About
A Managing Director at DTC, Michael Dell's early-stage VC, focusing on AI disrupters…
Activity
7K followers
-
Yair Snir shared thisAt the upcoming EY JOURNEY, I'll be joining Amnon Shashua, Founder & CEO of doubleAI, on stage to talk about deep reasoning, how far we are from AI systems that can develop true expertise, and what that will mean for the next frontier of AI. Looking forward to this discussion and a full day of engaging content at JOURNEY 2026.
-
Yair Snir shared thisProud to join the TECH1 Awards 2026 jury. I’m looking forward to reviewing the people, companies and achievements shaping the next chapter of Israeli tech. Know a founder, executive, company or breakthrough that deserves recognition? Nominations are open: https://tech1.co.il/awards [tech1.co.il] The winners will be announced at the TECH1 Gala on October 29 at Suzanne Dellal, Tel Aviv.
-
Yair Snir reposted thisYair Snir reposted thisCongratulations to AITAN and Terra Security for being named to the 25 Most Promising Startups 2026 list by N12 News Israel, Valley Bank, and LeumiTech. We couldn't agree more!
-
Yair Snir shared thisWhile the common AI focus is on models, tokens, and benchmarks, I’m actually focused on what happens when AI leaves the screen. My belief: the largest AI companies of the next decade will be built in the physical economy. Manufacturing. Logistics. Infrastructure. Energy. Defense. These industries represent trillions of dollars in economic activity, persistent labor shortages, and operational challenges that still lack technological solutions. Software was the first proving ground for AI. The physical economy has the potential to be the much larger opportunity. I shared more thoughts on the tech driving this shift this with CTech by Calcalist *link in the first comment
-
Yair Snir posted thisLightbits Labs introduced a disruption today with Inferra. 🚀🚀🚀 A foundational idea in infrastructure is that not all data needs to live in the fastest, most expensive layer of the stack. AI inference is reaching the same conclusion in what might be record time. Thinking of KV cache as a tiered data challenge as Lightbits is with Inferra could unlock big gains in GPU utilization with models that are substantially more efficient and economical to run. Compute constraints are our reality. For AI adoption to continue to scale across the enterprise at its current pace, changing the economics of inference is non-negotiable. Today’s announcement is exciting and exactly the type of innovative thinking Eran Kirzner, Avigdor Willenz and the Lightbits team consistently deliver.
-
Yair Snir posted this500,000 operational hours is a number that got our attention. 🚀 I don’t think I’ve met any other early-stage physical AI company with tech that has been tested and used that as extensively in real-world environments as AITAN’s has. Co-founders Udi Oster and Daniel Almog saw early on that the future success for drones, robots, and autonomous systems will be defined by how flexible they are to scale and how well they work together. In other words, they started AITAN to create the orchestration layer for Physical AI. Every major computing wave created a new software category. Physical AI will be no different. As autonomous systems proliferate, coordination, interoperability, and fleet management become the bottlenecks. AITAN is removing that complexity and is already helping to define the next generation of infrastructure. Dell Technologies Capital is proud to back Udi, Daniel, and the AITAN team co-leading the Series A alongside Deep33 and a group of investors who share our conviction in the future of Physical AI. Link to the PR in the comments
-
Yair Snir shared thisAI systems that can improve themselves and verify their own work are transformative. doubleAI's WarpSpeed just posted state-of-the-art results on NVIDIA's hardware design benchmark, solving 98.5% of evaluated design problems while maintaining formal correctness guarantees. Congrats to the team for continually pushing boundaries to bring expert-level AI to some of the world's hardest engineering challenges. 🚀 🚀 🚀Yair Snir shared thisDesigning a chip usually takes years, hundreds of specialists, and tens of millions of dollars. WarpSpeed, our AI expert, just reached state-of-the-art performance on NVIDIA's CVDP benchmark for chip design, surpassing Claude Code with Fable. The benchmark tests whether an agent can turn a specification into a real chip design, verify it, and optimize it. WarpSpeed solved 98.5% of the design tasks, ahead of every system it was measured against, and set a new record on optimization. Beyond the benchmark, it also autonomously designed a cryptographic chip and a lane-detection chip for cars. It's the same self-improving system behind our recent pandas, GPU kernel, and quantum circuit design results. The part that matters most: WarpSpeed verifies its own work. In the benchmark, many other AI systems scored well by hacking the evaluation and passing tests with designs that were actually wrong. WarpSpeed had 0% hacks, compared with 16.2% for Claude Code. Read the full technical blog: https://lnkd.in/dnnuyNqk
