Daily AI Brief

Adopted stories from the last 24 hours, judged on an absolute UTC basis (cutoff: after 2026-07-24 21:07 UTC).

Nvidia Bets Big on South Korea — $1B into Naver plus a $500B+ SK Group Partnership

Nvidia is investing $1 billion in Naver to more than triple its GAK Sejong “DSX AI factory,” expanding it from 55MW to 200MW by 2028, with Brookfield signing a non-binding term sheet for up to $9 billion. The broader partnership with SK Group exceeds $500 billion in total scale — building more than 2GW of AI data centers on the Korean peninsula and co-developing next-generation HBM memory with SK Hynix. Jensen Huang also announced an autonomous Genesis vehicle with Hyundai Motor Group, the relocation of AI researchers from California to South Korea, and a Korean large language model built with KAIST, underscoring a “sovereign AI” theme.

Reference: Bloomberg · GlobeNewswire

---

UK and US Safety Institutes Jointly Assess Moonshot’s Kimi K3 — Trails Leading US Frontier Closed Models on Cyber

The UK’s AI Security Institute (AISI) and the US’ CAISI released their first joint preliminary evaluation, finding that Moonshot AI’s open-weight Kimi K3 trails leading US frontier closed-weight models on cyber (offensive) capability. As Chinese open models surge and Washington weighs restrictions on Chinese models, the assessment signals that government bodies are now directly benchmarking the national-security implications of frontier models.

Reference: AI Security Institute (via Techmeme)

---

Hands-On with OpenAI’s “AI Keypad” — Fun for Some Coders, Puzzling for Most

TechCrunch went hands-on with OpenAI’s new hardware, an “AI keypad.” The device, which pairs a physical keypad with AI, is intriguing for specific workflows such as coding but leaves its purpose somewhat mystifying for everyday users. It shows OpenAI extending its experiments beyond software into physical developer interfaces.

Reference: TechCrunch

---

AMD’s “Advancing AI 2026” — Can It Break the CUDA Moat?

Semiconductor analysis outlet SemiAnalysis broke down AMD’s “Advancing AI 2026” event. Key points included agentic kernel generation, improvements in software quality, and the production ramp of the MI455X (Helios). The central question is how far AMD can narrow the software advantage — the “moat” — of Nvidia’s CUDA ecosystem, with remaining challenges such as unstable internal development clusters also flagged.

Reference: SemiAnalysis

---

Today’s Takeaways

• Nvidia’s large “sovereign AI” investment in South Korea (Naver, SK, Hyundai, KAIST) is the day’s biggest story, as AI infrastructure competition scales from companies to nations.
• The UK-US joint cyber evaluation of Kimi K3 and AMD’s pursuit of CUDA represent two defining currents: “AI security” and “AI hardware competition.”