TL;DR

Nvidia unveils Taiwan AI supercomputer, opens NVLink Fusion; US greenlights major UAE Nvidia chip deal.

Highlights

• Nvidia , Foxconn, TSMC , and Taiwan’s government will build a national AI supercomputer with 10,000 Blackwell GPUs, boosting Taiwan’s AI compute capacity1.
• Nvidia unveiled NVLink Fusion, enabling third-party CPUs/accelerators to directly connect with Nvidia GPUs; early partners include Fujitsu , Qualcomm , MediaTek, Marvell, and Alchip1.
• Nvidia launched the GB300 AI platform (Q3 2025), DGX Spark personal supercomputer, RTX Pro Server, and Omniverse Server, targeting both enterprise and developer AI workloads4.
• Nvidia is expanding in Asia with a new Taipei HQ (“Constellation”) and a Shanghai R&D center, maintaining a presence in China despite U.S. export controls1.
• The U.S. reversed chip export restrictions to the Gulf, approving a $200B US-UAE deal including up to 500,000 Nvidia AI chips annually; lawmakers cite national security concerns5.
• Nvidia ’s CEO denied AI chip diversion to China and confirmed no downgraded Hopper H20 chip for China; alternative designs are in development6.
• Qualcomm will develop custom data center CPUs designed for tight integration with Nvidia GPUs and software7.
• Apple will let EU users set third-party voice assistants by default, responding to Digital Markets Act requirements; Siri AI upgrades are delayed, and Apple is exploring Google Gemini integration810.
• U.S. officials are scrutinizing Apple ’s potential AI partnership with Alibaba in China due to data privacy and national security risks9.
• China launched 12 satellites for its “Star Computing” AI constellation, aiming for 2,800 satellites and 1,000 POPS of in-orbit AI compute11.
• Chinese startup DeepSeek’s R1 model matched top U.S. AI models at a fraction of the cost, sparking debate on AI development efficiency and valuations12.
• OpenAI launched Codex, an AI coding agent for ChatGPT, automating software tasks via cloud and GitHub integration; Google I/O and Microsoft Build will focus on AI agents14.
• Intel introduced Arc Pro B50/B60 GPUs and AI accelerators for professional workstations at Computex, intensifying AI hardware competition15.
• xAI’s Grok chatbot faced criticism for spreading misinformation, raising concerns over AI output controls and transparency13.

Commentary

Nvidia ’s Computex announcements underscore its push to remain central to global AI infrastructure, especially in Asia. The Taiwan supercomputer project, new GB300 platform, and NVLink Fusion’s open architecture all reinforce Nvidia’s strategy to anchor both sovereign and commercial AI buildouts14. By enabling third-party chip integration and expanding regional operations, Nvidia aims to secure its hardware and ecosystem leadership despite rising competition and ongoing U.S.-China trade restrictions12.

The U.S. policy shift allowing large-scale Nvidia chip exports to the UAE—after rescinding prior restrictions—opens a significant new market but also intensifies scrutiny over technology transfer and national security, particularly as China accelerates its own sovereign AI initiatives5. Nvidia ’s CEO addressed compliance, denying chip diversion to China and confirming alternative product plans for that market6.

China’s AI sector continues to advance rapidly. DeepSeek’s cost-efficient R1 model challenges assumptions about the capital required for frontier AI, while the Star Computing satellite constellation points to alternative, space-based compute architectures1112. These moves could pressure U.S. incumbents on both innovation and cost fronts.

On the software side, OpenAI’s Codex and upcoming launches from Google and Microsoft highlight the ongoing shift toward AI agents and automation in software development14. Apple ’s regulatory-driven changes in the EU, delayed Siri upgrades, and potential partnerships with Google and Alibaba reflect the challenges incumbents face in both compliance and foundational AI capabilities8910. The xAI Grok incident further spotlights the need for robust controls and transparency in AI outputs13.