1. Meta shifted Reality Labs investments from the metaverse to AI wearables and proprietary AI models as of Q4 2025.
2. Meta's AI smart glasses sales tripled in the past year, positioning them among the fastest-growing consumer electronics.
3. Zuckerberg predicts widespread adoption of AI glasses within several years, likening the moment to the smartphone transition.
4. Google is set to launch smart glasses in 2026 after a $150 million partnership with Warby Parker.
5. Apple is reportedly reallocating staff to smart glasses development, targeting a launch within one to two years, deprioritizing a lighter Vision Pro.
6. Snap announced on January 28, 2026, the spin-off of its AR glasses, Specs, into a new subsidiary for operational focus.
7. OpenAI is developing AI wearables, focusing on devices like AI pins or earbuds rather than glasses.
8. Apple is rumored to be working on an AirTag-sized AI device.
9. Meta currently leads the smart glasses market with multiple models, including Oakley smart glasses optimized for exercise.

1. Apple’s Foundation Models framework introduces a 3 billion parameter on-device language model for iOS, enabling GPT-like features with full privacy, no API costs, and offline capability.
2. Key features include text generation, summarization, entity extraction, sentiment analysis, tool calling, and guided generation, all with type-safe Swift integration and low latency optimized for Apple Silicon.
3. Requirements for development are Xcode 16+, Swift 6.0+, Apple Intelligence enabled, iPhone 15 Pro or later, and macOS Sequoia 15+.
4. The framework supports real-time streaming for text generation, fully type-safe summarization and entity extraction, sentiment analysis with structured outputs, and autonomous tool calling with parameter extraction.
5. The Foundation Models framework is built-in for iOS 18+ and does not function in the iOS Simulator; a physical device is required.
6. The evolution from iOS 7-12 (no on-device ML), through Core ML era (iOS 11-17, classification tasks), to GenAI (iOS 18+, full LLM capabilities) marks a shift from classification to characterization, cloud to on-device processing, and manual parsing to type-safety.
7. Foundation Models offer advantages over Core ML 3 NL: full text generation, structured outputs, native tool calling, real-time streaming, and general-purpose AI for iOS 18+; Core ML 3 NL remains for classification and multi-language support on earlier iOS versions.
8. Real-world use cases include smart journaling, fitness coaching, study assistants, travel planning, writing tools, and support chatbots—all leveraging on-device generative AI.
9. The framework eliminates cloud dependency, guarantees user privacy, removes API costs, enables type-safe guided generation, and supports offline experiences.
10. The GenAI era on iOS is production-ready as of January 28, 2026, with mature tools and platform support for entrepreneurs building AI-powered apps.

1. Airtable's valuation dropped from $11.7 billion in 2021 to about $4 billion on secondary markets, but the company retains half of its $1.4 billion raised and remains cash-flow positive.
2. Airtable employs over 700 people, serves 500,000+ organizations including 80% of the Fortune 100, and is launching its first stand-alone product, Superagent, after 13 years.
3. Superagent is an AI agent platform focused on multi-agent coordination, enabling parallel deployment of specialized agents for complex tasks and delivering interactive, high-quality outputs.
4. Superagent distinguishes itself from competitors by claiming a generally capable, long-running agent architecture, unlike most "LLM powered workflows."
5. The system leverages premium data sources such as FactSet, Crunchbase, SEC filings, and earnings transcripts for structured, citation-backed outputs.
6. In late 2025, Airtable hired David Azose (ex-OpenAI) as CTO and acquired DeepSky (formerly Gradient), an AI agent startup that raised $40 million; Superagent will be led by DeepSky's founders and operate semi-independently.
7. Pricing for Superagent is expected to range from $20 to $200 per user per month, with generous inference credits, prioritizing adoption over profit margin.
8. Liu positions Airtable as an "AI-native platform" and frames the valuation drop as a recruiting advantage due to more attractively priced equity.
9. Liu maintains strategic capital for acquisitions and does not anticipate needing additional fundraising.
10. Liu views Superagent as a significant but uncertain opportunity, emphasizing optionality and rapid adaptation as key leadership strategies in the current AI market.

