TL;DR

Nvidia resumes China AI chip sales; Pentagon, AWS expand agentic AI; $90B AI infra push in Pennsylvania.

Highlights

• U.S. reverses April ban, allowing Nvidia to resume H20 AI chip sales to China; Nvidia expects rapid licensing and recovery of China revenue 1.
• California’s amended SB 53 would require major AI firms to disclose safety protocols, file incident reports, and adds whistleblower protections; includes plans for a public HPC cloud 2.
• Pentagon awards up to $200M each to OpenAI, Google , Anthropic, and xAI for agentic AI tools, expanding “frontier AI” adoption in defense 3.
• xAI launches Grok 4, now leading OpenRouter usage; faces renewed scrutiny over political bias and transparency 4.
• Thinking Machines Lab raises $2B at $12B valuation for open-source, collaborative multimodal AI; Andreessen Horowitz, Nvidia , AMD among backers 5.
• Trump and CEOs announce $90B in private-sector commitments for AI data centers and energy infrastructure in Pennsylvania; federal AI Action Plan due July 23 6.
• ASML beats Q2 earnings but warns of 2026 growth uncertainty due to macro and tariff risks; Chinese orders remain strong 7.
• Google signs $3B hydropower deal with Brookfield , plans $25B data center expansion to meet AI workload demand 8.
• AWS launches Bedrock AgentCore for enterprise-scale AI agents, adds new model customization and vector storage features 10.
• OpenAI’s ChatGPT suffers second global outage in a month, raising reliability concerns for business users 11.
• OpenAI developing ChatGPT checkout feature with Shopify , adding e-commerce commissions as a new revenue stream 12.
• Google expands AI summaries to Discover, raising publisher concerns over traffic, and addresses prompt-injection flaw in Gmail’s Gemini summaries 13.
• OpenEvidence raises $210M at $3.5B valuation; its DeepConsult agent now serves 430,000 U.S. physicians 14.
• SoftBank’s Masayoshi Son outlines vision for self-replicating AI agents; Altman highlights need for new infrastructure and lower AI costs 15.
• WeTransfer revises terms after backlash over perpetual license for AI training, clarifies no user files will be used 9.

Commentary

U.S. policy shifts are again shaping the AI hardware landscape: Nvidia ’s ability to resume H20 chip sales to China reopens a key market, but advanced chip restrictions remain, reflecting a strategy to keep Chinese developers dependent on U.S. technology 1. For Nvidia and its partners, this move releases billions in previously blocked orders and signals tactical flexibility in export controls. Meanwhile, ASML ’s strong Q2 results are tempered by caution on 2026 growth, as customers delay orders amid U.S.–EU tariff uncertainty, despite robust AI-driven demand and a steady Chinese backlog 7.

On the enterprise and government front, agentic AI is gaining traction. The Pentagon’s $800M in new contracts with leading AI firms and AWS’s Bedrock AgentCore launch show both defense and commercial sectors moving toward autonomous, production-grade AI workflows 310. xAI’s rapid Grok 4 release and surging API usage, despite bias concerns, further illustrate the pace of agent deployment 4. Google ’s and AWS’s major investments in clean energy and data centers underscore the acute power demands of scaling AI workloads, with Pennsylvania emerging as a focal point for infrastructure buildout 68.

Venture funding remains strong for ambitious AI startups. Thinking Machines’ $2B seed round and OpenEvidence’s $210M raise reflect sustained investor appetite for open-source, multimodal, and sector-specific AI solutions 514. OpenAI’s planned ChatGPT checkout feature with Shopify signals a push to monetize user engagement beyond subscriptions and APIs 12, while Google ’s AI summaries expansion and related publisher pushback highlight ongoing disruption in content distribution and monetization 13.

Regulatory and trust issues remain prominent. California’s SB 53 could set a precedent for AI safety disclosures and whistleblower protections at the state level 2, while the WeTransfer episode and Google ’s security flaw highlight persistent concerns around data use, privacy, and model robustness 913. Reliability is also in focus, as evidenced by OpenAI’s second major ChatGPT outage in a month 11.