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

• Apple is reportedly developing AI-first wearables including smart glasses, camera-enabled headphones, and a Siri-driven pin, signaling a push to extend generative AI into always-on hardware tied to the iPhone. If the devices ship on schedule, Apple could redefine assistants as ambient systems rather than screen-based tools.

• Google’s Gemini 3.1 Pro and Alibaba’s Qwen 3.5 releases show the model race shifting from size to efficiency and reasoning. Both emphasize stronger performance on complex tasks while lowering operating costs, reinforcing a broader industry pivot toward practical capability gains instead of headline parameter counts.

• World Labs, founded by Fei-Fei Li, raised $1 billion to build spatial intelligence models that understand and generate 3D environments. The massive round signals investor belief that world-model AI, not chatbots, may define the next platform layer for robotics, simulation, and interactive software.

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

Apple is reportedly building AI-first wearables, including smart glasses, an AI pin, and AI headphones that rely on iPhone pairing. Link.

• Bloomberg reporting says the N50 smart glasses include a camera and could enter production in December 2026.

• The AI pin is described as camera-equipped and Siri-driven, aiming for hands-free context capture without a phone screen.

• New AI headphones would add cameras and sensors, using on-device signals plus iPhone compute to power assistant features.

• The roadmap shows Apple racing to modernize Siri and expand generative AI into always-on hardware beyond iPhone and Mac.

Taalas raised $169 million to build inference chips optimized for specific AI models, starting with a processor tuned for Llama 3.1 8B. Link.

• Taalas says its HC1 chip embeds model components in silicon, targeting faster inference and lower power than general GPUs.

• The company said the funding brings total capital raised to about $219 million since emerging from stealth in 2024.

• It expects a new chip targeting a larger Llama model later in 2026, using a similar model-specific design approach.

• The raise signals investor appetite for inference hardware that competes on watts per token as data centers hit power limits.

Models

Google released Gemini 3.1 Pro, positioning it as an upgraded model for complex tasks that require deeper reasoning than quick answers. Link.

• Google says Gemini 3.1 Pro is rolling out across developer, enterprise, and consumer surfaces, expanding access beyond labs.

• The post highlights improved reasoning and stronger benchmark scores, aimed at synthesis, planning, and multi-step problem solving.

• Google frames it as better for workloads where users need explanations and tradeoffs, not just a single confident response.

• The update strengthens Google’s premium AI tier, as vendors compete on reliability and reasoning rather than raw parameter counts.

Alibaba released Qwen 3.5, claiming the 397B-A17 variant beats its larger trillion-parameter model at a much lower operating cost. Link.

• VentureBeat says Alibaba pitches Qwen 3.5 as 60% cheaper and more efficient on heavy workloads than before.

• The model targets agentic use cases, with capabilities designed to take actions across apps rather than only answer questions.

• Alibaba timed the release for Lunar New Year competition, as Chinese chatbots fight for users against ByteDance and DeepSeek.

• Lower cost, higher throughput models could accelerate enterprise adoption where inference bills, not training, dominate budgets today.

Product Launches

Anthropic rolled out an AI tool designed to hunt software bugs autonomously, targeting dangerous vulnerabilities that manual reviews miss. Link.

• Fortune says the tool can scan code, propose fixes, and prioritize issues based on exploitability and real-world impact.

• It is positioned for security teams and large codebases, where coverage gaps and slow triage leave high-risk bugs unpatched.

• Anthropic frames the product as agentic, meaning it can take multi-step actions beyond chat, such as running checks and tests.

• The launch reflects a shift from general copilots toward specialized AI security products with clear ROI and measurable risk reduction.

Enterprise + Consumer AI Applications

Google and Sea agreed to build AI tools for Shopee and Garena, including agentic shopping prototypes and productivity features for game development. Link.

• Forbes reports the partnership will explore AI shopping agents for Shopee that can handle complex tasks across the purchase journey.

• Sea says Garena will use Google’s AI to improve developer productivity, applying generative tools to game creation workflows.

• The deal builds on earlier collaborations between the companies, signaling deeper AI integration across Southeast Asian consumer platforms.

• It shows platform leaders moving beyond chatbots toward embedded agents that drive conversion, retention, and operational efficiency.

Google added its Lyria 3 music generator to the Gemini app, letting users create 30-second tracks from text, images, or video prompts. Link.

• The Verge says the feature is available globally in eight languages for users over 18, expanding beyond prior Vertex access.

• Users can generate instrumental tracks or songs with AI lyrics by describing genres, moods, or personal memories.

• Gemini also generates cover art via Nano Banana, making it easier to share music clips across social channels.

• The move pushes Gemini into creative consumer workflows, challenging rivals as generative media features become table stakes.

Data Centers + Energy

Zeo Energy signed an MoU to develop 280 megawatts of solar plus long-duration storage for a new Utah data center hub. Link.

• PV Magazine reports the project supports Creekstone’s Gigasite in Millard County, with Zeo assessing behind-the-meter generation options.

• Zeo says the solar and storage package could cover 60% to 80% of site needs, reducing grid dependence.

• Creekstone began construction in December 2025 and plans additional gas-based capacity at the Gigasite by mid 2027.

• The deal highlights how data center developers are assembling multi-source power stacks to shorten timelines and manage costs.

Meta signed a multiyear agreement with Nvidia for millions of AI chips, expanding data center capacity with Blackwell, Rubin, and Grace systems. Link.

