🇺🇸Broadcom is discussing $60–70 billion in senior secured debt, potentially plus $30 billion in junior debt, for AI chips. #
Broadcom is reportedly negotiating with lenders over an AI-chip financing structure that remains under development. Under the proposal, a special-purpose vehicle would issue between $60 billion and $70 billion of senior secured debt, with Broadcom discussing a guarantee for part of that tranche.
The structure may also include approximately $30 billion of junior debt, bringing the total financing under discussion to as much as $100 billion. These figures describe a potential facility rather than completed or funded financing.
Blackstone and Apollo Global Management were reported to be in talks with Broadcom about participating. The companies had previously entered an “AI XPV platform” partnership in June involving $35 billion of financing for 1 gigawatt of Broadcom custom-chip infrastructure to be built and operated by Fluidstack, with the resulting compute purchased and used by Anthropic.
🇺🇸Anthropic is preparing for a potential IPO filing by August-end and may add Citigroup as a lead underwriter. #
Anthropic is preparing for a potential initial public offering filing as soon as the end of August, according to reports citing people familiar with the preparations. Chief Financial Officer Krishna Rao was reported to have led recent investor briefings, but no filing was included in the supplied source material.
Citigroup may join Morgan Stanley, Goldman Sachs and JPMorgan in the core underwriting group. The final composition of the underwriting syndicate remains subject to change, and additional banks could still be added.
Separate reports said Anthropic expected the offering to match or exceed SpaceX’s reported record IPO size. The supplied reports also attributed more than $11.5 billion of preliminary second-quarter revenue and a $65 billion annualized revenue run rate at the end of July to Anthropic; these are reported company figures rather than figures from a supplied regulatory filing.
🇦🇺SharonAI (Nasdaq: SHAZ) received customer acceptance for an initial deployment under a five-year, approximately $950 million contract. #
SharonAI Holdings disclosed in an SEC-filed press release that a global technology customer with a major Asia-Pacific presence accepted an initial AI-cloud deployment. The acceptance completed a milestone under a five-year infrastructure agreement with an initial total contract value of approximately $950 million and triggered the release of cash security held in escrow.
The accepted deployment is the first phase of the contract. SharonAI said it expects to deliver additional NVIDIA graphics processing unit clusters over the coming months and is deploying the customer’s AI-cloud solution across multiple Australian data centers.
The company expects revenue under the agreement to begin in stages during the third and fourth quarters of 2026. That revenue timing and the planned delivery of additional clusters are company expectations contained in the filed release, not completed operating results.
SharonAI describes itself as an Australian neocloud serving artificial intelligence and high-performance computing workloads. Its amended registration statement says its infrastructure includes access to NVIDIA B200 and B300 GPUs and anticipated GB300 GPUs, while also identifying NEXTDC as its primary data-center provider and stating that delays by NEXTDC could impair deployment and revenue generation.
🇺🇸Micron plans to invest $10 billion over ten years in Boise-based Micron Research Labs, focused on memory and AI research. #
Micron announced Micron Research Labs as a U.S.-based, long-horizon research institution headquartered in Boise, Idaho. The company plans to support the organization with a $10 billion investment over the next decade.
The research hub will focus on memory and artificial intelligence. Micron said it will bring together customers, academic institutions, government participants and the wider semiconductor ecosystem.
The announcement characterizes the $10 billion as a planned investment rather than capital already spent. The supplied source does not provide an annual spending schedule, construction timetable or allocation among facilities, equipment and research programs.
- Micron
- Micron Research Labs
🇭🇰Hong Kong Exchanges and Clearing (HKEX) cut weighted-voting-rights listing thresholds to HK$20 billion, or HK$6 billion for issuers exceeding HK$600 million in annual revenue. #
Hong Kong Exchanges and Clearing lowered the minimum market-capitalization threshold for companies using weighted voting rights, a structure under which some shareholders hold greater voting power than others. The standard threshold was reduced from HK$40 billion to HK$20 billion.
An alternative route now permits a minimum valuation of HK$6 billion when the issuer recorded more than HK$600 million in revenue during its latest financial year. Previously, that route required at least HK$10 billion of market capitalization and HK$1 billion of annual revenue.
