Bruno Posokhow

🚀 The $1 Trillion "What If" Investment Story: Sega & Nvidia What if a single, seemingly small investment in 1996 could have resulted in a $1 trillion valuation today? That's the staggering scenario surrounding Sega and Nvidia. The story began with a failed contract: Sega had tapped Nvidia to develop the graphics chip for the upcoming Dreamcast console. When the deal fell through, Sega made a crucial move, converting the final contract payment into a $5 million equity investment in the then-tiny graphics card company. This move turned a contract failure into a stake in one of the most explosive tech companies of the next three decades. It's a powerful reminder that sometimes, the most significant long-term wins emerge from pivoting away from short-term disappointments. Sega’s investment, while ultimately highly successful for the time, was short-lived. Following Nvidia’s successful Initial Public Offering (IPO), Sega decided to sell its entire stake. The company walked away with a respectable $15 million, achieving a solid 3x return on their $5 million investment in just about four years. While this was a great financial outcome for the period, the move represents one of the biggest "missed opportunities" in tech history. The decision highlights the constant tension facing corporate treasuries: locking in a safe, guaranteed return versus betting on the exponential, long-term growth of a promising asset. As Nvidia's CEO Jensen Huang often recounts, had Sega held onto that $5 million stake through the subsequent decades of computing, gaming, and the AI revolution, that investment would be worth an estimated $1 trillion today. This Epic story transcends gaming and finance; it's a profound lesson in the power of long-term compounding and the challenge of identifying and holding onto truly disruptive technology early on. It serves as a stark, unforgettable case study for every investor and corporate development leader: sometimes, the greatest wealth is generated by the stock you don't sell. #Nvidia #Sega #Investing #TechHistory #VentureCapital

220

25 Comments

Nvidia just revealed its first robot to us CEO Jensen Huang unveiled 'BLUE', an AI-powered robot that is a leap into the future of robotics and AI. What makes this more interesting is that this wasn’t a scripted demo but was a live showcase of Nvidia’s progress in the world of AI with robotics, proving they’re not just playing in this space but are indeed dominating it. This announcement signals a seismic shift in AI and robotics, with Nvidia leading the charge.

829

84 Comments

🥳 This is such a solid repo that covers all parts of the AI agent production pipeline. It dropped just a week ago and already has 6k stars ⭐ ! Nir Diamant has done it again! Thanks for putting this together. It’s a one-stop resource for anyone building real-world agents for production use-cases. It includes tutorials, notebooks, and examples for every layer : ⛳ Orchestration Design: Multi-tool, memory-aware workflows and agent-to-agent messaging ⛳Tool Integration: Connect agents to databases, web data, and external APIs ⛳Observability: Add tracing, monitoring, and debugging hooks ⛳ Deployment: Ship to containers, GPU clusters, or on-prem servers ⛳ Memory: Implement short- and long-term memory with semantic search ⛳ UI & Frontend: Build chat or dashboard front-ends ⛳ A gent Frameworks: Create stateful graphs, expose agents as REST endpoints, and package reusable tools ⛳Model Customization: Fine-tune LLMs for domain-specific behavior ⛳Multi-agent Coordination: Enable message passing and shared planning ⛳Security: Add real-time guardrails and injection protection ⛳Evaluation: Automate behavioral testing and metric tracking Even if you don’t use the code/notebooks as is, it gives you a clear sense of the key components involved in building a production pipeline and how you might approach each of them! Link: https://lnkd.in/eE9t4ba5

783

45 Comments

NVIDIA just dropped something that will change how we build voice-enabled experiences: Parakeet 🦜-TDT-0.6B-v2 The new open-source ASR model that:  ⚡ Transcribes 60 minutes of audio in 1 second ⚡ Beats most commercial alternatives, and it's free ⚡ Scores a 6.05% WER, topping Hugging Face's Open ASR Leaderboard ⚡ Offers word-level timestamps, punctuation, and supports fine-tuning ⚡ Runs on GPUs with as little as 2GB RAM AI is no longer just for labs. It's in your startup. Your product roadmap. Your next moonshot. 3️⃣ things this model signals for the future: 1/ Speed is no longer a luxury → It's the default. → What used to take hours is now done before your next coffee sip. 2/ Open source is eating the enterprise stack → CC-BY-4.0 license = full commercial use → Small teams can now punch above their weight 3/ Voice is back → Not as gimmicks, but as robust, real-time interfaces → Subtitles, assistants, call analysis, and podcasts, all reimagined ------------------------------------------------------------------------------ ♻️ Follow Aravind Kumaresan🚀 for insights at the intersection of AI, engineering, and real-world impact. 📌 Share this post! Your next product idea might be one transcript away.

