Company@X

Signal of the Day

River AI raised a staggering $1.1 billion, co-led by General Catalyst and AMP Public with backing from NVIDIA and AMD, to build a fully personal AI stack where users own the hardware, training data, and model weights. This monumental raise signals a major industry push toward private-by-design, sovereign AI architectures that challenge centralized cloud-computing monopolies.

Key Announcements

[River AI] · Source River AI announced a $1.1 billion funding round led by General Catalyst and AMP Public, with participation from NVIDIA, AMD, Y Combinator, and Temasek. The company’s training API is already live and powering workloads. This capital will accelerate the development of a completely personal AI stack designed to align deeply with individual values while giving users complete ownership of their models, data, and hardware.

[Neros Technologies] · Source Neros Technologies closed a $250M Series C funding round at a $2.5B post-money valuation, co-led by Sequoia Capital and the American Strategic Technology Fund. The funding will ramp up production at its new 250,000-square-foot Torrance factory, which currently ships 1,250 drones a week and is built to scale to 1 million units annually by 2028. Neros is expanding into a multi-product defense-tech firm, accelerating the deployable combat timelines for its Archer AI autonomous strike drone and Bandit counter-drone interceptor.

[OpenAI] · Source OpenAI released a preview of its ChatGPT desktop app for Linux, expanding its native footprint directly to developer workstations. The desktop app is available for Ubuntu (24.04 and 26.04 LTS), Debian 13, and Fedora (43 and 44). Delivered via .deb or .rpm packages for x64 or ARM64 architectures, the app allows users to integrate ChatGPT, ChatGPT Work, and Codex directly into local projects and browser workflows.

[webAI Intelligence Lab] · Source webAI Intelligence Lab open-sourced TwiL-LM3, a 3-billion parameter formal reasoning model. Despite its lightweight footprint, TwiL-LM3 outperforms OpenAI’s 120-billion parameter GPT-OSS-120B on four out of five formal logic benchmarks while running 2.6x faster on consumer hardware. Trained on proprietary, verified datasets rather than scraped web data, the model is built for edge deployment on devices ranging from iPhones to Raspberry Pis.

[NVIDIA] · Source NVIDIA introduced Nemotron 3.5 Lightning, an open 30-billion parameter Mixture of Experts (MoE) model with 3 billion active parameters. Designed specifically for high-speed, always-on AI agents, the model achieves four times the output speed of similar-sized models and supports up to a 1-million token context window. The model runs Hermes Agent by Nous Research via Microsoft Foundry, showcasing a strong partnership to streamline enterprise agent deployment.

[xAI] · Source xAI released Grok Bot in public beta, introducing autonomous AI teammates that can log into user tools and complete workflows independently. Alongside the bot release, xAI has exposed its Imagine Image 2.0 model via API and a dedicated playground. The image model supports precise editing, infographics, ads, and UI/UX mockups, expanding the platform’s multi-modal utility for developers.

[Unsloth AI] · Source Unsloth AI launched Unsloth Desktop, a new open-source desktop application that lets developers run and train models locally. The platform supports training models 2x faster with 70% less VRAM across NVIDIA, AMD, Intel, and Mac hardware setups. Key integrations allow developers to connect Claude Code and Codex to local LLMs, leveraging private web search and deep research on their own machines.

[Prodigy Research] · Source Prodigy Research launched its frontier AI quantitative finance trading research lab (YC S26). The startup trained a foundation model that beats Claude Fable and GPT-5.6 Sol at autonomous quant research and outperforms a top 10% Jane Street trader. During its live batch trading, the model yielded over 100% returns while major indices remained flat or down.

Also Noted

• [Google] (Source): Celebrated reaching 1 billion monthly active users on Gemini, with Josh Woodward revealing that 63% of users interact via voice and that 60+ new regional dialects are coming soon.
• [Kara Labs] (Source): Launched its YC S26 startup to grow ultra-high-purity diamond wafers that move heat 5x faster than copper to improve thermal management for AI chips and power-dense systems.
• [Tesla] (Source): Showcased the first Cybercab with integrated Starlink satellite connectivity, marking a strategic hardware convergence across autonomous vehicles and satellite networks.
• [Inkbox AI] (Source): Launched Agent2Agent (A2A), a protocol that gives autonomous agents their own emails, phone numbers, and iMessage identities to let them communicate and delegate directly across systems.
• [H.A.L.E.] (Source): Opened preorders for H.A.L.E. 1.0, a mobile bimanual robot starting under $10,000 for businesses and trained on 150,000 hours of real-world video.

📊 I can compile these fundraising rounds and strategic launches into a structured PDF report or a comparative table to map the capitalization and competitive landscape of these new “applied research” labs.