NVIDIA Fires Four Arrows at Computex
Ahead of COMPUTEX TAIPEI 2026 in Taiwan, NVIDIA made a barrage of major announcements at once during its GTC Taipei 2026 keynote. CEO Jensen Huang championed a "new era of Physical AI," laying out an end-to-end strategy spanning hardware, models, and developer tools.
Cosmos 3 — An Open Model for Physical AI Reasoning
Announced simultaneously on the NVIDIA Blog and Hugging Face Blog, Cosmos 3 is an open omni model specialized for Physical AI reasoning and action. It is positioned as a model that can reason with an understanding of real-world physical constraints, targeting applications such as robotics and autonomous driving.
A key characteristic is that it is offered as an open model hosted on Hugging Face, with the aim of lowering the barrier to entry for Physical AI research and development. Training Physical AI systems has traditionally required large-scale simulation environments, but by providing that foundation, Cosmos 3 could make it easier for more researchers and companies to enter the field.
RTX Spark — An AI PC with 128GB of Unified Memory
Co-developed with Microsoft, the Windows PC "RTX Spark" features a Blackwell-generation GPU and 128GB of unified memory, and is designed from the ground up for running AI agents locally.
According to the NVIDIA Blog, demand for personal AI agents is surging, with open source projects such as OpenClaw and Hermes rapidly gaining traction on GitHub. RTX Spark provides the hardware foundation for running these local AI agents smoothly. This marks NVIDIA's re-entry into the Windows PC market, putting it in direct competition with Intel and AMD.
Nemotron 3 Ultra — NVIDIA's New LLM
Nemotron 3 Ultra, whose announcement drew attention in the LocalLLaMA community, is the latest model in NVIDIA's in-house LLM series. Detailed benchmarks are still to come, but it signals that NVIDIA is getting serious about model development, not just GPUs.
Vera CPU — Anthropic, OpenAI, and SpaceX Among Early Adopters
According to Bloomberg, NVIDIA announced that Anthropic, OpenAI, and SpaceX will be the first major customers for its in-house designed CPU, Vera. Jensen Huang himself mentioned it during the Computex presentation.
For a company that has until now been GPU-centric, this is a clear statement of intent to move into the CPU space in earnest. For enterprise customers, being able to standardize on NVIDIA products for both GPUs and CPUs in AI data centers offers the benefit of simplified infrastructure management. How far NVIDIA can penetrate the server CPU market that Intel and AMD have built up over many years, however, will come down to real-world performance and cost-effectiveness.
Intel Counters — OpenVINO Physical AI Framework
As if to counter NVIDIA's Physical AI strategy, Intel announced the OpenVINO Physical AI Framework on the same day. It is an open source robotics library featuring an inference runtime optimized for Intel processors, and it will be incorporated into the integrated SDK, Robotics AI Suite.
While NVIDIA goes on the offensive with vertical integration of GPUs plus models, Intel is responding with open source and processor optimization. In the Physical AI and robotics space, the difference between the two companies' approaches is likely to split developers' choices.
Peter Thiel: "AI Threatens Technical Roles, Not Creative Thinkers"
In a Fortune interview, Peter Thiel warned that "AI threatens technical roles more than creative thinkers," sparking debate on Hacker News as well.
At a time when mainstream AI-threat narratives tend to fixate on creative jobs being taken away, Thiel's point is counterintuitive yet insightful. The argument: technical work that lends itself to standardization—such as coding and data analysis—is precisely the top target for AI-driven automation. Indeed, recent news has brought a string of examples showing how engineering roles are being transformed.
Chiba Gin Computer Service Cuts VB.NET Migration from 12.5 to 2.0 Person-Months with AI
Chiba Gin Computer Service has built an AI-driven development workflow that cut the migration effort for an existing VB.NET system from 12.5 person-months to 2.0 person-months, as reported by @IT.
Legacy system migration is a shared challenge for Japanese companies, and this is a noteworthy real-world example of AI enabling dramatic reductions in effort. It also echoes Thiel's point—but rather than "technical jobs being replaced by AI," it illustrates technical professionals using AI to achieve a leap in productivity.
A Quadriplegic Data Scientist Builds VibeETL
A story gaining traction in Reddit's LocalLLaMA community: a former data scientist who became quadriplegic developed a visual ETL tool, VibeETL, in three months. Built with Polars and React Flow, the high-speed data processing tool aims to be an alternative to Alteryx.
This example of continuing to build software despite physical limitations is instructive from two angles: the evolution of development tools, and accessibility.
---