ComputeLabs Research

AI & Compute Infrastructure — October 18, 2026

Edition of · 17 stories

Investors urged AI-cloud provider Nscale to reduce its $30 billion IPO valuation target to $16–19 billion; discussions remain unresolved.

The valuation remains under negotiation. Reports citing people familiar with the discussions said NVIDIA-backed data-center developer Nscale sought an approximately $30 billion initial public offering (IPO) valuation. Some prospective investors argued that $16–19 billion would attract subscriptions more readily, citing developments including the failed listing plans of Firmus, a separate Australian AI data-center company.

Read the full story · 3 sources →

Runjian subsidiary Wuxiang Yungu signed a $200 million annual-minimum AI inference-services contract with an unnamed overseas customer, starting October 19.

The announcement concerns a signed services contract. Runjian’s wholly owned subsidiary Wuxiang Yungu signed an “AI Token Cross-Border Sales Service Contract” with an overseas customer identified only as “Company X,” not as the social-media platform X. The stated term runs from October 19, 2026, through October 18, 2027.

Read the full story · 2 sources →

Anthropic pledged 5% of its shares to a proposed public-benefit fund, subject to shareholder approval.

The initial equity allocation remains conditional. Anthropic said it pledged 5% of its shares to the proposed fund, subject to shareholder approval. The more detailed account specifies that each 50% increase in valuation relative to an established baseline could trigger an additional 2% allocation at the board’s discretion, rather than an automatic transfer.

Read the full story · 3 sources →

Samsung plans to manufacture base dies for NVIDIA’s customized high-bandwidth memory (HBM) using TSMC’s 3-nanometer process.

The manufacturing arrangement is reported, not directly confirmed in the supplied primary materials. The Seoul Economic Daily, as summarized on Telegram, cited unnamed industry sources saying Samsung accepted NVIDIA’s request after an internal evaluation. The report describes this as Samsung’s first collaboration with TSMC at the HBM manufacturing stage.

Read the full story · 2 sources →

Fusion developer Taiyi Nuclear Energy unveiled a Wuxi stellarator research-and-validation project with planned investment exceeding RMB2.5 billion.

The project was presented on October 17. Taiyi Nuclear Energy held its controlled-fusion stellarator project launch at the Wuxi International Conference Center. The research-and-validation base is located in Wuxi’s high-tech district and was described in local reporting as Jiangsu’s first controlled-fusion project.

Read the full story · 2 sources →

RFdiffusion3, a generative molecular-design system, designs proteins and nucleic acids at atomic resolution, generating protein structures approximately 10 times faster than its predecessor.

The research team includes David Baker. Science and Technology Daily, citing the latest issue of Science, reported that the University of Washington professor and 2024 Nobel Prize in Chemistry laureate participated in developing RFdiffusion3, also called RFD3. The report describes it as a generative AI system for direct atomic-level molecular design.

Read the full story · 1 source →

Cadence and University of Notre Dame researchers published an AI-accelerated iterative solver for thermal modeling of heterogeneous 2.5D/3D chip packages.

The publication focuses on package-level thermal analysis. Semiconductor Engineering reported that researchers from Cadence Design Systems and the University of Notre Dame published “A Composable AI-Accelerated Iterative Solver for 3D-IC Thermal Modeling.” The work addresses heterogeneous 2.5D and 3D integrated-circuit packages.

Read the full story · 1 source →

All ComputeLabs Research editions