5 carefully selected reads across AI, business, and investing.
This week, the AI industry is centered on the massive hardware requirements of next-generation physical AI and the emergence of open-weight models. As NVIDIA launches its new Cosmos and Nemotron-3 suites, infrastructure providers are racing to develop cooling solutions for record-breaking thermal densities. Meanwhile, the software ecosystem faces strain as autonomous agents significantly increase the volume of code generated and shipped.
NVIDIA introduced Cosmos 3, an open-weight family of omnimodal world models that unify language, image, video, and action via a Mixture-of-Transformers architecture. Alongside this, the company announced Nemotron 3 Ultra, a 550B open-weight LLM positioned as a new U.S. state-of-the-art model. The release includes a full-stack ecosystem of weights, datasets, and fine-tuning recipes to support open physical AI.
Source: Latent SpaceThe mass adoption of autonomous coding agents led to a 1,400% increase in shipped code throughout 2026, placing unprecedented pressure on GitHub’s CI/CD infrastructure. GitHub COO Kyle Daigle addressed concerns regarding system uptime and the potential for flooding repositories with low-quality AI-generated contributions. The platform is currently adapting its services to remain the stable operating layer for an increasingly automated developer workflow.
Source: Latent SpaceZutaCore secured $100 million in funding from strategic investors including Samsung and Mitsubishi Electric to advance its two-phase liquid cooling technology. This investment highlights the industry's shift away from traditional cooling as power densities for AI clusters reach critical levels. The technology uses dielectric fluid to dissipate heat from high-performance chips, reducing the risk of hardware damage from leaks.
Source: Data Center RichnessCoolIT Systems has developed a new single-phase coldplate capable of cooling 15 kilowatts per chip, nearly four times the capacity of models released in early 2025. This advancement underscores the industry's long-term reliance on liquid cooling as future GPU platforms approach rack densities of 600kW. The design serves as a vital benchmark for managing the extreme thermal output of post-Blackwell hardware.
Source: Data Center RichnessAnthropic is preparing for a nearly $1 trillion IPO, supported by an annualized revenue run rate of $45 billion. The company's growth is attributed to the successful product-market fit of its coding agents, which have provided a tangible path to economic value beyond simple chatbot interactions. This pivot toward autonomous software engineering represents a major shift in the utility of modern AI models.
Source: Noahpinion