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

• TSMC’s Arizona fab produced the first U.S.-made Blackwell wafer for Nvidia—an anchor moment for domestic advanced chipmaking. The Phoenix line will span 2nm–4nm nodes (including AI-grade A16), tightening U.S. supply chains as AI hardware demand explodes.

• An AI Infrastructure Partnership of BlackRock, Nvidia, Microsoft, and xAI is acquiring Aligned Data Centers for $40B. With ~5 GW of current and planned capacity, the deal is about locking power, real estate, and racks at scale as hyperscaler demand outstrips supply.

• Anthropic projects a 2026 revenue run-rate of $20–$26B (after a targeted $9B in 2025), driven largely by enterprise adoption of the Claude portfolio. The surge positions Anthropic as a top rival to OpenAI and signals sustained demand for AI agents, coding, and knowledge workflows.

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

Nvidia and TSMC unveil first U.S.-made Blackwell GPU wafer at Arizona fab. Link.

• Advanced 4nm-class wafer made in Phoenix will power Nvidia’s next-generation Blackwell AI accelerators.

• Localizing production strengthens U.S. supply chains and aligns with domestic semiconductor incentives.

• TSMC Arizona will make 2nm, 3nm, and 4nm chips, including AI-grade A16 for high-performance computing.

• Move reflects surging AI hardware demand and efforts to secure American leadership in chipmaking.

Nvidia launches DGX Spark, a desk-side AI supercomputer targeting on‑prem developers. Link.

• Grace‑Blackwell platform pairs a 20‑core Arm Grace CPU with a Blackwell GPU sharing 128GB unified memory.

• Supports massive models up to roughly 200 billion parameters with 4‑bit weight compression techniques.

• Priced at $3,999 with ~240W power draw, units can be linked to scale memory and throughput.

• Full CUDA AI software stack support enables local prototyping, fine‑tuning, and lab‑friendly inference.

Software

Windows 11 gains hands‑free “Hey Copilot” and agentic PC Actions capabilities. Link.

• Copilot can organize files, launch apps, adjust settings, and assist inside apps with on‑screen context.

• Taskbar and search integration plus Copilot Vision provide conversational, in‑app guidance across workflows.

• User‑approved autonomy introduces a third input mode beyond keyboard and mouse for everyday PC tasks.

• Features reflect Microsoft’s push toward conversational and proactive Windows experiences.

Models

Anthropic projects a $20B–$26B 2026 run‑rate as enterprise adoption accelerates. Link.

• Company targets $9B 2025 run‑rate and plans to more than double revenue next year.

• Roughly 80 percent of sales come from business users, with 300,000+ enterprises onboard.

• Cheaper Haiku 4.5 and broader Claude portfolio drive traction as firms seek OpenAI alternatives.

• Valuation jumped following a $13B raise, positioning Anthropic closer to rival OpenAI.

Product Launches

Intel unveils Panther Lake, its first 18A process PC chip for AI notebooks. Link.

• New transistor design and improved power delivery target large performance gains over 20A generation.

• Integrated CPU and GPU aim at premium AI laptops, enabling richer on‑device machine‑learning features.

• Initial shipments expected late 2025 with broad availability in early 2026 if ramp remains on schedule.

• Milestone is central to Intel’s foundry turnaround and efforts to regain process leadership.

Intel debuts Crescent Island inference GPU to serve token‑heavy LLM workloads. Link.

• 160GB LPDDR5X and Xe3P architecture target energy‑efficient real‑time inference for agentic AI systems.

• Complements Gaudi3 training chip as usage shifts from batch training to ubiquitous deployment.

• Customer sampling slated for second half of 2026 as Intel broadens its accelerator portfolio.

• Move underscores Intel’s push to remain relevant in AI after losses and external investments.

Enterprise AI Applications

Meta partners with Arm to run ranking workloads on Neoverse data‑center CPUs. Link.

• Meta will co‑optimize PyTorch and FBGEMM libraries for Arm, improving efficiency at massive scale.

• Shift complements Nvidia Grace CPU deployments alongside GPUs in Meta’s expanding AI infrastructure.

• Arm’s low‑power, high‑throughput design aims to cut costs while serving 3+ billion users’ feeds.

• Deal reflects multi‑architecture strategies for recommendation systems beyond x86 processors.

Content Creation

Spotify partners with major labels to develop artist‑first generative AI music tools. Link.

• Universal, Sony, Warner, Merlin and others will collaborate on responsible AI products with opt‑in controls.

• Platform will help artists identify and monetize AI usage of their voice or music across content.

