Latest news with #H200
Straits Times
2 hours ago
- Business
- Straits Times
Nvidia CEO warns that Chinese AI rivals are now ‘formidable'
US restrictions on exports to China have effectively locked Nvidia out of the country. PHOTO: REUTERS WASHINGTON – Nvidia Corp. Chief Executive Officer Jensen Huang said that Chinese AI rivals are filling the void left by the departure of US companies from that market, and their technology is becoming more powerful. 'The Chinese competitors have evolved,' he said on May 28 in an interview with Bloomberg Television. Huawei Technologies Co., a Chinese tech company blacklisted by the US government, has become 'quite formidable,' he said. US restrictions on exports to China have effectively locked Nvidia out of the country, the largest market for chips, and as a result the company expects to lose out on US$8 billion (S$10.3 billion) in sales this quarter alone. During a quarterly earnings call on May 28 , Mr Huang spent much of the time arguing that the American government should ease the curbs. Rather than keeping AI technology out of Chinese hands - the intended purpose - local companies are just finding alternatives, he said. Tencent Holdings Ltd. and other major purchasers of his products can't be blamed for turning to Huawei because they can no longer depend on US suppliers, he said. 'Like everybody else, they are doubling, quadrupling capabilities every year,' Mr Huang said. 'And the volume is increasing substantially.' Mr Huang cautioned that the gap between US products and Chinese alternatives is decreasing. Huawei's latest AI chip is similar to the performance of Nvidia's own H200 – a component that was state-of-the-art until its replacement in recent months. Under new rules, Nvidia isn't able to even ship its H20 chip to China. That component is a downgraded version of the H200. It's not possible to degrade the product's capabilities further, he told Bloomberg Technology. Nvidia is considering potential alternatives to the H20, but has no current chip planned, Mr Huang said. When it does, the company will have to seek permission from Washington. 'You cannot underestimate the importance of the China market,' Mr Huang said. 'This is the home of the world's largest population of AI researchers.' Mr Huang said he wants all the world's AI researchers and developers to be using American technology. 'Irrespective of the near-term revenue success we have had, we can't ignore the fact that the Chinese market is very important,' he said. In the interview, Mr Huang was also asked about the US revoking some Chinese student visas and how that might affect Nvidia. 'I believe the administration still feels very strongly about the incredible importance of immigration,' said Mr Huang, who was born in Taiwan. 'Look, I'm an immigrant. I know many immigrants that came to the US to build a great life, and many of us have contributed greatly to the technology industry in the US.' Mr Huang said he believed that will continue. 'We would like the brightest to come here,' he said. BLOOMBERG Join ST's Telegram channel and get the latest breaking news delivered to you.
Yahoo
8 hours ago
- Business
- Yahoo
CoreWeave vs. Nebius: Which AI Infrastructure Stock Is the Better Buy?