1. OpenAI, valued at $500 billion, is seeking an additional $100 billion in investment, potentially raising its valuation to $830 billion.
2. Amazon may contribute at least $50 billion to this funding round, with CEO Andy Jassy leading negotiations with OpenAI CEO Sam Altman.
3. OpenAI is also in funding discussions with Middle Eastern sovereign wealth funds, Nvidia, Microsoft, and SoftBank.
4. The funding deal is expected to close by the end of Q1 2026.
5. Amazon's potential investment in OpenAI is notable due to its existing $8 billion investment in Anthropic and AWS's role as Anthropic's primary cloud and training provider.
6. Amazon recently launched an $11 billion data center campus in Indiana dedicated to running Anthropic models.

1. Harvey has acquired Hexus, a two-year-old startup specializing in tools for product demos, videos, and guides, as part of its expansion in the legal tech market.
2. Hexus founder Sakshi Pratap and her San Francisco team have joined Harvey, with India-based engineers to follow after a Bangalore office is established.
3. Pratap will lead an engineering team focused on accelerating Harvey’s offerings for in-house legal departments.
4. Hexus previously raised $1.6 million from Pear VC, Liquid 2 Ventures, and angel investors; acquisition terms are structured around long-term team incentives.
5. Harvey is valued at $8 billion as of fall 2025, after raising $160 million, bringing its total 2025 funding to $760 million.
6. Andreessen Horowitz led the latest round, joined by T. Rowe Price, WndrCo, Sequoia Capital, Kleiner Perkins, Conviction, and Elad Gil.
7. Harvey started 2025 with a $3 billion valuation after a $300 million Series D led by Sequoia.
8. Harvey claims over 1,000 clients in 60 countries, including most of the top 10 U.S. law firms.
9. Harvey originated from a July 4, 2022, cold email to OpenAI CEO Sam Altman, resulting in early investment from the OpenAI Startup Fund, now Harvey’s second-largest investor.

1. StepFun, a Shanghai-based large-model startup, secured over RMB 5 billion ($718 million) in a B+ financing round, marking the largest single financing in China's large-model sector in the past 12 months.
2. Investors included Shanghai SDIC Leading Fund, China Life Private Equity Investment, Pudong Venture Capital, Xuhui Capital, Wuxi Liangxi Fund, Xiamen ITG Group, Huaqin Technology, and follow-on investments from Tencent Holdings, Qiming Venture Partners, and 5Y Capital.
3. The capital will be allocated to foundational model development, construction of a major base model, and acceleration of StepFun’s AI+terminal strategy.

1. Meta will launch new AI models and products within months, following a foundational rebuild of its AI program in 2025.
2. AI-driven commerce, specifically agentic shopping tools, is a key focus area for Meta's upcoming offerings.
3. Meta emphasizes its unique advantage in leveraging personal data to deliver highly contextualized AI agent experiences.
4. Meta acquired Manus, a general-purpose agent developer, in December and will integrate its technology while continuing Manus services.
5. Meta projects capital expenditures of $115–$135 billion for 2026, up from $72 billion in 2025, primarily to support Meta Superintelligence Labs and core business.
6. The projected infrastructure spending remains below the $600 billion anticipated by 2028.
7. Despite investor concerns about ROI, Meta plans to deliver public-facing AI advancements and infrastructure in 2026.
8. TechCrunch Founder Summit 2026 will be held on June 23 in Boston, gathering over 1,100 founders for growth and scaling strategies.

1. NVIDIA and Mercedes-Benz will deliver premium autonomous rides via Uber’s global mobility platform using the new S-Class.
2. The new S-Class, marking 140 years of Mercedes-Benz innovation, features a level 4-ready architecture built on NVIDIA DRIVE Hyperion and full-stack DRIVE AV L4 software.
3. The S-Class is designed for future robotaxi operations, integrating NVIDIA Halos and parallel AI/classical driving stacks for safety and reliability.
4. The partnership enables a premium, chauffeur-style autonomous experience through Uber’s network.
5. NVIDIA DRIVE AV, trained on DGX systems and validated with Omniverse NuRec and Cosmos world models, addresses complex edge cases in real-world driving.
6. DRIVE AV is optimized for Mercedes-Benz platforms and sensor configurations, enabling real-time evaluation and selection of the safest outcomes.
7. NVIDIA DRIVE Hyperion provides sensor diversity, hardware redundancy, and software stack diversity for robust level 4 autonomy.
8. The platform incorporates defense-in-depth principles, including redundant compute, multimodal sensors (cameras, radar, lidar), and parallel safety stacks.
9. NVIDIA Halos safety system and DRIVE Hyperion eliminate single points of failure, supporting large-scale, reliable robotaxi deployment.
10. NVIDIA’s AI ecosystem, including Alpamayo open models, simulation tools, and datasets, advances autonomous driving research and production readiness.
11. The production-grade approach combines large-scale training, high-fidelity simulation, rigorous safety validation, and deep system integration for both level 2 and level 4-ready systems.
12. NVIDIA Alpamayo enables human-like, step-by-step reasoning for safe autonomous driving decisions.
13. The S-Class’s L4-ready architecture extends Mercedes-Benz’s safety leadership, combining end-to-end AI with classical stacks for predictable, reliable operation.
14. The Mercedes-Benz CLA was named Euro NCAP’s Best Performer of 2025, reflecting the shift toward active, intelligent safety.
15. The collaboration demonstrates how legacy automakers and AI pioneers can deliver advanced, safe, and autonomous vehicles without compromising quality.