The reform was described as the largest adjustment to Hong Kong’s weighted-voting-rights listing framework since 2018. The reported changes also expand eligibility to companies with innovative business models, in addition to the framework’s focus on technology and innovation issuers.
- HK$20 billion
- HK$6 billion
- HK$600 million
Waymo deployed its custom application-specific integrated circuit in its newest robotaxi; the chip delivers more than 1,000 TOPS. #
Waymo said its custom application-specific integrated circuit, or ASIC, has been installed in its newest-generation robotaxi. An ASIC is a chip designed for a defined workload rather than the broad range of tasks handled by a general-purpose processor.
The chip provides more than 1,000 trillion operations per second, or TOPS, of stated compute performance. Waymo combines it with other processors and company-designed sensors in the vehicle platform.
According to Waymo, the ASIC’s specialized accelerators extract relevant information directly from raw lidar, radar and camera streams. One identified workload is temporal denoising intended to improve perception under low-light conditions, while the combined platform supports reaction and navigation in complex urban environments.
Micron said data-center customer demand is approximately 50% above the supply volume it can commit. #
Micron Chief Executive Officer Sanjay Mehrotra said purchasing demand from data-center customers is approximately 50% greater than the volume the company can commit to supply. The comparison concerns customer requests against committed Micron supply, not a statement that Micron’s total manufacturing capacity is 50% below overall market demand.
Mehrotra said customers across Micron’s end markets would buy all of the products the company produces. He also said that operating at full production would still not satisfy the cited portion of demand.
The CEO described artificial intelligence as changing the role of memory because AI systems require more memory, higher-performing memory and lower-power memory. The source does not provide a product-level split among dynamic random-access memory, high-bandwidth memory, NAND flash or other memory categories for the 50% demand figure.
SK hynix identified Japan’s Miyagi Prefecture as a possible memory-fab location but emphasized that no investment decision has been made. #
Reports said SK hynix was considering Miyagi Prefecture for a memory-chip manufacturing facility. The proposed project was described as potentially representing the South Korean chipmaker’s first major manufacturing investment in Japan.
The reported investment could reach tens of trillions of South Korean won, and industry sources said SK Group Chairman Chey Tae-won had recently visited the region. These details came from media reports rather than an announced capital commitment.
SK hynix responded that any location with the necessary infrastructure could be considered as a candidate. The company explicitly said it had not made a decision, so Miyagi remains a potential site rather than a selected or approved fab location.
Cornelis Networks and NEC expanded their collaboration to deploy Cornelis high-performance networking across Japan and the Asia-Pacific region. #
Cornelis Networks and NEC Corporation announced an expanded collaboration covering artificial intelligence and high-performance computing infrastructure in Japan. The companies said organizations would use Cornelis high-performance networking solutions in these deployments.
The expanded arrangement also covers the broader Asia-Pacific region. It builds on work the two companies had previously conducted together in Europe.
The supplied announcement does not identify a specific customer, cluster size, network-port count, contract value or deployment date. It therefore establishes an expanded commercial and technical collaboration rather than confirming a particular completed infrastructure installation.
Duke Energy’s Carolinas resource plan targets 18.5 gigawatts of new solar capacity by 2041. #
Duke Energy’s resource plan for its North Carolina and South Carolina service territories targets 18.5 gigawatts of new solar generation by 2041. This is planned generation capacity, not current operating output or a statement that all projects have already received approval.
Duke told South Carolina regulators that its Carolinas territory continues to experience significant growth. The company also noted that North Carolina and South Carolina rank among the fastest-growing states.
The supplied source identifies solar alongside gas and nuclear resources in its coverage of the plan. It does not provide, in the supplied description, a project-by-project solar schedule or identify how much of the 18.5 gigawatts is contracted, permitted or under construction.
Federal Energy Regulatory Commission (FERC) approved Southwest Power Pool’s topology-optimization plan for reducing congestion through transmission reconfiguration. #
The Federal Energy Regulatory Commission approved a Southwest Power Pool plan to use topology optimization to reduce grid congestion. Topology optimization changes the configuration of existing transmission elements rather than referring to the construction of a new generating plant or data center.
The approach uses transmission reconfiguration to address constraints in the power network. The supplied source does not quantify expected Southwest Power Pool savings or identify the individual transmission facilities that could be reconfigured.
