Good morning and welcome back to The AI Rundown by Lightscape Partners.

• U.S. Reverses AI Chip Ban – Nvidia and AMD can now resume AI chip sales to China, potentially unlocking billions in revenue and signaling a thaw in U.S.–China tech tensions. Shares of Nvidia, AMD, and Chinese tech firms surged on the news.

• OpenAI’s IMO Gold Medal Performance – An experimental OpenAI model solved 5 of 6 problems at the International Mathematical Olympiad, far exceeding expectations and showcasing accelerating AI reasoning capabilities with implications for future technical and financial applications.

• AWS Launches Bedrock AgentCore – Amazon unveiled a modular enterprise platform for building and scaling AI agents, already adopted by major firms across finance and healthcare, positioning AWS to capture a large share of enterprise AI deployment spend.

If you haven’t yet, please support the newsletter by subscribing!

Hardware

U.S. Reverses AI Chip Ban, Allowing Nvidia and AMD to Resume Sales to China (Link)

• Policy Shift: Nvidia and AMD will resume sales of AI chips to China after the U.S. government reverses previous restrictions, potentially adding billions in revenue for Nvidia.
  

• Trade Negotiations: The decision follows recent U.S.-China trade discussions, suggesting a broader diplomatic thaw where tech exports and rare-earth minerals were central negotiation points.
  

• Market Reaction: The announcement boosted shares of Nvidia, AMD, and major Chinese tech firms like Alibaba, signaling positive market sentiment toward easing trade tensions.
  

• Strategic Implications: The reversal represents a significant diplomatic and commercial win for Nvidia CEO Jensen Huang, who argued previous restrictions benefited Chinese rivals like Huawei.
  

Models

OpenAI claims gold-medal performance at IMO 2025 (Link)

• OpenAI's experimental AI model achieved gold-medal performance at the 2025 International Mathematical Olympiad (IMO), solving problems 1 through 5, though it couldn't solve problem 6.
  

• This performance significantly exceeded earlier forecasts, surpassing expectations that AI would only reach 30% accuracy on math benchmarks by mid-2025.
  

• The milestone highlights rapid advancements in AI mathematical reasoning, showcasing remarkable progress compared to predictions made just a few years earlier.
  

Enterprise AI Applications

AWS launches Bedrock AgentCore (Link)

• AWS launched Bedrock AgentCore, a comprehensive platform enabling enterprises to build, deploy, and manage AI agents securely and at scale.
  

• AgentCore supports interoperability, working seamlessly with open-source frameworks like CrewAI, LangChain, LlamaIndex, and industry protocols such as MCP and Google’s A2A.
  

• Major companies, including Box, Itaú Unibanco, Innovaccer, and Epsilon, are already deploying production-grade AI agent applications across finance, healthcare, marketing, and content management.
  

• AgentCore is available in preview, priced on a consumption-based model with modular services, allowing enterprises flexible usage without upfront commitments.
  

Product Launches

OpenAI launches a general purpose agent in ChatGPT (Link)

• OpenAI launched "ChatGPT agent," a new general-purpose AI that automates computer-based tasks like calendar management, creating presentations, running code, and using connected apps.
  

• The agent integrates features from prior OpenAI tools, enabling it to navigate websites, synthesize research, and access services like Gmail and GitHub via APIs.
  

• It significantly outperforms previous OpenAI models, scoring 41.6% on Humanity’s Last Exam (double prior scores) and 27.4% on FrontierMath, with tool-assisted capabilities.
  

• OpenAI introduced enhanced safety measures, such as disabling the memory feature to prevent misuse and implementing real-time monitoring to mitigate risks related to biological and chemical threats.
  

Content Creation

Netflix starts using GenAI in its shows and films (Link)

• Netflix has begun incorporating generative AI into its productions, notably in the Argentine show "El Eternauta," marking its first use of GenAI-created footage.
  

• The AI-assisted scene of a building collapsing was completed 10 times faster and at lower cost compared to traditional visual effects methods.
  