1. Waabi raised $1 billion, including a $750 million Series C co-led by Khosla Ventures and G2 Venture Partners, and $250 million in milestone-based capital from Uber for deploying 25,000+ Waabi Driver-powered robotaxis exclusively on Uber.
2. Waabi's partnership with Uber marks its first expansion beyond autonomous trucking, aiming to scale across multiple self-driving verticals with a single AI technology stack.
3. Waabi's AI architecture enables simultaneous deployment in trucking and robotaxi markets, claiming capital efficiency and generalizability over competitors.
4. Waabi's technology leverages a closed-loop simulator, Waabi World, for training, testing, and validation, reducing reliance on large-scale real-world data and human intervention.
5. Waabi's approach allows faster and cheaper development, minimizing the need for large teams, fleets, data centers, and high-end chips compared to AV 1.0 systems.
6. Total funding for Waabi now stands at approximately $1.28 billion, following a $200 million Series B in June 2024; competitors Aurora Innovation and Kodiak Robotics have raised $3.46 billion and $448 million, respectively.
7. Waabi has conducted several commercial pilots with safety drivers in Texas and delayed its fully driverless truck launch to the next few quarters.
8. Waabi is collaborating with Volvo to build purpose-built autonomous trucks, revealed in October 2025 at TechCrunch Disrupt, with technology ready but awaiting full vehicle validation.
9. Waabi employs a direct-to-consumer model for autonomous trucks, enabling shippers to purchase directly, and plans to vertically integrate its technology with OEMs for scalability and safety.
10. Other Series C investors include Uber, NVentures (Nvidia’s VC arm), Volvo Group Venture Capital, Porsche Automobil Holding SE, BlackRock, and BDC Capital’s Thrive Venture Fund.
11. Uber has launched Uber AV Labs to collect data for AV partners and has partnered with multiple AV companies globally, including Waymo, Nuro, Avride, Wayve, WeRide, and Momenta.

1. Waymo robotaxi rides in the San Francisco Bay Area averaged $19.69 between November 27 and January 1, compared to Uber at $17.47 and Lyft at $15.47, based on over 94,000 simulated ride requests by Obi.
2. Since April 2025, Waymo’s average cost dropped 3.62%, Uber’s increased 12%, and Lyft’s rose 7%, narrowing the price gap.
3. Waymo’s average wait time rose to 5.74 minutes, Lyft’s was 5.14 minutes, Uber’s was 3.15 minutes, and Tesla’s was highest at 15.32 minutes.
4. Tesla’s ride-hail service, using a transportation charter permit and employee drivers, is not a true robotaxi and operates a modest Bay Area fleet of about 156–168 vehicles.
5. Tesla’s rides are cheaper but limited by fleet size and long wait times, and its pricing at scale remains uncertain.
6. Over half of 2,000 surveyed respondents in CA, NV, AZ, and TX who had taken an autonomous ride reported riding in a Tesla, and 31% preferred Tesla for autonomous rides, compared to 39.8% for Waymo.
7. Tesla’s brand preference is especially strong among men (56% prefer Tesla vs. 25% for Waymo), while women are evenly split between Tesla and Waymo.
8. Waymo is expanding into new cities, partnering with Uber and Lyft, and will soon offer rides in a new, lower-cost Zeekr-built vehicle called Ojai.
9. Nuro is supplying self-driving systems for Lucid Gravity vehicles in a premium robotaxi network with Uber, and Motional plans to launch a commercial robotaxi service in Las Vegas before the end of 2026.
10. Other companies, including Avride, are partnering with Uber to deploy robotaxis in additional U.S. cities, signaling intensifying competition in the autonomous ride-hailing market.

