News linked to both this project and an event.
To address HBM memory supply shortages, NVIDIA is considering reducing the specifications of the next-generation Rubin Ultra GPU. According to sources cited by The Information, NVIDIA has internally tested at least three versions of the Rubin Ultra GPU with lower HBM configurations over the past few weeks, requiring HBM capacity lower than the initially announced specifications. Possible downgrade paths include scaling down from HBM4E to HBM4, and from 12Hi HBM to 8Hi HBM. This move aims to advance product launch plans under supply constraints.
Western Digital Chief Product Officer Ahmed Shihab published a long article pointing out that as AI infrastructure expands rapidly, the core competition in the storage field should not be simply reduced to a contest between Flash and Hard Disk Drives (HDD), but lies in whether an AI storage architecture with long-term economic scalability can be built. The AI industry is currently facing a key question: whether the storage architecture chosen this year can support future data scale growth to the PB level or even the EB level. Many AI infrastructure designs do not fail due to insufficient performance, but fall into cost dilemmas after data scale expands. Ahmed Shihab added that Flash and HDD are not in a competitive relationship, but are complementary technologies for different workloads. High-performance scenarios, such as model weights, GPU spillover, KV cache, etc., require low-latency Flash support; while long-term storage needs such as training datasets, logs, checkpoints, compliance records, and large-scale historical data are more suitable for adopting HDDs with cost advantages. Storage architecture in the AI era will be more layered, rather than relying on a single storage medium. "Flash handles performance at critical moments, HDD handles data lifecycle. The direction of future AI storage development is not 'Flash replacing HDD', but precise layering based on different data lifecycles and business requirements." "True infrastructure is not about pursuing dazzling performance, but a reliable foundation capable of supporting long-term AI growth." US stock market trends show, Western Digital
Odaily News: AI research startup Mirendil has entered into a multi-year partnership agreement with Google Cloud to secure large-scale computing resources in support of its "Self-Improving AI" research and development. Under the agreement, Mirendil will gain access to TPU and NVIDIA GPU computing resources provided by Google Cloud, as well as managed AI training clusters, to develop AI systems capable of continuously optimizing their own capabilities. It is reported that Mirendil is focused on advancing "Recursive Self-Improvement" AI, in which AI systems enhance their own performance through iterative refinement, self-learning, and optimization. This direction is also a research area of interest among some of the top AI laboratories today.Benham Neyshabur, co-founder and CEO of Mirendil, revealed that the total value of the agreement exceeds $100 million, roughly equivalent to half of the $1 billion valuation seed funding round the company completed at the end of June. (TechCrunch)
Odaily News AMD, the semiconductor giant, announced the launch of its enterprise-grade AI programming platform, AMD Instinct Coder. The platform combines AMD chips, Supermicro servers, and Spectro Cloud software, aiming to help enterprises deploy AI coding assistants locally, reduce the cost of cloud-based AI models, and protect code and data security.AMD stated that Instinct Coder is an "out-of-the-box" end-to-end AI development platform, integrating AMD EPYC processors, AMD Instinct GPUs, Supermicro AI servers, Spectro Cloud PaletteAI Inference Launchpad software, and the AMD-optimized GLM-5.2 model. It can be used for software development scenarios such as code generation, application modernization, automated testing, and code review.AMD said that compared to relying on cutting-edge cloud-based AI models, Instinct Coder can help enterprises reduce total cost of ownership (TCO) by up to 70%, with the fastest payback period shortened to 6 months.AMD noted that more and more enterprises are looking to leverage AI to improve development efficiency, but face two major challenges: on one hand, the cost of invoking top-tier cloud models continues to rise; on the other hand, entrusting enterprise source code, intellectual property, and sensitive data to third-party services poses security and compliance risks.Through a local deployment model, Instinct Coder allows enterprises to maintain control over their data and code while providing more predictable infrastructure costs. The platform supports development tools such as Claude Code, OpenAI Codex, Visual Studio Code, and Cursor, with each node supporting up to 50 users (30 concurrent users).Additionally, the PaletteAI Inference Launchpad provided by Spectro Cloud enables AI workload management, model routing, request auditing, and cost monitoring, and supports invoking external models such as Anthropic, OpenAI, Google, or xAI when needed.AMD stated that Instinct Coder aims to help enterprises break free from the high costs of cloud-based AI services, accelerate AI-driven software development processes while ensuring data security and autonomous control.
