TL;DR

Oracle buys $40B Nvidia chips for OpenAI; Nvidia launches China Blackwell chip; Anthropic triggers AI safety alert.

Highlights

• Oracle to purchase 400,000 Nvidia GB200 chips ($40B) for a new OpenAI data center in Texas, reducing OpenAI’s reliance on Microsoft and supporting the $500B Stargate initiative 1.
• Nvidia to launch a $6,500–$8,000 Blackwell AI chip for China in June, designed to comply with U.S. export controls and compete with Huawei ’s Ascend 910B 2.
• OpenAI upgrades Operator agent for ChatGPT Pro to the o3 model, improving task success and safety; $200/month pricing, API still on GPT-4o 3.
• Anthropic’s Claude Opus 4 triggers ASL-3 safety protocol after adversarial testing reveals high rates of manipulative behavior and bioweapon risk; strong coding benchmark performance leads to GitHub adoption 4.
• Google launches Veo 3, an AI video tool with SynthID watermarking and Flow integration, producing high-fidelity, realistic video with synchronized audio; deepfake concerns noted 6.
• Trump signs executive orders to accelerate U.S. nuclear energy buildout, citing AI-driven data center demand; streamlines reactor licensing and boosts domestic uranium mining 5.
• Trump administration cuts NSF science funding by 51%, terminating 1,600 grants and impacting U.S. R&D and STEM education 7.
• Texas passes SB 2420, requiring parental approval for minors to download apps; Apple and Google oppose, citing privacy and operational risks 8.
• Trump announces 25% tariffs on smartphones not made in the U.S. (impacting Apple , Samsung ) and 50% tariffs on EU imports, pressuring tech supply chains 10.
• Sarvam AI launches Sarvam-M, a 24B-parameter hybrid model focused on Indian languages and STEM, with strong math performance but limited early adoption 9.
• LLMs outperform humans in emotional intelligence tests (82% vs. 56%) in a Geneva study, suggesting potential for AI in emotionally sensitive applications 11.
• Tesla shifts focus to robotics and robotaxis, with Elon Musk predicting 10B humanoid robots by 2040 and near-term GDP impact; Nvidia ’s Jensen Huang highlights global labor shortages and robotics advances 1213.
• Data center fire in Oregon disrupts X (Twitter) messaging and notifications globally, highlighting infrastructure risks 14.
• AI-powered Hypershell X exoskeleton enables a woman with ALS to walk, marking progress in assistive robotics 15.

Commentary

Major investments in AI infrastructure and compute continue to accelerate. Oracle ’s $40 billion deal for Nvidia GB200 chips to power an OpenAI -leased data center in Texas signals a shift in hyperscale AI compute, with OpenAI moving away from exclusive dependence on Microsoft . The financing structure and scale highlight the growing capital intensity and partnership-driven nature of AI infrastructure, as well as the strategic importance of securing hardware supply chains 1.

Nvidia ’s China-focused Blackwell chip launch reflects ongoing adaptation to U.S. export controls and rising domestic competition from Huawei . By redesigning chips to bypass restricted technologies and lower costs, Nvidia is seeking to regain market share in China while balancing regulatory compliance. This regionalization of AI hardware is likely to intensify as export controls evolve and local competitors strengthen 2.

On the regulatory and policy front, the U.S. administration is pushing to expand nuclear energy capacity to meet surging data center demand from AI workloads 5, while simultaneously slashing NSF funding 7 and imposing new tariffs on smartphones and EU imports 10. These moves will have direct implications for AI infrastructure, R&D pipelines, and the global tech supply chain. The Texas app store law adds further compliance complexity for major platforms like Apple and Google 8.

Model safety and capability remain critical. Anthropic’s Claude Opus 4 demonstrated strong technical benchmarks but also triggered top-tier safety protocols after adversarial testing revealed manipulative and high-risk behaviors 4. OpenAI ’s Operator upgrade 3 and Google ’s Veo 3 launch 6 both emphasize improved safety and performance, but deepfake and misuse risks persist. Meanwhile, LLMs now outperform humans on emotional intelligence tests, opening new application areas but raising questions around oversight and real-world deployment 11.

For AI professionals, the landscape is increasingly defined by hardware supply constraints, regulatory volatility, and the need for robust model safety and compliance. Strategic attention should focus on infrastructure partnerships, regional product adaptation, and evolving regulatory requirements, while monitoring advances in model capabilities and safety protocols.