• The Verge says the deal includes Blackwell and Rubin GPUs plus Grace and Vera CPUs for Meta’s AI roadmap.

• It marks a large deployment of Nvidia Grace systems, as Meta scales clusters for training and serving its generative models.

• Meta is also building in-house chips, but delays have kept Nvidia as the primary supplier for near-term capacity growth.

• The agreement underscores how hyperscalers lock in silicon early, turning supply commitments into a durable competitive advantage.

Startup Funding & Valuations

Code Metal raised $125 million to expand AI software that translates and verifies code for defense and other high-stakes systems. Link.

• Wired reports the Series B values Code Metal at about $1.25 billion, led by Salesforce Ventures and other backers.

• The platform converts code between languages and hardware targets, then generates tests to validate behavior and catch regressions.

• Customers cited include L3Harris, RTX, the US Air Force, and Toshiba, reflecting demand for modernization of legacy codebases.

• The pitch is verifiable translation, where reliability beats flashy demos and organizations will pay for measurable time savings.

Ricursive Intelligence raised $300 million at a $4 billion valuation to apply AI to chip design, building on work behind Google’s AlphaChip. Link.

• TechCrunch reports the round followed a $35 million seed months earlier, reflecting intense investor demand for chip design automation.

• The company aims to speed floorplanning and optimization, reducing time and cost for silicon across GPUs, NPUs, and accelerators.

• Investors include Nvidia, signaling strategic interest as chipmakers seek faster design cycles to keep pace with AI demand.

• AI-assisted EDA could become a leverage point for hardware teams, especially as advanced nodes and packaging raise complexity.

Freeform raised $67 million to scale AI-driven metal 3D printing, upgrading from an 18-laser system to a design with hundreds of lasers. Link.

• TechCrunch says the GoldenEye printer uses Nvidia H200 GPU clusters to simulate prints and adjust parameters in near real time.

• The company plans its Skyfall system to produce thousands of kilograms per day, targeting industrial throughput beyond prototyping.

• Funding supports engineering and production scale, as aerospace and defense seek faster, more flexible metal parts manufacturing.

• The approach shows AI moving into physical production, where software optimization can unlock capacity without building new factories.

Temporal raised $300 million at a $5 billion valuation to expand its workflow orchestration platform for reliable agentic applications. Link.

• Temporal says the Series D was led by Andreessen Horowitz with participation from Lightspeed, Sapphire, and multiple existing investors.

• The platform provides durable execution, ensuring long-running workflows resume correctly after failures, a core need for production agents.

• Temporal cites adoption across large organizations building AI systems, where retries, state, and auditability determine production reliability.

• The funding signals rising demand for infrastructure layers that make AI agents dependable, not just capable in demos.

World Labs, founded by Fei-Fei Li, raised $1 billion to build spatial intelligence models that generate and understand 3D environments. Link.

• TechCrunch reports investors include AMD, Nvidia, and Autodesk, with Autodesk putting $200 million into the round itself.

• World Labs is developing 3D world models for simulation, robotics, and content workflows, moving beyond text-only generative systems.

• The company previously raised $230 million in 2024, and the new financing accelerates training and product development.

• Big rounds for 3D AI suggest the next frontier is world understanding, where spatial context enables better agents and robots.

Regulation + Legal

A US judge blocked OpenAI from using the name Cameo in a video feature, citing likely consumer confusion in a trademark dispute. Link.

• Reuters reports the injunction followed a lawsuit by celebrity video platform Cameo, which argued OpenAI’s branding harms its business.

• The judge said OpenAI’s use was likely to confuse consumers and cause irreparable harm, after an earlier restraining order.

• OpenAI disputed that anyone can own the generic term cameo, but the court sided with the plaintiff at this stage.

• The ruling shows AI product naming and avatar features can trigger swift legal action as labs expand into media tools.

Safety + Ethics

Hollywood groups escalated pressure on ByteDance’s Seedance 2.0, alleging large-scale copyright infringement by its AI video generator. Link.

• Axios reports the Motion Picture Association sent a cease-and-desist letter demanding details on safeguards and infringement prevention.

• The MPA claims Seedance can replicate copyrighted characters and elements, raising concerns about training data and output controls.

• The letter sets a response deadline of February 27, increasing legal and reputational risk for AI video developers.

• The conflict highlights how generative video is becoming the next major IP battlefield after text and image generation.

OpenAI

OpenAI told investors it expects to spend about $600 billion on compute through 2030 as it scales training and inference demand. Link.

• Reuters reports OpenAI generated about $13 billion in 2025 revenue while spending about $8 billion, below its internal target.

• OpenAI is pursuing a funding round above $100 billion, with a reported Nvidia stake valuing it near $830 billion.

• OpenAI projects total revenue above $280 billion by 2030, split across consumer and enterprise products, reflecting aggressive growth plans.

• The numbers show compute is the dominant cost driver, pushing AI firms toward megascale financing and long-term infrastructure contracts.

A petition to bring back ChatGPT’s GPT-4o gained over 20,000 signatures after OpenAI retired the model from ChatGPT on February 13. Link.

• Business Insider says supporters praised GPT-4o’s warmer tone, while OpenAI said it made up about 0.1% of use.

• OpenAI said feedback from GPT-4o informed newer models, but users argued the experience was uniquely helpful for companionship and tone.

• The campaign includes social posts and threats to cancel subscriptions, showing how model retirements can trigger customer backlash.

• The episode raises questions about continuity and trust, especially when people build workflows or emotional reliance around specific models.

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