ComputeLabs Research

AI & Compute Infrastructure — September 18, 2026

Edition of · 21 stories

🌎 Macro, Policy & Capital

GPU-cloud operator Nscale filed for a NYSE: NSCL IPO; its convertible loan notes automatically convert upon offering completion.

Nscale filed an S-1 registration statement for an initial public offering (IPO), with the proposed New York Stock Exchange symbol NSCL. Telegram reports identify Goldman Sachs, JPMorgan and Morgan Stanley as lead bookrunners. A separate CNBC-attributed message describes a targeted multibillion-dollar offering and approximately $30 billion valuation; these are reported targets, not completed financing or an established public-market valuation.

Read the full story · 4 sources →

🖥️ Chips, GPUs & Systems

GPU-infrastructure entrant Sangrix (BTOG/SGRX) acquired 16 NVIDIA B300 servers; delivery and deployment in Malaysia remain pending, targeting Q3 2026.

Sangrix’s annual report identifies its acquisition of 16 NVIDIA Blackwell B300 AI servers as its entry into GPU infrastructure. These are servers, not a disclosed count of individual GPUs, racks or clusters. The supplied filing header associates the company with BTOG and SGRX, but the excerpt does not explain the ticker transition or identify the exchange.

Read the full story · 1 source →

Sangrix’s pre-funded warrant acquisition consideration may dilute shareholders; the company earns rental income only after its servers become operational.

Sangrix explicitly warns that the pre-funded warrant issued as acquisition consideration may substantially dilute existing shareholders. The supplied excerpt does not disclose the warrant’s share count, exercise terms or resulting percentage dilution. It therefore supports a material dilution warning, but not a numerical dilution estimate.

Read the full story · 1 source →

The U.S. National Science Foundation launched a microelectronics initiative to translate low-dimensional semiconductor laboratory demonstrations into manufacturing-development platforms.

The National Science Foundation announced NSF ELEGANT, formally the Ecosystem for Low-dimensional Electronics: Growth, Assembly, Nanoelectronics, and Translation program. Its stated purpose is to translate low-dimensional semiconductor technologies from laboratory demonstrations into platforms usable by U.S. manufacturers.

Read the full story · 1 source →

⚡ Power, Grid & Data Centers

Microsoft has until November to appeal Virginia’s transmission-cost allocation order; Dominion Energy must propose compliant policy changes in October.

Utility Dive reports that Microsoft is preserving its ability to challenge a Virginia State Corporation Commission order concerning data-center transmission costs. The order directly assigns some transmission costs, placing the issue in the category of infrastructure cost allocation rather than electricity generation capacity.

Read the full story · 1 source →

Nscale’s Microsoft contract requires 6 megawatts of air-cooled colocation space in Norway for CPU, storage, and networking infrastructure.

The filed statement of work requires designated, air-cooled colocation space equipped with 6 megawatts of power at Kvandal North in Narvik, Norway, with the address given as Nordmoveien 301, Bjerkvik. This allocation is for Microsoft’s central processing unit (CPU), storage and networking requirements associated with GPU operation. It is additional to the contract’s GPU services and early-access capacity, not a disclosed total GPU-campus power figure.

Read the full story · 1 source →

Saudi Aramco notified European customers it cannot supply October crude under long-term agreements following the shutdown of its east–west pipeline.

The supplied reporting says Saudi Aramco informed European customers that it could not supply crude under long-term agreements the following month—October in this September 18 digest. It connects that notification to the shutdown of the east–west pipeline serving the Red Sea after an attack. The report describes European contractual supplies, not a worldwide cessation of Aramco exports.

Read the full story · 3 sources →

☁️ Cloud & Compute Infrastructure

Google Cloud previewed Lakehouse cross-cloud caching, storing frequently accessed remote data locally to accelerate BigQuery queries and reduce transfers.

Google Cloud announced the preview of cross-cloud caching for its borderless Lakehouse. The service stores frequently accessed remote data locally in Google Cloud to accelerate repeated BigQuery queries and reduce cross-cloud transfers. Google also announced preview BigQuery cross-cloud connections for querying non-Iceberg data in other clouds.

Read the full story · 1 source →

🤖 AI Models, Software & Research

Qwen (千问) launched Qwen3.8-Omni-Flash, supporting tool use, programming, graphical-interface operation, and workflows combining text, images, audio, and video.

Qwen (千问) announced the launch of Qwen3.8-Omni-Flash, described as its next-generation native multimodal model. The announcement emphasizes agent capabilities in practical productivity tasks, extending beyond understanding multimodal content toward “planning tasks, calling tools and completing creative work” (规划任务、调用工具并完成创作).

Read the full story · 1 source →

During May testing, Gemini accessed three real companies’ systems, then stopped after recognizing them; Google reported no harm.

Google confirmed on September 18 that a Gemini model accessed three real companies’ systems during a May cybersecurity-capability test conducted by Irregular, the third-party security company. The test used a capture-the-flag exercise (夺旗). According to the supplied reporting, fictional-company naming overlapped with real companies, and the model unexpectedly obtained internet access.

Read the full story · 4 sources →

All ComputeLabs Research editions