93

4 Comments

Looking back at the whiteboard that started it all. Six years ago, we rented an Airbnb for a weekend to dream big about what would become Flux. We started by mapping out the hardware development pipeline — every bottleneck, every point where projects die. Then we gave ourselves permission to think idealistically. Our moonshot: go from requirements to a board design automatically. We didn’t know about LLMs yet, but we knew we wanted to make hardware creation radically easier — to make launching a hardware project as simple as starting a software one. Fast forward six years, and we’ve taken a massive leap toward that vision. Today, we’re launching the first AI hardware engineer — capable of taking your idea, crafting a plan, asking clarifying questions, and automating multi-step workflows that go from idea → schematic → placement → routing. It’s not perfect yet. But for the first time, the entire workflow is connected — and LLMs can finally understand hardware. We’re seeing a new paradigm form: AI agents as true team extensions. You’ll give them tasks, they’ll execute, and check in when they need your product judgment. Right now, it’s like an intern — fast, explainable, and eager to learn — but it’s improving quickly. Flux also lets you teach your AI to work the way you do: correcting mistakes, explaining decisions, and building shared context over time. That means a future where two or three engineers can build what once took hundreds — where small teams compete with giants, and iteration happens at the speed of software. I can’t wait for the coming hardware explosion — when building with atoms becomes as fluid and collaborative as building with bits. Software has its Cursor, Replit, Devin, and Claude Code. Now, hardware has Flux. 🚀 👉 Read more about the launch here: https://lnkd.in/gxhj8tFi

116

4 Comments

A tear down of the just launched Meta Ray ban display glasses version 1 ($800) showing its Geometrical waveguide and LCOS mini projector light engine, along with other important components such as battery (900 mWh or 248mAh capacity) and Snapdragon processor. Most important optical difference from other wearables is the the use of a Geometric waveguide than the traditionally used Diffraction waveguide in the lens. Most of the headsets and many glasses in the market so far (Magicleap, Hololens, Snap spectacles, etc.) all use diffraction waveguides. If you do not know what is a waveguide and the optics works, see the link in the comments, for educational understanding. As a background knowledge, the geometric waveguide uses tiny mirrors embedded in the lens sandwitched between the glass like material layers, to reflect the light thats totally internally reflected after projecting from the side projector, while the diffraction waveguide uses diffraction gratings to diffract the light and are made with mostly lithography machines (Nano-tech). Geometric waveguide offers the benefit of no color dispersion that is typically is seen in diffraction waveguides where the person looking at the glass from outside notices a rectangular color tint in the glass. Geometric waveguide also offers better color quality due to use of the mirrors in the materials than diffraction gratings. #optics #diffraction #light #meta #ar #display https://lnkd.in/etekyrvP

6

1 Comment

🚨 What exactly happens inside a Streaming Multiprocessor (SM) on an NVIDIA GPU? Most developers know that CUDA cores live inside SMs. But few understand the architectural depth and scheduling mechanics that make SMs the real brains of parallel execution. Each SM isn’t just a cluster of arithmetic units. It’s a tightly coordinated hardware subsystem with: 🧠 Warp schedulers that dynamically issue instructions to avoid stalls 📦 Register files optimized for massive thread context switching ⚙️ Tensor Cores, SFUs, and LSUs designed for specialized matrix ops, transcendental math, and memory coalescing 🧵 128+ CUDA cores executing in SIMT (Single Instruction, Multiple Threads) fashion 💡 All backed by L1 cache and shared memory for intra-warp cooperation And it gets deeper: warp scheduling strategies (e.g., GTO vs. Two-Level) and resource partitioning in SMs are what separate good kernel design from great. ➡️ I broke down this entire hierarchy—from GPC to CUDA cores—in Part2 of my detailed blog post designed for engineers and system-level developers. 🔗 Full post in comments. Dive in if you’re ready to truly understand the architectural backbone of NVIDIA GPUs. Comment "Simple but deep dive into NVIDIA GPU architecture" to help bring out Part3 and rest sooner :) #NVIDIA #CUDA #GPUarchitecture #StreamingMultiprocessor #WarpScheduling #TensorCores #ParallelComputing #DeepLearning #GPUs #EngineeringLeadership #BloggingForEngineers #AI

329

27 Comments

🚀 Miko is about to get a whole lot smarter at home. We’ve built our own visual SLAM (vSLAM) stack—enabling Miko to map, localize, and navigate any home environment, regardless of where it starts. But here’s what truly sets this apart: 🧠 It does all this using significantly less compute than typical home robots. ⚡ It’s also faster, more efficient (stay tuned for the bench marking results) Turning the vision of intelligent home robotics—at scale—into reality. Simit Pradhan Prashant Iyengar Chintan Raikar #Robotics #vSLAM #AI #Miko #AutonomousNavigation