CoreWeave, Inc. CRWV and Nebius Group N.V. NBIS are emerging AI-focused cloud infrastructure providers positioning themselves as agile alternatives to traditional hyperscalers like Amazon Web Services and Azure, aiming to capitalize on the growing demand for AI cloud solutions. The rapid proliferation of AI is transforming the entire tech scene, and AI infrastructure has become a high-stakes battleground for tech companies. Per an IDC report, spending on AI infrastructure is expected to top $200 billion by 2028. This uptrend in spending benefits both CoreWeave and Nebius, but not equally. So, if an investor wants to make a smart buy in the AI infrastructure space, which stock stands out? Let us delve a little deeper into the companies' strengths and weaknesses to see which is the better stock pick? CoreWeave is an AI-focused hyperscaler company, and its cloud platform has been developed to scale, support, and accelerate GenAI. Businesses have been increasing spending on AI inference/fine-tuning, AI workload monitoring, and training infrastructure, including training compute, AI servers, AI storage, cloud workloads and networking. This increasing demand for AI cloud platforms, including integrated software and infrastructure, bodes well for CRWV. In the last reported quarter, revenues of $981.6 million beat the Zacks Consensus Estimate by 15.2% and jumped 420% year over year. In the first quarter earnings call, CRWV highlighted that AI is forecasted to have a global economic impact of $20 trillion by 2030, while the total addressable market is anticipated to increase to $400 billion by 2028. It recently unveiled the next generation of its CoreWeave AI object storage. This is purpose-built for training and inference, offering a production-ready, scalable solution integrated with Kubernetes. Apart from scaling capacity and getting adequate financing for infrastructure, CRWV is also expanding its go-to-market capabilities. Moreover, the buyout of the Weights and Biases acquisition has added 1,400 AI labs and enterprises as clients for CoreWeave. CoreWeave now has a data center network with 33 data centers across the United States and Europe, supported by 420 megawatts of active power. CRWV also works with NVIDIA Corporation NVDA to implement the latter's GPU technologies at scale. CoreWeave was one of the first cloud providers to deliver NVIDIA H100, H200, and GH200 clusters into production for AI workloads. The company's cloud services are also optimized for NVIDIA GB200 NVL72 rack-scale systems. Nonetheless, management's commentary surrounding higher capital expenditures and high interest expenses is likely to have unnerved investors. CRWV expects capex to be between $20 billion and $23 billion for 2025 due to accelerated investment in the platform to meet customer demand. Higher capex can be a concern if revenue does not keep up the required pace to sustain such high capital intensity. High interest expenses could weigh on profitability. In the first quarter, interest expense came in at $264 million, topping expectations. The company now guides interest expense to remain elevated, at $260-$300 million in the current quarter. CoreWeave's 77% of total revenues in 2024 came from the top two customers. This intense customer concentration is a major risk, especially if the client migrates, the revenue impact could be material. Apart from this evolving trade policy, macro uncertainty and volatility remain additional headwinds. Nebius posted 385% year-over-year revenue growth in the first quarter of 2025, driven by accelerating demand for its AI infrastructure services. NBIS is focusing on technical enhancements that increase reliability and reduce downtime to boost customer retention. In the first quarter, Nebius significantly upgraded its AI cloud infrastructure through improvements to its Slurm-based cluster. These enhancements included automatic recovery for failed nodes and proactive system health checks designed to identify issues before they impact jobs. This directly lowers downtime and boosts capacity availability. According to the company, these changes led to an estimated 5% improvement in the availability of nodes for commercial use. Nebius is making substantial investments in improving its object storage capabilities, and the upgraded storage system ensures that big data sets can be easily accessed and saved quickly during model training, directly lowering time-to-result for end users. NBIS successfully graduated multiple platform services like MLflow and JupyterLab Notebook from beta to general availability. Nebius expanded integrations with external AI platforms like Metaflow, D Stack and SkyPilot, enabling customers to migrate tools with nominal friction. Nebius is focusing on building a global footprint, with capacity in the United