Odaily News: Bitdeer announced that its subsidiary Tydal Data Center AS has signed a 16-year data center hosting and services agreement with Volta Tydal AS, which will provide 121 MW of IT load capacity (approximately 133 MW total power) at the Tydal AI/HPC campus in Norway, all to be deployed with NVIDIA GPUs to serve a leading AI laboratory. The total expected contract payments during the base term are approximately $4.7 billion, with an 8-year renewal option attached, bringing the potential total contract value after renewal to approximately $8 billion. (Stocktitan)
Odaily News: SK Hynix and SanDisk have unveiled the first standard specification for High Bandwidth Flash (HBF), a next-generation storage technology based on NAND flash. SK Hynix presented the specification at FMS 2026, held from August 4 to 6 in Santa Clara, California, USA.HBF stacks NAND flash vertically in a manner similar to High Bandwidth Memory (HBM) to enhance capacity and data transfer speeds. The specification defines two stacked configurations—8-layer and 16-layer NAND dies—with a maximum capacity of 512GB and bandwidth offered in three tiers, supporting 0.4TB to 3.0TB per second.HBF connects to processors using the industry-standard UCIe interface, enabling compatibility with different processor types such as GPUs and CPUs. The specification has been released through the Open Compute Project (OCP), and the HBF Consortium currently includes Google and AI semiconductor company Tenstorrent.SK Hynix will also showcase its in-development 10th-generation (V10) 375-layer 4D NAND wafers and products for the first time at this event, with plans to begin mass production of high-performance, high-capacity enterprise SSDs (eSSD) utilizing this technology early next year.
KIOXIA officially announced its first GP series SSD, the KIOXIA GP1, supporting GPU direct access to high-speed flash memory. This product achieves ultimate random read performance of 100M IOPS. KIOXIA stated that the KIOXIA GP1 series aims to support emerging AI storage architectures that incorporate high-speed flash media into video memory systems. This method enables AI systems to access larger datasets at a cost far lower than adding HBM, while improving GPU utilization.
HIVE Executive Chairman Frank Holmes stated that the company's cluster of 504 NVIDIA B200 GPUs at Bell Canada's Manitoba AI fabric generates approximately $2.90 per GPU per hour in revenue; by comparison, HIVE's Bitcoin mining equipment generates approximately $0.12 per hour. HIVE's total revenue for fiscal year 2026 reached $298 million, up 158% year-over-year; digital currency revenue from Bitcoin mining grew 164%. During the same period, average hash rate stood at 22.2 EH/s, up 290% year-over-year, accounting for approximately 3% of the Bitcoin network's total hash rate, and the company mined 2,885 BTC. HIVE's BUZZ HPC division, which houses its AI and high-performance computing business, generated revenue of $19.5 million, up 94% from $10 million in the prior fiscal year. The company also signed GPU cloud agreements worth approximately $220 million with Bell and AI company Cohere, and raised $75 million through a note issuance to fund AI infrastructure expansion. HIVE is building a 320-megawatt AI data center in the Greater Toronto Area, with plans to eventually house more than 100,000 GPUs. The company stated that if the facility becomes fully operational in the second half of 2027, it could generate approximately $360 million in annualized recurring revenue.
According to TechFlow Research, Morgan Stanley released a research report on July 27, quantifying for the first time the Incremental Return on Invested Capital (ROIC) of Generative AI investments. The report constructed three estimation frameworks: the ROIC for hyperscale cloud service providers' GPU leasing business is approximately 31%, the ROIC for proprietary infrastructure model API business is approximately 46%, and the ROIC for third-party compute API business is approximately 25%. Under base case assumptions, a single 1 gigawatt (GW) data center is configured with approximately 410,000 NVIDIA GB300 GPUs, with a utilization rate of 75% and an hourly leasing price of $8.5. The combined capital expenditure of the three major cloud giants is expected to exceed $1.4 trillion. Morgan Stanley maintains an Overweight rating on Microsoft, Amazon, Meta, and Google. The report points out that as AI moves from the training phase to the inference phase, demand for GPU compute power will continue to grow. Providers with self-built compute infrastructure will achieve considerable profits by leveraging their pricing power in an ecosystem where compute is scarce. If Morgan Stanley's calculations hold true, the hundreds of billions of dollars in AI capital expenditure will shift from being perceived as "costs" to "growth assets".