• Company tightened policies by removing mass AI spam tracks and labeling AI‑created songs.

• New research lab and product team will co‑design features with rights‑holders to protect creators.

Data Centers + Energy

BlackRock, Nvidia, Microsoft, and xAI acquire Aligned Data Centers for $40B. Link.

• Aligned operates ~80 existing or planned facilities with about five gigawatts of capacity.

• Deal is the first by the AI Infrastructure Partnership to lock in large‑scale AI server supply.

• Move underscores the race among AI leaders to secure power, real estate, and racks at scale.

• OpenAI has separately signed for 26 GW, signaling unprecedented industry capacity build‑outs.

Startup Funding & Valuations

Liberate raises $50M at a $300M valuation to automate insurance operations with AI. Link.

• Battery Ventures led the all‑equity round with Canapi and existing backers Redpoint, Eclipse, and Commerce Ventures.

• Voice assistant and agent network sell policies, process claims, and update accounts end‑to‑end.

• Funding scales deployments as insurers embed AI to cut costs and boost sales across markets.

• Company plans global expansion as adoption of full‑stack insurance automation accelerates.

Viven raises $35M seed to build privacy‑controlled employee “digital twin” LLMs. Link.

• Khosla Ventures led with Foundation Capital and FPV Ventures; co‑founders previously led Eightfold AI.

• Twins train on emails, documents, and Slack so colleagues can query when teammates are unavailable.

• Privacy controls expose twin query history and restrict sensitive information sharing inside firms.

• Genpact among early adopters preserving institutional knowledge and reducing workflow delays.

General Intuition raises $133.7M seed to train agents with spatial‑temporal game video. Link.

• Khosla Ventures and General Catalyst led, with Raine; OpenAI previously pursued Medal’s data for acquisition.

• Two billion game clips per year will teach AI how objects move to improve agents and robots.

• Initial focus spans smarter NPCs and autonomous robotics such as search‑and‑rescue drones.

• Funds expand research team and infrastructure for general‑purpose agent development.

Lila Sciences adds $115M, bringing Series A to $350M for AI science factories. Link.

• Flagship Pioneering incubated lab aims to automate discovery via AI‑controlled robotic laboratories.

• Leased 235,500 square feet in Cambridge to build large AI Science Factory facilities.

• Platform will open to commercial customers through enterprise software offerings.

• Company reports thousands of discoveries across life sciences, chemistry, and materials.

Regulation + Legal

New York state courts issue interim rules for AI use by judges and staff. Link.

• Policy limits generative AI to court‑approved tools and requires staff training on proper use.

• Bars inputting confidential or privileged case data into non‑private AI platforms.

• Emphasizes preventing biased or incorrect outcomes and preserving human judicial judgment.

• Guidance follows nationwide incidents of fake citations and misuse by litigants and lawyers.

U.S. Supreme Court asked to decide if AI‑generated art can be copyrighted. Link.

• Computer scientist Stephen Thaler seeks review after lower courts required human authorship.

• Case centers on an image produced by his DABUS system that the Copyright Office refused.

• Petition argues current stance chills innovation as generative AI output proliferates.

• Resolution could set key precedent for protection and incentives in AI‑assisted creativity.

Safety + Ethics

OpenAI pauses MLK deepfakes on Sora and adds opt‑out for public‑figure likenesses. Link.

• Sora’s early public feed drew backlash over disrespectful and misleading AI video content.

• OpenAI temporarily disabled MLK content while implementing additional safeguards.

• Estates and representatives can now opt out of appearance in AI‑generated outputs.

• Incident highlights persistent risks of generative media misuse and moderation limits.

OpenAI

OpenAI partners with Broadcom to codesign 10 GW of custom AI accelerators. Link.

• Effort aims to embed model insights into hardware and secure compute supply from 2026 onward.

• Financial Times estimates total costs in the hundreds of billions as infrastructure scales.

• Arrangement follows AMD six‑gigawatt deal and Nvidia investment up to one hundred billion.

• Strategy diversifies silicon partners to mitigate supply constraints and vendor reliance.

OpenAI outlines multi‑year plan to fund and monetize trillion‑dollar AI infrastructure. Link.

• Company reportedly generates about thirteen billion dollars annually but plans twenty six gigawatts of capacity.

• Roadmap includes government contracts, new consumer services, hardware, and reselling cloud compute.

• Goal is to bridge the gap between current income and massive capex commitments tied to Stargate.

• Success seen as critical for broader market given interlocking contracts across the AI ecosystem.

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