1. On January 2026, Moonshot AI released Kimi K2.5, a 1 trillion-parameter MoE model with 32 billion active parameters, positioning it as a leading open-source AI contender for 2026.
2. Kimi K2.5 is a next-generation open-source multimodal model excelling in agentic reasoning, vision, and large-scale execution, integrating text, images, videos, and tools.
3. The model features a self-directed agent swarm paradigm, autonomously spawning and coordinating up to 100 sub-agents for thousands of parallel operations without manual orchestration.
4. Kimi K2.5 enables rapid generation of professional PowerPoint presentations, extracting and formatting data from diverse internet sources, and providing source citations.
5. The model can create comprehensive, editable Excel workbooks with multiple linked sheets, dynamic formulas, data validation, and conditional formatting, all generated from natural language prompts in under 2 minutes.
6. Kimi K2.5 produces professionally structured, ready-to-use PDFs with advanced formatting, logical flow, and embedded tables, outperforming other AI tools in document assembly.
7. Outputs from Kimi K2.5 are immediately practical, client-ready, and demonstrate a shift from AI as an assistant to an operator, significantly reducing the gap between user intent and finished assets.
8. Despite its capabilities, verification of AI-generated content remains essential due to the potential for errors.

1. DDR5 RAM prices surged to nearly $800 for a 2x 32GB DDR5-6000 kit in December 2025, surpassing the $749 price of a PS5 Pro.
2. RAM prices are projected to rise another 50% in Q1 2026, with further surges possible later in the year.
3. The primary driver of RAM shortages and price hikes is AI demand, alongside manufacturing challenges and declining DDR4 production.
4. Notebook shipments are shifting to lower-tier 8GB models, reducing availability of 16GB RAM laptops.
5. TrendForce reports conventional DRAM contract prices could rise by up to 60% in Q1 2025, heavily impacting DDR5.
6. Refurbished PCs and older laptops, such as Lenovo ThinkPads under $300 and M3 MacBook Airs around $693, offer cost-effective alternatives.
7. Seasonal sales, like Amazon’s Big Spring Sale in March 2025, provide opportunities for discounted laptops and PCs.
8. Targeting DDR4-based systems or purchasing DDR4 RAM, such as Corsair Vengeance LPX 32GB for $210, can mitigate costs.
9. RAM and motherboard bundles, like a TeamGroup combo for $230 on Newegg, offer savings over separate purchases.
10. ChromeOS devices, such as the Acer Chromebook Plus Spin 514 and CX15 Chromebook, are affordable and require less RAM due to cloud reliance.
11. Linux distributions, including Ubuntu and Q4OS, enable efficient use of older hardware and offer resource savings over Windows 11.

1. Microsoft launched the Maia 200 chip, designed for scalable AI inference, following the Maia 100 from 2023.
2. Maia 200 features over 100 billion transistors, delivers over 10 petaflops at 4-bit precision, and about 5 petaflops at 8-bit, significantly surpassing its predecessor.
3. The chip targets optimization of inference costs, a growing concern for AI companies as they scale operations.
4. One Maia 200 node can run the largest current AI models with capacity for future, larger models.
5. Maia 200 is part of a trend among tech giants to develop proprietary chips, reducing reliance on Nvidia GPUs.
6. Microsoft claims Maia 200 achieves 3x the FP4 performance of Amazon Trainium3 (launched December 2025) and exceeds FP8 performance of Google’s seventh-generation TPU.
7. Maia 200 is already deployed in Microsoft’s Superintelligence team and supports Copilot chatbot operations.
8. Microsoft has released the Maia 200 software development kit to developers, academics, and frontier AI labs as of Monday.