据韩联社报道，韩国国家 AI 计算中心将于 8月 3 日在全南光州海南市溔라시도企业城市举行开工典礼，正式启动建设。该项目由三星 SDS 联合体（成员包括三星 SDS、NAVER Cloud、三星物产、Kakao、三星电子、KT 等）承建，总投资 2.4 万亿韩元，规划用地面积约 4.9 万平方米。中心计划 2028 年前完成 1.5 万张 GPU 部署，2030 年前进一步扩展至 5 万张，供电规模最终将达 80MW。全南光州特别市预计该项目将带动 6.4 万亿韩元经济效益及 1.95 万个就业岗位。
Axe Compute Inc. (Nasdaq: AGPU) today announced a new five-year contract with a customer valued at over $1.5 billion to deploy a large-scale dedicated AI infrastructure cluster in the United States based on the NVIDIA Blackwell architecture. Secured through Axe Compute's Build program, the contract will provide over 9,200 NVIDIA Blackwell B300 GPUs to construct a dedicated cluster designed, deployed, owned, and operated by Axe Compute.Combined with previously announced Build agreements, this new contract brings the total value of contracts signed by Axe Compute in 2026 to over $3 billion. Axe Compute expects to receive over $534 million in customer prepayments related to this agreement and previously announced contracts within the next 30 days. These payments are expected to cover most of the associated GPU and infrastructure capital expenditures.The Axe Build program is expected to start generating monthly revenue this quarter, with additional clusters becoming operational in the fourth quarter of 2026. It is anticipated that this agreement, along with previously announced contracts, will bring the annualized run rate to over $696 million after deployment is complete, nearly double the $385 million run rate the company reported earlier this month.It is reported that Axe Compute Inc. is an artificial intelligence infrastructure platform based on a new cloud architecture.
: Julian Schrittwieser, a technical staff member at Anthropic, recently posted on social media, joking about the shift in attitude some tech giants have shown towards "open source" in recent years. He expressed anticipation for Nvidia and Microsoft to further open up their core technologies, stating: "It's great to see Jensen Huang become a supporter of open source. Looking forward to the open-source release of CUDA and GPU drivers. Also looking forward to Satya Nadella supporting open source, hoping to see Windows and Microsoft Office open-sourced in the future."Julian Schrittwieser also responded to external speculations that he opposes open weight models, indicating that this interpretation is inaccurate. He believes that open models can actually play an important role in many scenarios. However, Schrittwieser pointed out that what is interesting is that some companies, which have long been very cautious or even opposed to open source in the past, are now rapidly shifting towards supporting open ecosystems.In recent years, with the intensification of competition in artificial intelligence, open-source models, open weights, and developer ecosystems have become critical areas of competition in the tech industry. Companies including Nvidia and Microsoft have continuously emphasized their support for open ecosystems. However, the definition of "open" remains controversial within the industry—some companies open up model interfaces, toolchains, or ecosystems, while their core commercial assets remain closed source.Julian Schrittwieser's remarks have also sparked discussion, focusing on whether the "selective open sourcing" promoted by tech giants represents genuine openness or a new ecological competitive strategy.
Axe Compute (NASDAQ: AGPU) announced that the company has signed new customer contracts totaling over $1.3 billion in the United States and Europe, exceeding the full-year 2026 signing target of $1 billion ahead of schedule. The relevant contracts were secured through the Axe Compute Build program, under which the company will design, deploy, hold, and operate dedicated AI infrastructure for customers, who can independently select GPU types, deployment regions, and infrastructure configurations.
: Bitcoin mining company Bitdeer Technologies Group mined 990 BTC in June, representing a 7.5% increase from 921 BTC in May and a 388% surge compared to 203 BTC in the same period last year. As of the end of June, its proprietary mining hashrate reached 73 EH/s, up from 70.2 EH/s in May and 16.5 EH/s in the same period last year. Bitdeer's proprietary mining machine count increased by 12,000 units to 243,000 units in June, while operating 56,000 self-owned mining machines under joint mining arrangements at third-party data centers. The related hashrate increased by 59% month-over-month to 15.9 EH/s. The company held 150 BTC at the end of June, down from 171 BTC at the end of May. Bitdeer stated that its AI Cloud annualized revenue run rate increased from $69 million in May to approximately $76 million, with GPU utilization rising from 90% to 95% and externally subscribed GPUs increasing from 3,305 units to 3,517 units. This run rate is calculated by annualizing the daily revenue from contractual GPU orders still active at the end of the month and does not represent recognized annual revenue or revenue projections. Bitdeer has completed customer delivery for a five-year GPU cloud contract, with the deployment including two NVIDIA GB300 NVL72 clusters. The company has also signed a 10-year lease for a data center in Johor Bahru, Malaysia, providing 21.7 megawatts of information technology capacity. Scheduled for delivery in the first quarter of 2027, the facility will support 128 NVIDIA GB300 NVL72 systems.
Odaily, July 21 - Microsoft and French AI startup Mistral AI announced an expansion of their strategic partnership on July 21, signing a multi-billion dollar agreement centered on European AI infrastructure to enhance AI computing capabilities in the region. Under the agreement, Microsoft will leverage Mistral's expanded European GPU infrastructure to support its cloud computing and AI services. This infrastructure will be based on thousands of Nvidia Vera Rubin GPUs.In terms of products, the Mistral Medium 3.5 and OCR 4 models are now available on the Microsoft Foundry platform, and Mistral Medium 3.5 has also been integrated into Microsoft Copilot Studio for agent applications, document processing workflows, and industry-customized workflows. Additionally, the two parties will expand AI deployment options through Azure and Azure Local, supporting cloud, cloud-connected, and fully offline environments, targeting industries with high data compliance requirements such as finance, healthcare, and manufacturing.