-
Yair Snir shared thisCongratulations to lakeFS for being named a Gartner Cool Vendor 🏆 🙌 Dr. Einat Orr, Oz Katz and the lakeFS team recognized early on that data versioning, governance, and reproducibility would be critical for the enterprise. Their approach of bringing software engineering versioning principles to data has only become more relevant as AI adoption takes off. Enterprise-scale orgs can ship AI workflows with confidence using lakeFS. Proud to be able to support this team at DTC! Dell Technologies CapitalYair Snir shared thislakeFS was named as a Cool Vendor in the 2026 Gartner Coolest Vendor Innovations in Data Management! We are honored! Here's the context that matters: enterprises are rushing to deploy AI agents for building pipelines, monitoring data quality, and documenting lineage. But per Gartner, most enterprise data isn't organized, contextualized, governed, or automated enough for autonomous agents to actually act on it. Testing, reproducibility, traceability, and version control are quietly becoming the bar for shipping AI safely, not nice-to-haves. Dr. Einat Orr, lakeFS CEO and co-founder: "lakeFS closes a critical gap in AI data management by supporting reproducibility, governance, and trust while increasing efficiency in AI agent and model delivery." We work with teams at Arm, Bosch, Lockheed Martin, NASA - National Aeronautics and Space Administration, Volvo Group, and the U.S. Department of Energy (DOE), and this recognition lands right after our recent work on AI-ready data, AI data governance, and agentic AI. Free copy of the report here at the link in comments. 👇🏼 #DataVersionControl #AgenticAI #MLOps
Yair Snir commented on a post
4w
https://www.calcalistech.com/ctechnews/article/b1dz1kbdme
-
Yair Snir liked thisYair Snir liked thisI am beyond thrilled to announce that we have closed an $18M Series A led by Dell Technologies Capital! Also joining in the round were Hitachi Ventures and our previous investors Wipro, BGV, FUSE, Tola Capital, Gaia Ventures, and more. The market is full of meaningless vanity metrics. Companies are counting tokens, lines of generated code and agent runs, and struggling to work out the impact AI has on their business. Ascerta is designed to cut through the noise to deliver meaningful insights to leaders looking to optimize their business for an AI-first world. This is an emotional milestone. When we started on this journey 3 years ago, it was a side project while we were still employed at our other jobs. As the scale of the problem started showing itself, we set out to tackle it in earnest and... its been the most exhilarating and challenging journey of my life to build a team and a company capable of staying on top of the rush of AI innovations while staying ahead of the market. As I write this, AI spending is the fastest growing line item in enterprise history, and the curve is just starting. Business has changed forever and Ascerta is, by far, the world’s most capable platform to rise to the occasion. Today we welcome Dell and Hitachi into the fold. Their combined connections to enterprises the world over will help us enable enterprises to unlock the full potential of AI. I would like to especially thank Raman Khanna from Dell Technologies Capital and Gayathri Radhakrishnan from Hitachi Ventures for their foresight and support. The gloves are off. Time to start Round 2.
-
Yair Snir liked thisYair Snir liked thisSoon 100% of American children will automatically have their own 530A Trump Account, even if they dont know about it. 👶 🗽
-
Yair Snir liked thisYair Snir liked thisA few years ago, when a production table broke, the first question was "who ran that job?" Usually someone raised their hand. With autonomous agents, nobody raises their hand. The agent made hundreds of changes while we were in a meeting, and it will happily make hundreds more. We've been here before, in a way. In software, we stopped relying on careful people never making mistakes. We built version control, isolated environments, and the ability to roll back. Agents need the same thing for data, and I think they need it more, because they move faster than any review process we designed for humans. Governance for agents needs two histories: what the agent did, and what happened to the data. I suspect many teams only have the first one today. Oz Katz, my co-founder and our CTO, is talking about exactly this on October 22 with Ravind Kumar from MinIO. They'll cover where agent memory should live and where governance belongs in the architecture, and they'll show a live demo on MinIO AIStor and lakeFS. Link to the event registration is in the comments below.
