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

• Amazon’s talks to invest more than $10 billion in OpenAI are really a compute story in disguise. If it happens, it broadens OpenAI’s capital base and could shift negotiating leverage among hyperscalers competing to host frontier workloads.

• The US interagency review on whether to approve Nvidia H200 shipments to China signals a potential policy swing, with big implications either way. A green light could reshape accelerator supply planning and revenue, while a block keeps the focus on limiting China’s access to top-tier training and inference.

• Lovable’s $330 million raise at a $6.6 billion valuation is a clear marker that AI-assisted software building is moving past novelty. Investors are pricing in durable usage and the enterprise features, safety layers, and unit economics needed to make natural-language development a default workflow.

Stay tuned as we explore these stories and their implications for the future of AI, technology, and innovation.

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

Hardware + Software

US antitrust agencies clear Nvidia’s investment in Intel, removing a key regulatory hurdle for the chipmakers’ tie up. Link.

• The Federal Trade Commission posted a notice indicating the agencies approved the deal after a required waiting period.

• Nvidia had disclosed the investment earlier in 2025, pitching it as strategic amid surging AI infrastructure and chip demand.

• The clearance lands as regulators scrutinize AI chip competition, where Nvidia remains the dominant supplier for training and inference.

• Intel gains momentum for partnerships and financing as it funds manufacturing upgrades and refreshes its roadmap for AI workloads.

GitHub adds C++ code editing tools to Copilot Chat in public preview, expanding structured edits for large native codebases. Link.

• The update supports editing and applying changes with tooling that better matches C++ project layouts and compilation oriented workflows.

• Developers can drive targeted modifications from chat, reducing manual patching time when updating headers, implementations, and build files.

• The preview focuses on practical refactors and bug fixes, helping teams iterate faster without leaving their existing IDE routines.

• GitHub continues broadening Copilot beyond text suggestions, moving toward tool-like operations that act directly on repository code.

Models

Google launches Gemini 3 Flash, a new Gemini model positioned for fast responses while keeping frontier level capabilities. Link.

• Google frames Gemini 3 Flash as speed first, aiming to make everyday learning, planning, and building feel noticeably more responsive.

• The release targets broad use, spanning drafting, coding, and multimodal tasks where latency is a key constraint for product teams.

• Google highlights faster reasoning and stronger understanding, positioning the model as a default option for many interactive experiences.

• Gemini 3 Flash extends Google’s model lineup, giving developers and consumers a clearer tradeoff between cost, speed, and capability.

Nvidia releases the open Nemotron 3 model series, signaling a push to be a major model maker alongside its chip business. Link.

• Nemotron 3 ships in three sizes, Nano, Super, and Ultra, targeting a range of deployment constraints and agent workloads.

• Nvidia emphasizes transparency, including tools to fine tune models and greater visibility into data and training resources.

• The move counters a shift toward closed research, while competing with frequent open releases from Chinese model developers.

• By growing an open model ecosystem, Nvidia aims to keep developers anchored to its platform as rivals build alternative silicon stacks.

Enterprise AI Applications

Coursera and Udemy strike a merger agreement valued around $2.5 billion to build a larger AI-first online learning platform. Link.

• The deal combines Coursera’s catalog and enterprise footprint with Udemy’s marketplace scale, widening distribution for professional upskilling.

• The companies framed AI as central, aiming to personalize learning paths and improve content discovery across career and team training.

• A combined platform could raise switching costs for employers by bundling analytics, credentialing, and skills development into one workflow.

• The move signals consolidation as education platforms race to embed AI tutoring features and defend margins against commoditized course content.

Google Cloud and Palo Alto Networks expand their partnership in a security services deal reported to approach $10 billion. Link.

• The agreement deepens commercial and technical ties, positioning Google Cloud as a key platform for Palo Alto’s security offerings.

• A source said it is Google Cloud’s largest security services deal, reflecting rising spend on AI era cyber defenses.

• The pact includes moving Palo Alto services onto Google Cloud and jointly building AI driven security offerings for customers.

• The partnership highlights enterprise demand for integrated cloud, data, and security stacks that simplify procurement and deployment.

Content Creation

Adobe and Runway team up to bring next generation AI video tools to creators, pairing Runway models with Adobe Firefly workflows. Link.

• The partnership uses Adobe’s Creative Cloud footprint as a distribution channel for Runway’s generative video capabilities.

• Adobe said creators will get access to Runway’s latest model in Firefly, with controls aimed at professional video production needs.

• The move intensifies competition in text to video as studios and brands adopt synthetic footage pipelines for marketing and media.

• For users, tighter integration can reduce tool switching, making generative video features feel native inside familiar editing workflows.

Data Centers + Energy

Japan plans its biggest data center hub in Toyama prefecture, targeting 3.1 gigawatts of capacity to serve AI demand. Link.

• The Nanto city project is coordinated with private firm GigaStream Toyama and aims to diversify infrastructure beyond Tokyo and Osaka.

• A first phase delivering 400 megawatts is scheduled for readiness by late 2028, according to a document cited by Reuters.