Nasdaq-listed Bitcoin mining company Bitdeer released its unaudited production and operations update for June 2026. The company mined 990 Bitcoins in June, a 388% increase year-over-year; self-mining hash rate reached 73 EH/s, with total managed hash rate at 86.1 EH/s and 243,000 self-owned mining machines. Its Sealminer manufacturing facility in Nevada has broken ground, with completion expected by the end of 2026.On the AI cloud business front, monthly ARR rose to $76 million, with GPU utilization at 95%, and delivery of the NVIDIA GB300 NVL72 cluster has been completed. Bitdeer's total global controllable power capacity has reached 3.02 GW, with multiple crypto data centers simultaneously advancing plans to transition into AI computing centers. (Globenewswire)
AMD has officially launched its first rack-scale AI system for artificial intelligence, Helios. This is seen as AMD's key product to rival Nvidia's Grace Blackwell and Vera Rubin AI systems. It has been adopted by enterprise customers including Microsoft, Meta, OpenAI, and Oracle, as AMD seeks to mount a more direct challenge to Nvidia in the AI infrastructure market.Microsoft announced it will deploy the AMD Helios system in its Azure data centers, becoming the latest customer to adopt the platform. AMD stated that Helios is expected to begin shipping to customers later this year and will be used to support frontier AI model inference, Azure AI services, and enterprise-level AI applications. Data from market research firm Futurum Group shows that Nvidia currently holds over 95% of the data center GPU market share, while AMD holds approximately 4.5%. Analysts believe that if Helios deployment goes smoothly, AMD has the potential to capture 20%-25% of the market share in the future, corresponding to a potential market space worth hundreds of billions of dollars. (CNBC)
Kimi posted on platform X, stating that the market response to K3 has far exceeded expectations since its launch, with user demand in the past 48 hours approaching the upper limit of existing GPU computing capacity. To ensure the user experience for existing subscribers, new subscription services have been temporarily suspended. Current subscribers will not be affected, and the company is working to expand computing resources as quickly as possible, gradually resuming new user subscriptions in batches.Additionally, the future membership system will be split into two more targeted products: "Kimi Membership," designed for Kimi's web interface, app, and office scenarios, and "Kimi Code Membership," focused on programming workflows, to achieve more precise resource allocation and improve service stability.
: The South Korean government plans to invest approximately 4 trillion won next year to procure 10,000 of NVIDIA's latest Vera Rubin GPUs, aiming to support the expansion of public-facing services such as "AI for All" and the construction of a national-level AI infrastructure. The Ministry of Science and ICT plans to extend the "Project for Strengthening the Foundation for Utilizing AI Computing Resources" and configure all GPUs introduced next year as Vera Rubin. This project entrusts private companies, such as cloud service providers operating data centers, with the entire process of GPU procurement, construction, and service provision, funded by the government budget. Over the past two years, through this project and other means, the South Korean government has secured over 30,000 GPUs, providing them to both the public and private sectors. Compared to the previous generation Blackwell architecture, the Vera Rubin offers up to a 6-fold improvement in training performance and over an 8-fold improvement in inference performance. A source from the Ministry of Science and ICT stated that the specific GPU procurement plan and budget size have not yet been finalized. Industry insiders mentioned that current data center space is insufficient, making it challenging to secure large-scale new resources by next year. The government is currently evaluating various alternative plans.
analyst KawzInvests stated that Moonshot AI's upcoming Kimi K3 could become a significant event in the open-source AI space, and the infrastructure demand behind it may drive growth for AI cloud service platforms. Kimi K3 has approximately 2.8 trillion parameters, making it one of the ultra-large-scale open-source models. According to Moonshot's official evaluation, the model's performance is only slightly behind frontier models like Claude Fable 5 and GPT 5.6 Sol, and the full model weights are expected to be released on July 27.KawzInvests pointed out that a model of this scale cannot run on an ordinary laptop or even a single server; users need a computing cluster composed of a large number of GPUs to complete model loading and inference. When top-tier open-source models are made available for free, the biggest beneficiaries might not be ordinary users, but rather platforms that offer model hosting and inference services. For example, $DOCN (DigitalOcean) already supports serverless inference services for models like Kimi K2.6. Developers do not need to deploy hardware; they can call the model via API and pay per Token. Additionally, the platform hosts over 70 models and covers GPU leasing, model fine-tuning, and AI Agent development tools.As more large-scale open-source models are released, developers' demand for low-barrier AI infrastructure will continue to increase. Model hosting, inference services, and GPU cloud platforms may become key beneficiaries in the open-source AI wave.