States, Europe, and the Middle East. It added three new regions, including a strategic data center in Israel, in the last reported quarter. Like CoreWeave, NBIS' partnership with NVDA (also an investor in the company) is another plus. Nebius will be one of the first AI cloud infrastructure platform to offer the NVIDIA Blackwell Ultra AI Factory Platform and become a launch partner for NVIDIA Dynamo. Nonetheless, the intense competition from behemoths remains a concern, along with profitability issues. Despite its exceptional top-line growth, NBIS remains unprofitable, with management reaffirming that adjusted EBITDA will be negative for the full year 2025. Though it added that adjusted EBITDA will turn positive at 'some point in the second half of 2025.' Like CoreWeave, NBIS has also raised its 2025 capital expenditure forecast to approximately $2 billion from the previous estimate of $1.5 billion, primarily due to some planned fourth-quarter spending shifting into early first quarter. In the past month, CRWV has skyrocketed 189.6% while NBIS has surged 66.9%. Image Source: Zacks Investment Research Currently, the stocks carry a Zacks Rank #3 (Hold) each. While both CoreWeave and Nebius are aggressively scaling to capture the surging demand for AI infrastructure, NBIS appears to be the more compelling investment opportunity at this stage. Despite its ongoing investments and negative EBITDA in the near term, NBIS's full-stack AI cloud platform and expanding global footprint position it well for growth. NBIS is relatively better placed than CRWV, although both appear to be on a level playing field in terms of Zacks Rank. Consequently, NBIS seems to be a better pick at the moment. You can see the complete list of today's Zacks #1 Rank (Strong Buy) stocks here. Want the latest recommendations from Zacks Investment Research? Today, you can download 7 Best Stocks for the Next 30 Days. Click to get this free report NVIDIA Corporation (NVDA) : Free Stock Analysis Report Nebius Group N.V. (NBIS) : Free Stock Analysis Report CoreWeave Inc. (CRWV) : Free Stock Analysis Report This article originally published on Zacks Investment Research ( Zacks Investment Research Error in retrieving data Sign in to access your portfolio Error in retrieving data Error in retrieving data Error in retrieving data Error in retrieving data
Yahoo
20-05-2025
- Business
- Yahoo
Sify announces Pay-Per-Use Colocation Pricing at all NVIDIA-certified AI-Ready Hyperscale Data Center Campuses across India
CHENNAI, India, May 20, 2025 (GLOBE NEWSWIRE) -- Sify Technologies Limited (NASDAQ: SIFY), India's leading Digital ICT solutions provider, announced the launch of its unique Pay-per-use model to support the burgeoning requirements for AI Cloud Services. This follows the recent expansion of Sify's portfolio of DGX-Ready Data Centers, certified for up to 130 KW/rack capacity under NVIDIA's DGX-Ready Data Center program – Sify's latest hyperscale data centers in Chennai and Noida have now been certified by NVIDIA, joining Sify's Navi Mumbai data center which was certified in 2024. Sify will offer this innovative, colocation pricing program at all three of these campus locations. The hourly pricing model is inclusive of hosting, power and infrastructure costs. By pricing its services on an hourly basis, Sify is removing the entry-cost barrier and fixed-cost infrastructure risk, enabling its GPU Cloud partners to set up and respond quickly to the growing AI market. Sify's converged ICT ecosystem will also offer an extensive array of services like Global connectivity, White Glove IT infrastructure and Managed Services. Initially, Sify will support the NVIDIA H100, H200, B200, GB200 NVL72, and GB300 NVL72 platforms, including liquid-cooled variants. Speaking on this market disruption, Sharad Agarwal, CEO, Sify Infinit Spaces Limited, the data center subsidiary of Sify Technologies Limited, said 'By investing ahead in state-of-the-art infrastructure, Sify already has the most extensive footprint of NVIDIA-certified data centers and a network connectivity with the lowest latency to hyperscale clouds. Now, by introducing colocation pricing on an hourly basis, we aim to make it much faster and easier to deploy these platforms in India to support on-demand applications. Sify's colocation partners can bring the latest NVIDIA GPUs to India, while Sify will manage all of the local infrastructure to support this dynamic and rapidly evolving market.' 