• Developers are pitching the site as disaster resilient, aiming to attract hyperscale operators and AI related cloud workloads.

• The scale underscores how AI services are driving grid and site strategy, with power access becoming a primary constraint for growth.

Startup Funding & Valuations

Fluency raises $40 million Series A to scale its AI driven digital advertising operating system and automation engine. Link.

• The round was led and fully funded by Integrity Growth Partners, according to Fluency CEO and co-founder Mike Lane.

• Fluency centralizes creative development, campaign execution, and reporting so advertisers can adjust multi channel strategy from one place.

• Funding will expand AI and engineering work, including an agentic RAG tool that suggests and applies blueprint changes automatically.

• Lane said Fluency has been profitable since 2020, and plans to grow headcount from under 120 to about 200.

Databricks raises $4 billion at a $134 billion valuation as demand for its AI and data platforms accelerates. Link.

• The round signals continued investor appetite for late stage AI infrastructure leaders that can monetize enterprise data modernization trends.

• Databricks is positioning its platform as a foundation for building and governing AI applications on top of large data estates.

• Fresh capital can fund acquisitions and compute spending, helping the company compete with cloud hyperscalers and rival data stacks.

• The valuation step up reflects confidence that AI driven analytics workloads will remain sticky, even as CIOs scrutinize platform consolidation.

Mirelo raises $41 million from Index and a16z to add voice and sound to AI-generated video, solving a common quality gap. Link.

• The startup targets creators and brands that generate video quickly, but still need believable audio, dubbing, and sound design output.

• Funding can expand model training and licensing, where high quality voice data and rights management are key differentiators.

• Better integrated audio reduces post production overhead, making short form marketing videos cheaper to iterate and localize at scale.

• The round underscores a shift from novelty generation to production readiness, where finishing tools drive real customer spend.

Vibe-coding startup Lovable raises $330 million at a $6.6 billion valuation to scale its AI-assisted software building workflow. Link.

• The round highlights strong demand for tools that let nontraditional builders ship production software faster with natural language guidance.

• Lovable’s valuation jump suggests investors expect durable usage, not a short lived trend, as teams standardize on AI coding habits.

• Capital likely supports model costs, safety layers, and enterprise features needed to move from individual creators into larger organizations.

• The deal adds to a growing cohort of AI dev tools startups attracting large rounds as incumbents race to bundle similar capabilities.

Regulation + Legal

Adobe is sued in California federal court for allegedly using authors’ books to train its SlimLM models without permission. Link.

• Author Elizabeth Lyon filed a proposed class action in Northern District of California, seeking damages for affected copyright holders.

• The complaint alleges Adobe used pirated copies of books, extending a wave of litigation over training data for generative models.

• SlimLM is described as a small language model designed to assist with document tasks, including on mobile devices, Reuters reported.

• The case adds legal pressure on enterprise AI vendors, raising risks around dataset provenance, disclosures, and customer contract warranties.

The US launches an interagency review on whether to approve Nvidia H200 AI chip shipments to China, sources told Reuters. Link.

• The review follows a Trump pledge to allow sales with a proposed 25% government fee, shifting from prior export prohibitions.

• Commerce, State, Energy, and Defense provide input, while the final decision sits with the White House, per the report.

• National security critics argue H200 availability could strengthen Chinese military relevant AI capabilities and erode US strategic advantage.

• Nvidia faces strong China demand, and the decision could reshape both supply planning and geopolitical risk for advanced accelerator exports.

Safety + Ethics

UK performers back industrial action by a 99.6% vote, pushing for stronger rules on AI scanning and digital likeness use. Link.

• Equity said the indicative ballot supports resisting on set scanning without clear consent, usage limits, and compensation safeguards.

• The union wants protections that cover both initial capture and downstream reuse across games, film, and advertising workflows.

• A landslide vote strengthens Equity’s leverage in talks with studios and platforms seeking scalable synthetic performances.

• The dispute shows creators pushing enforceable boundaries as AI tools make voice and image replication cheaper and more widespread.

OpenAI

Amazon is in talks to invest more than $10 billion in OpenAI as the ChatGPT maker seeks funding for heavy data center spending. Link.

• The Guardian reported the potential deal as OpenAI looks for capital to support rapid growth and rising infrastructure commitments.

• A large Amazon check would diversify OpenAI’s funding sources, adding a major new partner alongside existing Microsoft relationships.

• The talks underline how AI leaders are competing on compute access, where financing and long term supply agreements shape roadmaps.

• If completed, the investment could deepen OpenAI’s cloud options and shift leverage among hyperscalers racing to host frontier models.

OpenAI introduces a new app directory and SDK to help users discover and integrate third-party tools inside ChatGPT. Link.

• The app directory centralizes tool discovery, helping users find workflow add ons without relying on external lists or social sharing.

• OpenAI’s SDK supports building and publishing integrations, aiming to standardize how apps connect into ChatGPT experiences.

• A cleaner distribution channel can boost quality control, as OpenAI can set policies and surface trustworthy integrations more consistently.

• The launch signals continued platformization, with ChatGPT shifting from a single interface into an ecosystem for specialized AI workflows.

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