For comparison, Utility Dive reported that the Midcontinent Independent System Operator’s power market had saved nearly $100 million during the year using a similar topology-reconfiguration approach. That figure applies to the Midcontinent Independent System Operator, not to Southwest Power Pool’s newly approved plan.
West Texas regulators examined additional transmission as oil producers and proposed data centers drive sharply rising regional electricity demand. #
Texas grid and utility-regulatory executives appeared before the Texas House State Affairs Committee to discuss the need for additional transmission lines in West Texas. The hearing addressed rising power demand in the region as both oil and gas producers and technology companies pursue greater electricity use.
Diamondback Energy, identified as one of the Permian Basin’s largest oil producers, expects its own electricity demand to double over the next ten years even if oil output remains unchanged. Technology companies are also considering data centers in the Permian Basin, adding another category of proposed load.
Citigroup analysts warned that the region could face outages as soon as the following summer. That statement was an analyst warning, while the transmission discussion itself concerned the need for more grid infrastructure; the supplied source does not identify approved transmission projects or committed data-center megawatts.
Russia plans new nuclear plants, including small modular reactors, targeting 30 gigawatts of added generation capacity. #
Russian President Vladimir Putin said Russia plans to construct new nuclear power plants. The program includes small modular reactor projects, which use reactor units smaller than conventional large nuclear generating units.
The stated target is 30 gigawatts of additional electricity-generation capacity. The figure refers to generation capacity and not to electrical consumption, data-center load or reactor count.
The supplied report does not provide a completion deadline, site list, financing amount or split between conventional reactors and small modular reactors. It therefore describes a national target and planned projects rather than completed or fully specified capacity.
BBT Group opened a 78-kilowatt data center at Caritas Hospital, supporting several Baden-Württemberg hospitals’ digital services. #
BBT Group launched a data center at Caritas Hospital in Bad Mergentheim, Germany. The facility has a stated capacity of 78 kilowatts.
The data center will support digital services used by several hospitals operated by the group in Baden-Württemberg. The source identifies a healthcare information-technology role rather than an artificial-intelligence training cluster or public cloud region.
The supplied source does not specify server count, rack count, redundancy tier, cooling system, storage capacity or accelerator hardware. The 78-kilowatt figure should therefore be treated as the stated facility capacity without converting it into an assumed IT-load or GPU count.
- BBT Group
- Caritas Hospital
SharonAI disclosed 212 megawatts of secured AI-factory capacity, including 120 megawatts under multi-year take-or-pay agreements. #
SharonAI said it has secured 212 megawatts of AI Factory capacity. Of that total, 120 megawatts is contracted under multi-year take-or-pay agreements, which require payment under agreed contractual terms rather than representing uncontracted pipeline capacity.
The company’s accepted deployment for its global technology customer represents progress toward bringing part of this secured and contracted capacity online. SharonAI did not state in the supplied release how many megawatts the accepted initial deployment represents.
The company’s amended registration statement says it has secured up to 87 megawatts through NEXTDC and relies on NEXTDC to host most of its GPU infrastructure. It also warns that construction, financing or power-supply delays at NEXTDC could prevent SharonAI from deploying GPUs, fulfilling customer contracts or generating anticipated revenue.
SharonAI’s filing describes its platform as supporting AI and high-performance computing with graphics processing units, central processing units, storage and high-bandwidth networking. The 212-megawatt figure concerns secured AI-factory capacity and should not be read as 212 megawatts already energized or occupied by operating GPU clusters.
Google AlloyDB’s preview four-level ScaNN index supports vector-search workloads exceeding 10 billion vectors. #
Google introduced a preview four-level tree architecture for the ScaNN index in AlloyDB, its fully managed PostgreSQL-compatible database service. Google said the architecture enables the index to operate efficiently at a scale of 10 billion vectors and supports datasets above that level.
Earlier AlloyDB ScaNN tree indexes used two- or three-level configurations. At 10-billion-vector scale, Google identified index-construction compute requirements, query-traversal work and sampling memory usage as major constraints.
The four-level design uses hierarchical partitioning to reduce the number of vectors examined during a query. Google described the basic search complexity as O(N^1/2) for two levels, O(N^1/3) for three levels and O(N^1/4) for four levels.
