ComputeLabs Research
Edition of · 18 stories
HPCwire reports that Gimlet Labs completed a $300 million financing round led by Andreessen Horowitz at a $3 billion valuation. The article describes a completed financing, rather than a proposed fundraising target or an undrawn financing facility.
Financial_Express reports that NVIDIA committed to purchasing $1.5 billion of SB Energy Class N common stock through a private placement. Separate messages state that NVIDIA’s total investment associated with the offering would be $3 billion; the private-placement amount is included within that total, not additional to it.
Financial_Express, citing The New York Times and four people familiar with the transaction’s marketing, reports that SoftBank-backed SB Energy postponed its planned September initial public offering (IPO). Those sources said investors questioned the sought valuation of $50 billion or more, and banks had difficulty finding sufficient buyers within the desired price range.
VNET (世纪互联) announced on September 21 that the strategic investment transaction had closed, according to Financial_Express. PJ Millennium I Limited and PJ Millennium II Limited together acquired 650,424,192 Class A ordinary shares from sellers, leaving the buyers with approximately 38.1% of the company’s outstanding ordinary shares.
According to a governor’s-office statement reproduced by Financial_Express, Governor Greg Abbott directed the Texas Commission on Environmental Quality (TCEQ) to suspend issuance of permits for data-center project applications. The quoted statement says no such permits would be issued until the specified audits were completed.
Huawei (华为) presented the additional UnifiedBus details at HUAWEI CONNECT 2026 in Shanghai on September 21. The announcement was delivered by Yang Chaobin, Huawei’s executive director of the board and chief executive of its Information and Communications Technology (ICT) Business Group.
Inspur Information (浪潮信息) introduced the Yuannao (元脑) SD200 Ultra at the 2026 Artificial Intelligence Computing Conference (2026 人工智能计算大会), identified as AICC2026, on September 21. Financial_Express reports that the supernode AI server uses domestic AI chips, without naming the chip supplier or model.
iPronics announced a two-year strategic technology collaboration with the Barcelona Supercomputing Center (BSC) on September 21. The announcement identifies iPronics as a developer of silicon-photonics-based optical circuit switching for AI infrastructure.
IonQ announced a strategic partnership with SDT, Inc., a Korea-based quantum-technology and software company, on September 21. Under the multiyear collaboration, IonQ will supply SDT with a Superion 256 system.
Georgia Power and Google announced a binding agreement under which Google would subscribe to capacity from nuclear uprates at the Hatch and Vogtle plants. Financial_Express reports that the agreement remains subject to approval by the Georgia Public Service Commission.
Utility Dive reports that New York regulators approved a 90-mile transmission line with a stated project cost of $3.3 billion. The developers are New York Transco and the New York Power Authority.
NVIDIA introduced DSX Ready as a program to qualify power and cooling products for AI factories. Its announcement emphasizes selecting equipment that fits the computing architecture and complete facility design.
Utility Dive reports that the Public Utility Commission of Texas adopted data-center interconnection rules that were less restrictive than the standards it proposed in March. The changes included eliminating a proposed nonrefundable interconnection fee and extending energization milestone deadlines.
Google Kubernetes Engine (GKE) Pod snapshots save a workload’s running state, including central processing unit (CPU) and GPU memory, and restore it on demand. Google reports startup-latency reductions of up to 89%, with 70-billion-parameter models loading in 37 seconds and 8-billion-parameter models in 15 seconds.
Google benchmarked its multi-cluster GKE Inference Gateway across 17,000 compute nodes in the United States and Europe. The deployment used three GKE clusters in `us-east5`, `us-west8`, and `europe-west4`, with `us-east5` serving as the configuration cluster.
NVIDIA’s developer announcement introduces TensorRT multi-device inference integration in NVIDIA Dynamo-Triton. The stated capability is model serving across multiple GPUs.
Telegram reports identify SpaceXAI as the issuer of the Grok 4.7 launch announcement and give pricing of $2 per million input tokens and $6 per million output tokens. Those are separate input and output usage rates, not a single blended price.
Researchers at Georgia Tech, NVIDIA Research, and Stanford University published a technical paper titled “BOOST: Concurrent Access to Host Memory and HBM to Accelerate LLM Inference.” Semiconductor Engineering reports that BOOST is a runtime system targeting large language model (LLM) inference.