319

18 Comments

Luma AI Ray3’s native HDR is interesting. It generates video in 10, 12, and 16-bit high dynamic range, preserving shadow nuance and highlight brilliance in vivid color. SDR can be converted to HDR, and EXR exports ensure integration with existing pipelines while giving unprecedented control in post. That said - speaking from a slightly jaded, feature-pipeline perspective - the real bottleneck in AAA workflows was always logistics. Moving camera rushes, EXRs, and massive datasets around remains painfully costly and time-consuming. Even with faster networks, hard limits persist. Originally, I thought the smarter play would be to replicate the classic proxy offline/online approach: Work with low res 8 bit generated proxies the whole time to keep timelines light, then “up-res” the bit-depth only at the finishing stage for grading and VFX. Instead, what’s emerged feels like the invention of an entirely new camera format. That's also how some of the streamers seem to be viewing it too after some conversations at IBC.

51

17 Comments

As AR glasses move closer to mainstream adoption, the choice of display technology becomes crucial. In Avegant's recent white paper, we delve into the comparison between LCoS (Liquid Crystal on Silicon) and microLED light engines for AR applications. #augmentedreality #display #LCoS #microLED #AI #AIGlasses #AR

183

6 Comments

Meta’s $100 Billion Bet: Can New Leadership Revitalize the Quest? In a move that feels like a high-stakes poker game, Meta is pushing all its chips to the center of the table, announcing a staggering $100 billion investment into its XR product lineup. This includes everything from augmented reality (AR) smart glasses to virtual reality (VR) headsets and the mixed reality Quest series. News Flash: Mark Rabkin, the seasoned VP of VR/MR, is stepping down after an 18-year run (6 years on Quest). His departure paves the way for Ryan Cairns previously at the helm of Portal and Meta for Work, to take the reins. The timing is intriguing. The Quest line has been treading water, not quite making the splash Meta hoped for. Yet, with this colossal investment announcement and fresh leadership, could we be on the brink of a renaissance in Meta’s immersive tech journey? https://lnkd.in/gDjdNB-k #Meta #VirtualReality #AugmentedReality #XR #TechLeadership #Innovation

38

8 Comments

Snap OS June/AWE Update - AI 💛 AR 🧠 OpenAI, Gemini, and Snap-Hosted Open-Source Integrations - Get access credentials to OpenAI, Gemini, and Snap-hosted open-source LLMs from Lens Studio. Lenses that use these dedicated integrations can use camera access and are eligible to be published without needing extended permissions and experimental API access. 📍 Depth Caching - This API allows the mapping of 2D coordinates from spatial LLM responses back to 3D annotations in a user's past environment, even if the user has shifted their view. 💼 SnapML Real-Time Object Tracking Examples - New SnapML tutorials and sample projects to learn how to build real-time custom object trackers using camera access for chess pieces, billiard balls, and screens. 🪄 Snap3D In Lens 3D Object Generation - A generative AI API to create high quality 3D objects on the fly in a Lens. 👄 New LLM-Based Automated Speech Recognition API  - Our new robust LLM-based speech-to-text API with high accuracy, low latency, and support for 40+ languages and a variety of accents. 🛜 BLE API (Experimental) - An experimental BLE API that allows you to connect to BLE devices,  along with sample projects. ➡️ Navigation Kit - A package to streamline the creation of guided navigation experiences using custom locations and GPS locations. 📱 Apply for Spectacles from the Spectacles App - We are simplifying the process of applying to get Spectacles by using the mobile app in addition to Lens Studio. 🈂️  Translation Lens - Get AI-powered real-time conversation translation along with the ability to have multi-way conversations in different languages with other Spectacles users 🆕  New AI Community Lenses - New Lenses from the Spectacles community showcasing the power of AI capabilities on Spectacles: 🧚‍♂️ Wisp World by Liquid City - A Lens that introduces you to cute, AI-powered “wisps” and takes you on a journey to help them solve unique problems by finding objects around your house. 👨‍🍳 Cookmate by Headraft: Whip up delicious new recipes with Cookmate by Headraft. Cookmate is your very own cooking assistant, providing AI-powered recipe search based on captures of available ingredients. 🪴 Plant a Pal by SunfloVR - Infuse some fun into your plant care with Plant a Pal by SunfloVR. Plant a Pal personifies your house plants and uses AI to analyze their health and give you care advice. 💼 Super Travel by Gowaaa - A real-time, visual AR translator providing sign and menu translation, currency conversion, a tip calculator, and common travel phrases. 🎱 Pool Assist by Studio ANRK - (Preview available now, full experience coming end of June) Pool Assist teaches you how to play pool through lessons, mini-games, and an AI assistant. Read more on our sub-reddit (https://lnkd.in/gCUi2Hkk)

300

24 Comments