Latest / UpNext AI / NVIDIA and Korea’s Sovereign AI Push, Anthropic’s Power Deal, and OpenAI’s Hacktron Partnership
Transcript
- Welcome to the UpNext AI podcast. It's Monday, November 9th, 2026, and here's what matters in AI today. NVIDIA and Korea’s Ministry of Science and ICT are expanding their collaboration across AI infrastructure, models, skills and security. The announcement reaches from semiconductors and data centers to the software used to serve Korean AI models. For organizations building domestic AI capacity, it brings several parts of the stack into one government-backed effort. One piece involves NVIDIA NVLink Fusion, which could give Korean chip companies a route into a broader AI infrastructure ecosystem. NVIDIA says Korean NPU and XPU companies can explore participation; it has not described that participation as complete. The company also announced a collaboration with SK Telecom and Rebellions, supported by the ministry, to advance infrastructure for Korean models and workloads using NVLink Fusion. On the model side, NVIDIA plans to work with three selected consortia and four Korean foundation-model builders through Korea’s AI for All initiative. It says its Dynamo software will help optimize and scale inference—the work of running models for users. Training and certification programs, hackathons and technical seminars are also part of the plan. The breadth is the news here. Korea and NVIDIA are linking chip development, model-serving software and workforce training in a sovereign AI initiative. Several components are collaborations or opportunities to explore, so their eventual scale remains to be seen. That infrastructure push also has a startup track. At a forum hosted with Korea’s science ministry in Seoul, NVIDIA announced an investment in Korean interconnect company Point2 Technology. It says additional investments are planned in Motif Technologies, RLWRLD and Upstage. The companies cover different needs. Point2 develops radio-frequency-based links for moving data between chips and racks in AI clusters. Motif builds foundation models for agents; RLWRLD works on robotics foundation models; and Upstage makes language models and document AI tools for businesses. NVIDIA leaders and vice minister Ryu Je-myung met with more than 20 Korean startups at the forum, spanning models, physical AI, infrastructure and semiconductors. For NVIDIA, the move puts startup financing alongside the chip and software partnerships it is building in Korea. Power is another part of the infrastructure picture. The Financial Times reports that Anthropic has struck a deal worth $3.4 billion to generate gas-fired power for data centers. The report says the AI lab is bypassing traditional suppliers to secure scarce equipment as energy demand rises. That puts energy procurement directly into a frontier lab’s expansion plans. Its significance is the size of the commitment and the route Anthropic is taking to obtain equipment. The report’s headline and summary do not establish when power will be delivered, so the deal should be read as a procurement development, not as capacity already online. For the research note, consider what happens when an AI system recommends an action in the physical world. In a preprint from earlier in the week, Yoko Hasebe proposes separating high-level AI reasoning from a deterministic controller that can reject commands when a risk threshold is crossed and switch to a safe fallback. The writeup brings that control design together with proposals for cryptographic hardware, local computing infrastructure and intellectual-property protections. Its most useful idea for operators is the separation of roles: let an AI system interpret a situation, while a dedicated control layer governs physical actuation. This is a proposed architecture spanning several technologies; it is best read as a design framework rather than a deployed safety result. ...Are you building apps with voice? Elevate your app's voice capabilities with ElevenLabs. Their API is a game changer for embedding dynamic, responsive voice interactions in your applications, providing unprecedented realism, flexibility and latency. Visit up next dot fm slash eleven to check out their latest offerings. ... A few more developments before we go. OpenAI says it is partnering with security firm Hacktron to find vulnerabilities in its systems, with particular attention to connections with third-party services. The two teams will choose testing priorities together, and OpenAI is expanding Hacktron’s access to its frontier models through Daybreak Red. A quick disclosure: we use models from OpenAI to help write the UpNext AI podcast. Developer Simon Willison has a small workflow fix for AI-written pull requests: put a short, human-approved explanation first, then collapse the longer agent-generated description using GitHub-supported details and summary tags. Reviewers can scan the purpose of a change without losing the technical record. And TechCrunch reports that Crusoe has sold its troubled Wyoming data center project to Google. The transfer illustrates a practical challenge in building large AI facilities: a project can change hands even as demand for computing capacity remains strong. Before we wrap up, a quick note: this podcast is generated with the assistance of AI and is intended for informational purposes only. All referenced articles, research, and commentary remain the property of their original authors and publishers. And that's your briefing for today. If you enjoyed this episode, share it with a friend or colleague—and follow UpNext AI on Apple Podcasts, Spotify, or wherever you get your podcasts! Full source links are in the episode notes, and we'll be back tomorrow with what's up next!