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

• Microsoft faces significant setbacks as its in-house AI chip "Braga" is delayed until 2026, still trailing NVIDIA’s industry-leading Blackwell GPUs. This highlights the enormous challenge—even for tech giants—in developing custom silicon that can keep pace with generative AI’s rapid advancements.

• The U.S. Senate struck down a controversial provision that would have banned state-level AI regulations for ten years, preserving individual states' authority to craft their own rules. This decision ensures ongoing regulatory complexity, with AI companies now facing potentially dozens of differing local laws rather than a single federal standard.

• Meanwhile, Google expanded the global rollout of its generative video model Veo 3, available now to Gemini Pro subscribers across 159 countries. This strategic move positions Google squarely against rivals like Runway and Meta, intensifying competition in the rapidly evolving AI-driven video generation market.

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

OpenAI starts using Google's TPUs, diversifying beyond exclusive NVIDIA reliance. Link.

• OpenAI rents Google Cloud TPU capacity to power inference workloads like ChatGPT responses cost-effectively.

• Google retains its newest TPU versions internally but validates its platform’s competitiveness through this deal.

• This move diversifies OpenAI’s infrastructure, reducing dependency risks on a single vendor’s GPUs.

• It underscores cloud providers’ strategic importance as AI compute demand surges and supply chains tighten.

Microsoft’s AI chip "Braga" faces significant delays, trails NVIDIA's performance. Link.

• Microsoft’s in-house Braga chip is delayed until 2026, missing performance targets versus NVIDIA’s Blackwell GPUs.

• Braga’s original design focused on vision workloads, limiting its competitiveness for current generative AI use cases.

• This setback highlights challenges of custom silicon projects amid rapid AI performance advancements.

• The delay amplifies pressure on Microsoft to refine future designs or remain reliant on third-party GPUs.

Models

Google expands global rollout of generative video model Veo 3 within Gemini suite. Link.

• Gemini AI Pro subscribers in 159 countries can generate up to three 8-second videos daily from text prompts.

• Planned enhancements include image-to-video functionality to give users more creative control over content.

• The launch positions Google competitively against emerging video-generation startups like Runway and Meta.

• Built-in safety measures like watermarking and moderation guardrails aim to prevent deepfake misuse.

Consumer AI Applications

Meta tests proactive AI chatbots to boost engagement on Messenger and WhatsApp. Link.

• Meta’s AI personas can proactively message users within two weeks of prior interaction if invited by engagement.

• Users control outreach frequency and privacy settings to balance personalization with user comfort.

• This approach mirrors strategies from apps like Character.AI and Replika to sustain user interest.

• The initiative raises important questions about consent and the ethics of unsolicited AI communication.

Enterprise AI Applications

Grammarly acquires Superhuman email app to launch AI-driven productivity suite. Link.

• Grammarly integrates Superhuman’s AI email drafting and triage features into its writing assistant ecosystem.

• The acquisition provides Grammarly with premium interface access and user data to refine its AI models.

• This move challenges big-tech productivity offerings by combining advanced writing AI with seamless email workflows.

• It underscores a consolidation trend in productivity AI as companies embed agent-based features across core applications.

Amazon deploys its one-millionth robot and will soon match human staff levels. Link.

• Amazon’s fulfillment centers now operate with roughly one robot per human worker after 13 years of automation.

• The new DeepFleet AI system optimizes robot movements and task assignments in real time to boost throughput.

• DeepFleet leverages SageMaker-trained models and internal logistics data to reduce bottlenecks by an estimated 10%.

• This milestone signals a future of seamless human-robot collaboration in logistics at unprecedented operational scale.

Content Creation

Cloudflare launches Pay Per Crawl marketplace to monetize AI scraping. Link.

• The platform allows publishers to charge AI bots per API request, block crawlers, or customize access rules.

• Cloudflare aims to help content owners reclaim revenue lost to generative AI-driven content extraction.

• By mediating licensing fees, the marketplace could standardize costs for AI model training across the web.

• This move may reshape generative AI training economics by making data licensing a standard cost factor for developers.

Data Centers + Energy

xAI granted permits for controversial gas-powered data center amid environmental scrutiny. Link.

• Shelby County approved 15 gas turbines producing 247MW to support xAI’s Memphis data center power needs.

• The Southern Environmental Law Center is preparing a lawsuit over prior unpermitted generator operations.

• Permits cap emissions of NOx, CO, particulate matter, and volatile organic compounds under Clean Air Act standards.

• This controversy highlights tensions between AI compute demands and environmental sustainability efforts.

Startup Funding & Valuations

Levelpath raises $55M Series B to scale AI-native enterprise procurement platform. Link.

• Battery Ventures led the round with Benchmark and Redpoint joining, reflecting strong confidence in Levelpath’s growth.

• The platform uses AI to analyze vendor contracts and recommend cost-saving alternatives, reducing off-platform procurement risks.

• Founders from Scout RFP aim to quadruple revenue by accelerating product development and customer acquisition with new funds.

• Levelpath targets a fragmented $7B market dominated by Coupa and Ariba, capitalizing on demand for simplified AI workflows.

Genesis AI emerges from stealth with $105M seed for robotics foundation models. Link.

• The seed round, co-led by Eclipse Ventures and Khosla Ventures, highlights strong investor belief in robot intelligence.

• Genesis AI leverages proprietary physics simulations to train models, cutting real-world data collection costs dramatically.

• Founded by CMU alumni and ex-Mistral AI researchers, the company focuses on automating repetitive lab and household tasks.

• This capital positions Genesis to compete with Physical Intelligence and other startups in enterprise robotics solutions.

Savvy Wealth secures $72M Series B to scale AI-driven advisor platform. Link.

• Industry Ventures led the investment, bringing Savvy Wealth’s total funding to $106M as it enhances its AI models.

• The service combines generative AI with human advisors to improve portfolio analysis and automate client communications.

• Investors see AI as a force multiplier, allowing advisors to serve more clients while maintaining personalized advice.

• Savvy Wealth aims to disrupt large incumbents by enabling advisors to manage significantly larger client rosters efficiently.

Regulation + Legal

US Senate strikes state-level AI regulation ban from Trump megabill. Link.

• Senators voted to remove a provision that would have barred state AI regulations for ten years.

• Tech leaders, including OpenAI’s Sam Altman, supported the moratorium, citing innovation concerns.

• Its removal preserves states’ authority to craft AI rules, likely resulting in a patchwork of local laws.

• The decision exposes deep legislative divides over federal vs. state roles in governing emerging AI technologies.

OpenAI

OpenAI warns Robinhood’s tokenized shares aren’t real equity. Link.

• OpenAI clarified that Robinhood’s “OpenAI tokens” have no endorsement and do not represent actual company shares.

• The tokens launched as part of Robinhood’s European expansion, leading to confusion about private-equity tokenization.

• OpenAI’s swift public warning highlighted legal risks and IP concerns surrounding tokenized representations of startups.

• This episode underscores regulatory scrutiny as financial firms explore crypto models for unlisted company investments.

Ilya Sutskever takes CEO role at Safe Superintelligence amid talent war. Link.

• Co-founder Ilya Sutskever replaced departing CEO Daniel Gross to lead the stealth AI alignment startup.

• The company, valued at $32B, reaffirmed its commitment to independent research on safe AI systems.

• This leadership change comes amid aggressive recruitment efforts by Meta’s Superintelligence Labs and other rivals.

• Sutskever’s appointment underscores intense competition for top AI talent among leading research organizations.

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