1. Nvidia GH200 is a hybrid superchip combining a 72-core Arm Grace CPU and Hopper/H200 GPU via NVLink-C2C, offering up to 624 GB unified memory and 900 GB/s bandwidth, targeting memory-bound AI workloads in 2026.
2. GH200’s unified memory architecture eliminates PCIe bottlenecks, enabling efficient handling of large LLMs, RAG, exascale simulations, and multimodal AI by supporting massive key–value caches and embeddings.
3. HBM3e revision increases GPU memory to 144 GB and bandwidth to 4.9 TB/s, extending single-chip model capacity by 50% and pushing GH200 into multi-GPU territory.
4. NVLink switch enables scalability, connecting up to 256 GH200 superchips for 1 exaflop compute and 144 TB unified memory, as used in exascale systems like JUPITER and Helios.
5. GH200 outperforms H100 in MLPerf inference by 1.4×–1.8×, achieves 7.6× higher throughput and 8× lower cost per token on Llama 3.1 70B inference, and delivers 2.7×–5.7× speedups over A100 in RAG workloads.
6. GH200 instances cost $4–$6/hour on hyperscalers, slightly above H100, but improved throughput and memory efficiency lower total cost for memory-bound tasks.
7. GH200 excels in LLM inference/training, RAG, GNNs, recommendation systems, video generation, and scientific HPC, supporting context windows of 128k tokens and enabling real-time, deep-context chatbots.
8. JUPITER supercomputer employs 24,000 GH200 chips for over 90 exaflops, each node with 288 Arm cores and 896 GB memory, accelerating climate and physics simulations.
9. GH200 is accessible via on-prem DGX systems, AWS/Azure/Google Cloud, specialist GPU clouds (Lambda, Baseten, Fluence), and Clarifai’s enterprise-grade hosting with smart autoscaling, GPU fractioning, and cross-cloud orchestration.
10. NVLink switch offers lower latency and higher bandwidth than InfiniBand, allowing multiple GH200s to function as a single logical GPU.
11. Decision matrix: choose GH200 for memory-bound workloads needing unified memory; H100/H200 for compute-bound or x86-dependent tasks; B200/Rubin for even larger memory/cost efficiency when available.
12. GH200 systems can draw up to 1,000 W per node; adequate cooling and power infrastructure are required.
13. ARM software stack requires recompilation for some libraries; containerised environments (e.g., Clarifai local runner) mitigate deployment friction.
14. Supply constraints and high demand may limit GH200 cloud availability; pre-ordering or using specialist providers is advised.
15. Future trends: HBM3e, Blackwell (B200/GB200), Rubin platform (88-core CPU, 1.5 TB LPDDR5X, 1.8 TB/s NVLink-6), and NVL72 racks will further increase memory, reduce inference cost by 10×, and cut training GPU count by 4×.
16. Industry adoption is strong, with Google, Meta, Microsoft, and SoftBank investing in GH200 systems; storage and networking vendors are adapting to unified memory architectures.

1. Nvidia's monthly GPU sales for AI data centers have made it the most valuable company globally.
2. ASML, the exclusive supplier of EUV photolithography equipment, is critical for advanced semiconductor manufacturing.
3. ASML reported 32.7 billion euros in net sales in its latest quarterly earnings.
4. ASML set a new record with 13 billion euros in new bookings last quarter, more than double the previous quarter.
5. The surge in ASML's new orders is driven by robust AI-related demand, as confirmed by CEO Christophe Fouquet.
6. Customers anticipate sustained AI infrastructure growth and are investing in chip manufacturing capacity accordingly.
7. There is uncertainty about whether all orders will be fulfilled, but no significant reduction in projected infrastructure spending is evident.

1. Tesla will end production of the Model S sedan and Model X SUV, with final versions produced next quarter as announced on the January 2026 earnings call.
2. Tesla will continue supporting existing Model S and Model X owners indefinitely.
3. The Fremont, California factory space used for Model S and X will be repurposed for Optimus robot production.
4. Production of the Cybertruck at the Austin factory will continue despite poor sales and missed targets.
5. Model S launched in 2012 with a $57,400 base price, collecting over 10,000 reservations before first deliveries in June 2012.
6. Model S received multiple battery options, software-limited range upgrades, and "Ludicrous Mode" enabling 0-60 mph in 2.8 seconds.
7. Model S and X received major refreshes in 2021, but sales have stagnated due to increased competition in the luxury EV market.
8. Model X, launched in 2015, featured complex "Falcon Wing" doors and was described by Musk as the "Fabergé of cars" due to production challenges.
9. Musk stated in 2019 that Model S and X were "niche" vehicles of "minor importance" to Tesla's future.
10. Cybertruck, revealed in 2019 with a planned $40,000 base price and 250,000 units/year target, faced delays and has sold only a few thousand units per quarter.
11. Tesla grouped Model S, X, and Cybertruck as "other models" in sales reports, obscuring individual performance.
12. Tesla is shifting focus toward autonomy and robotics, driving the discontinuation of Model S and Model X.