'India is rapidly emerging as a pivotal player in the global AI race, with its deep talent pool, cost advantages, and rapidly advancing digital infrastructure. Sify's pay-per-use colocation model, built on NVIDIA-certified platforms, is a timely innovation that removes traditional barriers to AI adoption. By enabling global enterprises to tap into India's AI capabilities through scalable, high-performance infrastructure, Sify is well positioned to support domestic innovation and position India as a global hub for AI workloads and transformation,' quoted Rajiv Ranjan, Associate Research Director, IDC Asia Pacific. About Sify Technologies A multiple times award winner of the Golden Peacock from the Institute of Directors for Corporate Governance, Sify Technologies is India's most comprehensive ICT service & solution provider. With Cloud at the core of our solutions portfolio, Sify is focussed on the changing ICT requirements of the emerging Digital economy and the resultant demands from large, mid and small-sized businesses. Sify's infrastructure, comprising state-of-the-art Data Centers, the largest MPLS network, partnership with global technology majors and deep expertise in business transformation solutions modelled on the cloud, make it the first choice of start-ups, SMEs and even large Enterprises on the verge of a revamp. More than 10000 businesses across multiple verticals have taken advantage of our unassailable trinity of Data Centers, Networks and Digital services and conduct their business seamlessly from more than 1700 cities in India. Internationally, Sify has presence across North America, the United Kingdom and Singapore. Sify, Sify Technologies and are registered trademarks of Sify Technologies Limited. Forward Looking Statements This press release contains forward-looking statements within the meaning of Section 27A of the Securities Act of 1933, as amended, and Section 21E of the Securities Exchange Act of 1934, as amended. The forward-looking statements contained herein are subject to risks and uncertainties that could cause actual results to differ materially from those reflected in the forward-looking statements. Sify undertakes no duty to update any forward-looking statements. For a discussion of the risks associated with Sify's business, please see the discussion under the caption 'Risk Factors' in the company's Annual Report on Form 20-F/A for the year ended March 31, 2024, which has been filed with the United States Securities and Exchange Commission and is available by accessing the database maintained by the SEC at and Sify's other reports filed with the SEC. For further information, please contact: Sify Technologies LimitedPraveen KrishnaInvestor Relations & Public Relations+91 20:20 Media Nikhila Kesavan+91 Weber ShandwickLucia Domville+1-212 546-8260LDomville@
Yahoo
15-05-2025
- Business
- Yahoo
atNorth to Host 6G AI Sweden's National AI Cloud
State-of-the-art AI infrastructure to be housed at atNorth's SWE01 data center in Stockholm. STOCKHOLM, May 15, 2025 /PRNewswire/ -- atNorth, the leading Nordic colocation, high-performance computing, and artificial intelligence service provider, has announced its hosting of infrastructure that will support the development of a state of the art National AI Cloud in partnership with 6G AI Sweden. 6G AI Sweden has an agreement to acquire Nvidia's latest AI-powered chips, the H200 and Blackwell GB200, which will support the development of a state-of-the-art National AI Cloud. This strategic move will also further the business's mission to deliver cutting-edge AI innovation across various industries. Located at atNorth's SWE01 data center in Stockholm, the National AI Cloud will ensure data sovereignty under Swedish jurisdiction and will be fully compliant with GDPR. Moreover, atNorth's SWE01 site leverages renewable energy sources and heat reuse technology to minimize environmental impact, reflecting 6G AI Sweden's commitment to sustainability. "This agreement is an important milestone for establishing 6G AI Sweden as a leading provider of sovereign AI infrastructure in Sweden," said M. A. Zaman, Founder & Chairman of 6G AI Sweden AB. "By incorporating Nvidia's powerful AI technologies and hosting our infrastructure at atNorth's SWE01 data center we can develop Sweden's first National AI Cloud and empower businesses to embrace the future of AI in a responsible way". "We look forward to hosting Sweden's leading National AI Cloud at our SWE01 data center," said Anders Fryxell, Chief Sales Officer at atNorth. "This partnership reflects a shared ethos of driving innovation while minimizing environmental impact. Together, we are committed to supporting AI development in a sustainable and responsible way." This news follows the announcement of atNorth's latest heat reuse partnership with Finnish retails giant, Kesko Corporation and the launch of its 2024 Sustainability Report. The business has also recently announced the securing of land in the Municipality of Sollefteå in Långsele, Sweden, for a potential new mega site to complement its existing metro sites in the country, and to contribute to its land bank. About