• Co-CEO Ted Sarandos emphasized AI as a powerful creative tool that expands advanced effects capabilities beyond big-budget productions.
  

• Netflix also plans to integrate generative AI into personalization, search, and interactive ads, aiming for rollout later this year.
  

YouTube's AI Content Crackdown Divides Creators, Pleases Marketers Ask ChatGPT (Link)

• YouTube updated its policies to target "inauthentic content," causing uncertainty among creators who automate content production using AI.
  

• Marketers generally welcome the changes, believing the crackdown will raise content quality and maintain advertiser confidence.
  

• The policy shift aims to clearly differentiate repetitive or low-value AI-generated content ("AI slop") from genuinely transformative content.
  

• While some creators fear demonetization or termination, most view the update positively, expecting it to drive better-quality, more original videos.
  

Data Centers + Energy

Meta and Google Ramp Up AI Infrastructure Arms Race (Link)

• Meta plans to invest hundreds of billions of dollars in massive AI "superclusters," aiming for gigawatt-scale computing power to secure a dominant position in AI, while aggressively hiring top talent.
  

• Google announced a $25 billion investment focused on AI data centers, coupled with a $3 billion upgrade of hydropower plants to sustainably power its data centers, emphasizing balanced infrastructure and clean energy.
  

• Experts note contrasting strategies: Meta prioritizes scale and compute supremacy with high execution risk, while Google blends aggressive expansion with sustainability, potentially shaping the future leadership landscape in AI.
  

Startup Funding & Valuations

Lovable Hits $1.8B Valuation With $200M Series A, Just 8 Months After Launch (Link)

• Stockholm-based AI coding platform Lovable raised a $200M Series A at a $1.8B valuation, led by Accel, just eight months after launch.
  

• The platform, which lets users create websites and apps via natural language, now has 2.3M active users, 180K paying subscribers, and $75M ARR with a lean 45-person team.
  

• Most traction comes from non-technical users building prototypes, though Lovable aims to support full-scale production apps; 10M projects have been created to date.
  

• High-profile backers and enterprise users include Klarna, HubSpot, and investors like Slack’s Stewart Butterfield and HubSpot’s Dharmesh Shah, as Lovable expands into enterprise and global markets.

Safety + Ethics

Meta Rejects EU’s Voluntary AI Code Ahead of AI Act Enforcement (Link)

• Two weeks before the EU AI Act takes effect, the European Commission introduced voluntary General-Purpose AI (GPAI) guidelines, asking developers of large AI models (10^23–10^25 FLOPs) to pledge transparency, copyright compliance, and additional safety measures.
  

• Meta refused to sign the GPAI Code, calling it an overreach that creates legal uncertainties beyond the AI Act’s requirements, and warning it could stifle AI innovation and businesses in Europe.
  

• Meta’s Llama 4 Behemoth (trained with 5e25 FLOPs) falls into the high-risk category, but Meta plans to operate outside the voluntary framework, despite previous EU fines for data privacy and antitrust violations.
  

• While the Code is optional, EU regulators warned that non-signatories must prove alternative compliance when the AI Act takes effect on August 2, facing potentially greater scrutiny from the EU’s AI Office.

OpenAI

Pentagon Awards $200M to Anthropic, Google, OpenAI, and xAI for National Security AI (Link)

• The U.S. Department of Defense awarded up to $200 million in contracts to Anthropic, Google, OpenAI, and xAI to accelerate adoption of advanced AI for national security.
  

• The funding will support the development of AI agents across multiple mission areas to enhance U.S. defense capabilities and maintain strategic advantages.
  

• xAI launched “Grok for Government,” offering AI tools to federal agencies via the GSA schedule, following controversy over its chatbot’s offensive outputs.
  

• OpenAI, already a past DoD contractor, recently launched OpenAI for Government to provide AI services to federal, state, and local government agencies.

Thank you for reading the AI Rundown by Lightscape Partners. Please send any questions, comments, or suggestions to [email protected].