-
Yair Snir liked thisYair Snir liked thisIn 2023, VAST Data and CoreWeave started with 10 PB across two data centers. Today, that partnership powers over 1 Exabyte of data infrastructure across 32+ global data centers. When training frontier models and running agentic AI at scale, compute efficiency is everything. By coupling CoreWeave’s flexible GPU cloud with the VAST AI OS, enterprise workloads run on a zero-copy architecture designed to feed GPUs at full speed—eliminating idle cycles from training to inference. If you are in San Francisco this week for CoreWeave Fully Connected, see this architecture in action as VAST Co-founder & VP of Technology Alon Horev presents "Data Architecture for AI: From Training Pipelines to Agent Memory" to learn how storage, metadata, and KV cache architectures must evolve across the AI lifecycle.
-
Yair Snir liked thisYair Snir liked thisOn a hard problem, an AI's first answer is rarely its best. The real gains come from letting it try many times and keeping the best one. But most training collapses a model onto a single answer, so you can ask a thousand times and get the same one back, throwing away one of the most powerful tools these models have. Our team just published ArgMaxRL, a training method that keeps that variety alive, so a model produces many strong, different options instead of collapsing onto one. It builds on MaxRL from Andrea Zanette's lab at CMU, which showed how to do this for pass-or-fail tasks. We extended it to continuous tasks, where quality is a scale rather than right or wrong, like code that runs 1.2x versus 8.5x faster. And it stays simple: a few lines of code you can drop into an existing training loop. This is the research underneath WarpSpeed, our system that generates many candidate solutions and keeps the best, whether it is rewriting pandas, writing GPU kernels, or designing chips. Full technical blog - https://lnkd.in/dHBZQBJG
-
Yair Snir liked thisYair Snir liked thisMost data governance was designed around a person doing the work: someone who moves at human speed and leaves a trail you can read afterward. Autonomous agents change that. They touch data far more often than any approval process expects, and when an agent gets something wrong, one audit log isn't enough. You need two histories: what the agent did, and what happened to your data. On October 22, lakeFS CTO and co-founder Oz Katz and Ravind Kumar from MinIO will walk through what it takes to run agents in production, and how enterprise teams in regulated environments are building that foundation now: → Why human-speed governance can't keep up once agents become a new type of data consumer → Why agent memory should live on infrastructure you control → Where to put governance so you get isolation, reproducibility, and guardrails without slowing teams down → A reference architecture for agentic AI on enterprise data → A live demo of governed agent workflows on MinIO AIStor and lakeFS 📅 October 22 | 12:00 PM ET | Virtual Save your spot at the link in comments 👇🏼
-
Yair Snir liked thisYair Snir liked thisAfter a long run at Intel Capital, I have decided to move on and do something different. I plan to stay in the industry and will make an announcement at the appropriate time. In the meantime, I will continue to work closely with great companies like Beep, Inc., LILT AI and MinIO. Of course, I continue to talk to other founders to explore ways I can help. Thanks to the great team Intel Capital and, in particular, to my great friend and partner, Anthony Lin (林立中) with whom I worked so closely. Thanks also to Andy Fligel and Nick Washburn for their help in creating such a fun and productive team. Best of luck to all of you!
-
Yair Snir liked thisSiMa.ai's founder and CEO Krishna Rangasayee on today's Series C announcement.Yair Snir liked thisSeven and half years ago, we bet that AI at the edge needed its own silicon, not a shrunk-down version of what powers data centers. Today that bet gets bigger. Thanks to our investors we just closed $150 million in Series C funding to go straight into what’s next: a platform that hits 1,000 dense TOPS, doubles the performance of today’s leading GPU, and does it all under 80 watts. Take it as IP, a chiplet, or a finished chip. It’s the same software stack underneath, purpose-built for Physical AI from day one. Our platform already ships in volume across robotics, automotive, and drones. This is the leap that takes us deeper into humanoids, high-end drones and the hardest ADAS problems out there. The foundation is proven. The software is ready. This is the industry’s best, your way. Read the full announcement here: https://lnkd.in/gA4Cf6eA #PhysicalAI #Semiconductors #Robotics #Automotive #Drones #AI #SeriesC #SiMaAI
Experience & Education
Publications
-
Zinc-ion binding and cytokine activity regulation pathways predicts outcome in relapsing-remitting multiple sclerosis
Clinical & Experimental Immunology
Explore more posts
Explore collaborative articles
We’re unlocking community knowledge in a new way. Experts add insights directly into each article, started with the help of AI.
Explore More