atNorth atNorth is a leading Nordic data center services company that offers cost-effective, scalable colocation and high-performance computing services trusted by industry-leading organizations. The business acquired leading High Performance Computing (HPC) provider, Gompute, in 2023 enabling a compelling full stack offering tailored to AI and other critical high performance workloads. With sustainability at its core, atNorth's data centers run on renewable energy resources and support circular economy principles. All atNorth sites leverage innovative design, power efficiency, and intelligent operations to provide long-term infrastructure and flexible colocation deployments. The tailor-made solutions enable businesses to calculate, simulate, train and visualize data workloads in an efficient, cost-optimized way. atNorth is headquartered in Reykjavik, Iceland and operates eight data centers in strategic locations across the Nordics, with a site to open in Ballerup, Denmark in 2025, as well as its tenth under construction in Kouvola, Finland and its eleventh site in Ølgod, Denmark. The business has also secured land for a future mega site in the Sollefteå Municipality in Sweden. For more information, visit or follow atNorth on LinkedIn or Facebook. Press Contact:Anders Fryxell, CSO atNorthE-mail: This information was brought to you by Cision The following files are available for download: Release 6G AI SwedenAB LI View original content: Error in retrieving data Sign in to access your portfolio Error in retrieving data Error in retrieving data Error in retrieving data Error in retrieving data
Geeky Gadgets
14-05-2025
- Geeky Gadgets
TPUs vs GPUs the AI Hardware Decision : Why Your Hardware Choice Matters More Than Ever
What if the key to unlocking faster, more efficient AI development wasn't just in the algorithms you write, but in the hardware you choose? For years, the debate between Google's Tensor Processing Units (TPUs) and NVIDIA's Graphics Processing Units (GPUs) has divided developers, researchers, and tech enthusiasts alike. Both are engineered for artificial intelligence, yet their architectures and capabilities diverge in ways that can make or break your AI project. With NVIDIA's GPUs dominating the market and Google's TPUs offering specialized performance for certain tasks, the choice isn't as straightforward as it seems. Understanding the nuances of these technologies is no longer optional—it's essential for anyone navigating the rapidly evolving AI landscape. In this guide, Trelis Research explore the core differences between TPUs and GPUs, from memory architecture to cost efficiency, and how these impact real-world AI workloads. You'll discover why NVIDIA's H100 and H200 GPUs are often favored for scalability and affordability, while Google's TPU V6E shines in specific low-latency scenarios. We'll also delve into critical factors like parallelization techniques, software optimization, and deployment flexibility, offering insights that could transform how you approach AI hardware decisions. By the end, you'll have a clearer picture of which technology aligns best with your goals—and why the debate between TPU and GPU is far from over. TPU vs GPU Comparison Key Hardware Differences The fundamental differences between TPUs and GPUs stem from their hardware architecture and memory capabilities. NVIDIA's H100 GPU features an impressive 80 GB of VRAM with high-bandwidth memory (HBM), while the H200 takes this further with 141 GB of VRAM and even faster memory speeds. In contrast, Google's TPU V6E is equipped with only 32 GB of VRAM, which can be a significant limitation for memory-intensive tasks. Another critical distinction lies in interconnect speeds. TPUs have slower interconnects, which can hinder their ability to efficiently manage large-scale, distributed workloads. NVIDIA GPUs, with their advanced architecture, are better suited for handling such tasks, offering greater flexibility and scalability for developers. Performance: Speed and Scalability Performance is a pivotal factor when comparing AI hardware, as it directly impacts the efficiency and scalability of workloads. TPUs and GPUs exhibit notable differences in concurrency handling, throughput, and cost efficiency: Time to First Token: TPUs excel at generating the first token quickly under low concurrency levels. However, as concurrency increases, their performance diminishes, making them less suitable for large-scale applications requiring high parallelism. TPUs excel at generating the first token quickly under low concurrency levels. However, as concurrency increases, their performance diminishes, making them less suitable for large-scale applications requiring high parallelism. Token Throughput: NVIDIA GPUs, particularly the H200, outperform TPUs in overall throughput. This