The design also incorporates Top-K branch selection, Spilling with Orthogonality-Amplified Residuals, or SOAR, centroid adjustment and a balanced tree shape. Google said these additions are intended to preserve performance and limit recall loss while improving build efficiency and memory use.
Cloudera Data Engineering added native NVIDIA GPU acceleration for Apache Spark 4.1 using the cuDF library. #
Cloudera announced native NVIDIA graphics processing unit acceleration for Apache Spark 4.1 within Cloudera Data Engineering. The integration uses cuDF, a CUDA-X library for GPU-accelerated dataframe processing.
Cloudera described the cuDF plug-in as providing performance gains without requiring customers to rewrite application code. The integration applies GPU acceleration to Spark data-processing pipelines rather than presenting the GPU itself as a database or storage system.
The companies positioned the capability as a way to accelerate AI-ready data pipelines and lower cloud-compute spending. Those benefits are vendor claims in the announcement; the supplied source description does not include benchmark configurations, measured speedups or specific cost reductions.
- Cloudera Data Engineering
- NVIDIA GPU acceleration
- Apache Spark 4.1
iDLab aims to unify access across U.S. National Science Foundation-supported computing centers and two cloud platforms. #
The University of California, Los Angeles is leading iDLab, a national initiative intended to create a unified environment for research computing, scientific data and applications. The environment will span several national computing centers supported by the U.S. National Science Foundation and two cloud platforms.
The National Center for Supercomputing Applications is one of five institutions assisting the initiative. The project focuses on simplifying access across multiple computing environments rather than deploying a single new supercomputer or centralized cluster.
The supplied announcement does not name the two cloud platforms or enumerate all participating computing centers. It also does not provide a budget, aggregate accelerator count, storage capacity or launch timetable for the unified environment.
Google Antigravity joined eligible Gemini Enterprise subscriptions with pooled quotas, project spending caps, audit logging, and administrative controls. #
Google made Antigravity available as part of eligible Gemini Enterprise application subscriptions, including Standard, Plus and Standard Emerging Market licenses. Administrators can enable Antigravity and Android Studio for eligible users through the Gemini Enterprise administrative environment, while new extensions support development environments including Visual Studio Code.
The subscription includes shared token pools so unused prepaid quota can be consumed by teams with greater demand. Administrators can also opt into usage above pooled quotas at standard consumption-based rates, subject to monthly spending caps.
Project-level monthly budget caps are available through the billing console, while Google said per-user and team controls would roll out later in 2026. Centralized metrics cover token consumption, application programming interface calls and developer activity.
Security controls include workspace sandboxing and restrictions on browser and Model Context Protocol server access. Administrators can enable audit logging for prompts, agent responses and metadata, while Antigravity falls under Google Cloud’s standard security, compliance and data-privacy protections for Gemini Enterprise subscriptions.
- Google Antigravity
- Gemini Enterprise
Unitree Robotics is building a physical-AI self-evolution system that searches research and open-source solutions, then generates robot-control code. #
Unitree Robotics Chairman, General Manager and Chief Technology Officer Wang Xingxing described the company’s work on “physical AI robot self-evolution” (物理 AI 机器人自进化) at the 2026 World Robot Conference. He said AI models currently represent Unitree’s largest area of investment in both funding and personnel.
The proposed system uses frontier large models operating within defined rules, experience constraints and tool boundaries. The models are intended to search for frontier papers, high-quality research results and open-source solutions.
After retrieving those materials, the system generates robot-control code automatically. Wang presented this as an area Unitree is actively developing; the source does not report a commercial launch, performance benchmark or autonomous deployment in production robots.
The U.S. Department of Energy selected a Fermilab-led initiative applying AI to adaptive resonance control in particle accelerators. #
The U.S. Department of Energy selected a project led by Fermi National Accelerator Laboratory for funding through the department’s Genesis Mission. Fermilab will work with partners from other national laboratories, universities and industry.
The initiative will apply artificial intelligence and machine learning to precise, adaptive resonance control in particle accelerators. Resonance control concerns maintaining accelerator behavior under changing operating conditions rather than adding general-purpose AI-compute capacity to the power grid or a commercial data center.
The project aims to reduce accelerator operating costs by millions of dollars, according to the announcement. The supplied source does not state the funding award, project duration, participating accelerator facilities or a quantified performance target.
