Susana Schwartz
Not his first rodeo: SoftBank CEO's move to cash out $5.8 billion in Nvidia shares for OpenAI isn't his first risky move
This isn’t the first time CEO Masayoshi Son has taken a gamble: after the .com bubble burst in 2000, Son lost an astounding $70 billion of his personal fortune, toppling him from his position as the world’s wealthiest man to near bankruptcy. He then made an extremely risky $20 million investment in fledgling internet startup Alibaba — an investment that eventually ballooned to more than $200 billion.
Yesterday, Son announced to shareholders that he’d sold nearly $6 billion of Nvidia shares, as well as $9.17 billion in T-Mobile shares, and taken a margin loan against SoftBank’s Arm Holdings stake. With this all-in OpenAI push, he is making a pivot toward the application layer of AI. And by simultaneously retaining SoftBank’s controlling stake in chip designer Arm, it seems he is striving to build his own large-scale, integrated AI ecosystem. These moves also complement SoftBank’s investment in AI architecture for 5G and AI for Ran work. As we had reported a few weeks ago, SoftBank had made a $22.5 billion commitment, contingent on OpenAI’s corporate restructuring with Microsoft. OpenAI did its part, laying the groundwork to eventually go public, and now Softbank is doing its part to move toward a total investment of approximately $34 billion into OpenAI by end of December.
After yesterday’s announcement, Nvidia stocks went down 2%, as did SoftBank’s – at times as much as 10%. For some, this announcement increased speculation about whether the AI bubble was about to pop, for others, it looked like a strategic pivot rather than a signal of collapse.
Susana Schwartz
Technology Editor
RCRTech
AI Infrastructure Top Stories
First DC in Kuwait: Ooredoo Kuwait launched a sovereign AI data center with Nvidia, equipping its local data center with H200 GPU units. This supports Kuwait’s Vision 2035 digital strategy to power national applications in healthcare, education, and finance.
Pt. 2 of ‘AI readiness in fiber networks’: Following Monday’s checklist 1-10 for AI readiness, RCR Wireless today released points 11-20, focusing on the raw optics and geographic reach that make a global fiber network smart, controllable, clever.
Edge computing in the age of AI: Joe Madden, founder and lead analyst of Mobile Experts, announced a new report and examination of telcos’ role in “Sovereign AI” or “Sovereign Edge Computing,” as nations pursue control over computing workloads.
AI-Powered Telecom Infrastructure
Supermicro, in collaboration with NVIDIA, delivers AI-powered infrastructure tailored for telcos, enhancing operational efficiency, network management, and customer experiences. Explore now
AI Today: What You Need to Know
Meta to focus on AI-driven sciences: Mark Zuckerberg announced his philanthropy work will shift in focus toward AI-driven sciences, noting researchers don’t want more lab space or headcount, but rather, ‘They just want GPUs.” His partner Priscilla Chan noted the goal is to get to the 10,000 range for GPUs.
AI infrastructure expansion in Europe: Microsoft and Google will invest more than $10 billion in AI infrastructure in Portugal and $6 billion in Germany — all part of a plan to invest about $16 billion in AI infrastructure in Europe.
Nebius GPUs for Meta: Nebius Group’s $3 billion deal with Meta takes some of the sting out of soft Q3 results. Nebius will supply AI infrastructure for Meta Platforms, with the first phase beginning next month. The GPU Services will be deployed in two tranches during December 2025 and February 2026, along with associated storage and connectivity services.
Nvidia’s next-gen Rubin AI system: Nvidia’s upcoming Vera Rubin NVL144 AI rack system features higher power GPUs and CPUs, driving its liquid cooling cost to about US$55,710, 17% higher than the previous NVL72 model.
Cold-plate cooling solution by Amazon: Amazon CEO Andy Jassy said the company has built its own solution for cooling, since it cannot wait for liquid cooling facilities to be built. Amazon’s hardware solution will be a direct-to-chip cold plate, which aims to land both liquid and air-cooled racks in the same facilities.
