, that DeepSeek had access to 60,000 Nvidia chips, including 20,000 that have been restricted from China. The committee said that DeepSeek and other Chinese A.I. companies may have circumvented U.S. restrictions on buying Nvidia chips by purchasing them through intermediaries in Singapore. It cited a report by Reuters in February, that Singaporean authorities arrested three people for illegally exporting advanced Nvidia chips to DeepSeek in China. In the letter opening its investigation into Nvidia, the committee asked the company to provide details about every customer from 11 Asian countries, including Malaysia and Singapore, who purchased 500 A.I. chips or more since 2020. It also asked for details about the companies who were expected to use those chips. The committee has subpoena power and expects Nvidia to provide responses within two weeks, said a committee official who spoke on the condition of anonymity. It typically spends more than four months on an investigation before generating a report on an issue and calling a hearing, which it may do with Nvidia. The investigation comes as the Trump administration evaluates a proposed government rule to provide more oversight over Nvidia's global chip sales. The Biden administration rule, which Mr. Lutnick and other Trump officials are reviewing, would dictate where A.I. chips could be shipped and require more disclosures about the customers using American technology. Tech companies have protested the rules, saying they threaten sales and U.S. technology influence around the world. In a statement, Representative John Moolenaar, a Michigan Republican and the committee's chair, said that DeepSeek's use of Nvidia's chips could be considered 'a national security failure.' 'U.S. companies should not enable a hostile regime's drive to dominate A.I. and challenge the free world,' he said. Nvidia has previously said that its record of revenue from Singapore does not 'indicate diversion to China,' and it is a result of its practice of reporting the country where purchases are made, not necessarily where chips are shipped. 'We insist that our partners comply with all applicable laws, and if we receive any information to the contrary, act accordingly,' John Rizzo, a Nvidia spokesman, said in a January statement. The congressional report also found that DeepSeek used data from OpenAI to accelerate the development of the Chinese company's A.I. system. In a memo from OpenAI to the committee, which was reviewed by The New York Times, OpenAI said DeepSeek employees had bypassed guardrails meant to prevent that kind of data harvesting. Using data from one A.I. system to build another, a technique called distillation, is common. But if a company takes data from proprietary technology, the practice could be illegal. (The New York Times has sued OpenAI, claiming copyright infringement of news content related to A.I. systems. OpenAI has denied the claims.) OpenAI called on Congress to 'safeguard U.S. technological leadership.'
Yahoo
10-04-2025
- Business
- Yahoo
Fangzhou Showcases AI Governance Breakthroughs at Guangzhou Closed-Door Summit on Vertical-Specific AI Models
GUANGZHOU, China, April 9, 2025 /PRNewswire/ -- Fangzhou Inc. ("Fangzhou" or the "Company") ( a leading Internet healthcare solution provider, emerged as the sole digital health innovator invited to the Modern Industrial System Development Summit hosted by the Guangzhou Industrial Development Research Institute on April 1, 2025. At the closed-door forum, executives from Alibaba, Baidu, academia, and government agencies discussed strategies to position Guangzhou as a national leader in domain-specific large language AI models, a key component of China's "new quality productive forces." Dr. Xie Fangmin, founder, chairman, and CEO of Fangzhou commented, "At Fangzhou, we are dedicated to advancing domain-specific AI models which are able to solve real-world healthcare challenges while maintain the highest standards of safety and efficacy." Comprehensive Safety Architecture At the summit, Fangzhou's executive unveiled the Company's groundbreaking approach to ensuring medical AI accuracy and reliability. To address critical risks such as flawed symptom assessments or false drug information, Fangzhou has engineered an innovative three-tiered safety system. The security framework leverages a "robust knowledge database + multi-model collaboration + stringent safety controls," aiming to reduce hallucinations and mitigate risks by ensuring data source integrity, optimizing model training, and validating outputs. Blueprint for Vertical AI Leadership At the summit, Fangzhou proposed a two-tiered approach to accelerating Guangzhou's development of domain-specific AI expertise. On the R&D front, Fangzhou recommended strengthening interdisciplinary talent development in medical AI, fostering high-impact use cases, and accelerating technology commercialization. For industry deployment, the strategy should focus on bolstering infrastructure, emphasizing practical use cases, and fostering ecosystems, with targeted support for specialized and innovative companies. AI-Powered Expansion and Forward-Looking Strategy Fangzhou's expansion plans are closely tied to its implementation of AI models such as DeepSeek-V3 and strategic collaborations with Tencent Health and Tencent Cloud, aimed at advancing AI-enabled healthcare solutions. The Company intends to enhance its technological strength and accelerate the commercialization of its innovations, ultimately aiming to provide better experiences for both patients and users. About Fangzhou Inc. Fangzhou Inc. ( is China's leading online chronic disease management platform, serving 49.2 million registered users and 223,000 physicians (as of December 31, 2024). The Company specializes in delivering tailored medical care and precision medicine solutions. For more information, visit About the Modern Industrial System Development Summit The Modern Industrial System aims to establish Guangzhou as a hub for strategic sectors like AI and biotech by 2025. Vertical-specific models are prioritized under China's 2024 State Council AI guidelines to counter "homogeneous competition" in generic AI. Media ContactFor further inquiries or interviews, please reach out to:Xingwei Zhao Associate Director of Public Relations Email: pr@ Disclaimer: This press release contains forward-looking statements. Actual results may differ materially from those anticipated due to various factors. Readers are cautioned not to place undue reliance on these statements. View original content to download multimedia: SOURCE Fangzhou Inc.