makes them ideal for high-demand AI models that require consistent and large-scale processing capabilities. NVIDIA GPUs, particularly the H200, outperform TPUs in overall throughput. This makes them ideal for high-demand AI models that require consistent and large-scale processing capabilities. Cost per Token: NVIDIA GPUs are more cost-effective. The H200 offers the lowest cost per token, followed by the H100, while TPUs are comparatively more expensive for similar workloads. These performance metrics highlight the scalability and cost advantages of NVIDIA GPUs, particularly for developers managing complex AI models or large datasets. NVIDIA GPUs vs Google TPUs: Which is Best for Your AI Project? Watch this video on YouTube. Enhance your knowledge on AI development by exploring a selection of articles and guides on the subject. Parallelization: Maximizing Efficiency Parallelization techniques are essential for optimizing hardware performance, especially in AI workloads. Both TPUs and GPUs support pipeline and tensor parallelization, but their effectiveness varies significantly: Pipeline Parallelization: This technique divides model layers across multiple devices, reducing VRAM usage. However, it increases the time to first token, making it less suitable for latency-sensitive tasks where quick responses are critical. This technique divides model layers across multiple devices, reducing VRAM usage. However, it increases the time to first token, making it less suitable for latency-sensitive tasks where quick responses are critical. Tensor Parallelization: By splitting matrices within layers, tensor parallelization enhances performance but demands substantial VRAM, particularly for storing key-value (KV) caches. NVIDIA GPUs, with their larger VRAM capacities, handle this method more effectively than TPUs. The larger memory capacity of NVIDIA GPUs gives them a distinct advantage in handling parallelization techniques, allowing them to deliver better performance and efficiency for complex AI workloads. Cost Efficiency Cost is a decisive factor for many developers, and NVIDIA GPUs consistently outperform TPUs in terms of cost-efficiency. The H200 GPU offers the lowest cost per token, followed closely by the H100. While TPUs deliver strong compute performance, their higher operational costs make them less appealing for budget-conscious developers. For most AI workloads, NVIDIA GPUs strike a better balance between performance and affordability, making them the preferred choice for developers seeking cost-effective solutions without compromising on efficiency. Software Optimization The role of software optimization in hardware performance cannot be overstated. NVIDIA GPUs benefit from a robust ecosystem of open source libraries, such as VLM, which are specifically optimized for their architecture. These libraries enable better compute utilization and practical performance, allowing developers to maximize the potential of their hardware. In contrast, TPUs often face software limitations that restrict their ability to achieve peak performance. This lack of optimization reduces their effectiveness in real-world applications, further tilting the balance in favor of Nvidia GPUs for most AI development scenarios. Accessibility and Deployment Accessibility is another critical factor when choosing AI hardware. Nvidia GPUs are widely available across multiple platforms, including RunPod, AWS, and Azure, offering developers flexibility in deployment. This multi-cloud support ensures that Nvidia GPUs can be integrated into a variety of workflows and environments. On the other hand, TPUs are restricted to Google Cloud, with limited access to higher configurations like V6E-16 or V6E-32. This lack of multi-cloud compatibility makes TPUs less attractive for developers seeking scalable and versatile solutions, further limiting their appeal in competitive AI markets. Future Outlook The future of AI hardware is poised for significant advancements, and Google's upcoming TPU V7E is expected to address some of the limitations of the V6E. Improvements in VRAM capacity and interconnect speeds, coupled with enhanced software optimization, could make TPUs more competitive with NVIDIA GPUs. However, until these advancements materialize, NVIDIA's H100 and H200 GPUs remain the superior choice for most AI workloads. Their combination of high performance, cost-efficiency, and accessibility ensures they continue to lead the market, offering developers reliable and scalable solutions for their AI projects. Media Credit: Trelis Research Filed Under: AI, Guides Latest Geeky Gadgets Deals Disclosure: Some of our articles include affiliate links. If you buy something through one of these links, Geeky Gadgets may earn an affiliate commission. Learn about our Disclosure Policy.
