Patrick Chase

Sustained ~252–291 GiB/s of HBM memory throughput on H100s under decode load — essentially hitting the hardware roofline. This matters because HBM throughput, not FLOPs, is the real bottleneck in LLM inference. By keeping memory nearly fully saturated, I’ve unlocked far higher efficiency and throughput than standard engines. The result: 0.9–1.36M tokens/sec with ~0.1 ms first-token latency  #AI #LLM #GPURouter #H100 #Inference #CostEfficiency #Innovation #DeepLearning #AIInfrastructure #HighPerformanceComputing NVIDIA AMD OpenAI Google Shilpa Kolhatkar Keith Strier a16z speedrun

5

Trying something new this week - a studio industry roundup with my take on what matters. The studio model is no longer an experiment being run in a handful of places. It is infrastructure being built across geographies, sectors, and institutional types. 5️⃣ stories from the past seven days:  1️⃣ OSS Ventures hit a EUR 40M first close on a EUR 75M follow-on fund -- capital raised specifically to back the 30 companies their studio created since 2019. The "build then fund" lifecycle is maturing. B'More Venture Studio with a similar thesis: studio operations first, then dedicated capital to scale what's working.  2️⃣ Delta40 raised $20M from 54 investors across 13 countries to launch Africa's first integrated venture studio and fund. The LP base is almost entirely DFIs and foundations. Impact capital is discovering the studio model as a deployment vehicle, something I seeing across geographies (conversations with Michael Bob Starr and Damias McDonald this week alone) 3️⃣ Start Holdings launched Start.vc out of NYC and Barcelona after a $200M EV infrastructure exit. Another operator-turned-studio-founder. This is the pattern we see quite often in our work at 9point8 -- experienced operators like Jason Goldsmith who've built and exited deciding the studio model is how they want to build next.  4️⃣ UBC HATCH admitted 10 new deep-tech spinouts for Winter 2026 -- mostly climate tech. University venture builders are quietly becoming one of the fastest-growing segments. Follow Evan Allen and Jaimie Testai to keep track of its rapid evolution.  5️⃣ The INVEST Act is advancing through the Senate. If passed, it expands qualifying VC fund size from $10M to $50M and loosens general solicitation rules for accelerator and university events. For studio-funds -- and for folks in the The Venture Studio Forum community tracking policy -- this could be the most significant capital formation legislation since the JOBS Act. If this roundup format is useful, I'll keep it going. What other noteworthy news hit your desk this week, and what stood out to you?

13

1 Comment

Voice is a super interesting modality right now - maybe the first modality we're seeing move to open source models across a number of scale ups / enterprises. Reliability concerns, high costs, and open source model performance are pushing engineers to do their own fine tuning vs. relying on third-party vendors of proprietary models. Many of these orgs have already been collecting their own first-party data and now with third-party vendors like Extrian, David AI, etc they can train really high quality models. RL has been insanely hyped, but it's been unclear how long it will take scale ups and enterprises to actually lean in. Voice AI might be hitting that inflection point faster than expected.

45

7 Comments

GPUs have gotten faster, and models have gotten larger. But storage is still stuck in the past. Modern AI workloads need instant access to huge datasets. Instead, teams waste time on slow retrieval, idle GPUs, and costly pipelines. Archil is changing this. It delivers object-storage scalability at block-storage speeds. The platform allows for seamless, high-performance access to massive datasets. By eliminating cold starts and accelerating throughput 30X, Archil enables teams to fully utilize their infrastructure and move faster on training, analytics, and AI deployment. Archil’s founder Hunter Leath brings rare technical depth and operating experience from building Amazon EFS and optimizing Netflix’s cloud performance. This is a team building the foundational layer that AI workloads desperately need. I’m thrilled to lead Archil’s seed at Felicis with Nancy Wang and support their mission to reinvent cloud storage for the AI era. https://lnkd.in/gWueXhzz

65

2 Comments

Reinforcement learning isn’t the bottleneck. Infrastructure is. Most teams want RL. Few want to build the plumbing from scratch. 🚀 AgileRL Raises Seed Funding To Accelerate Reinforcement Learning at Scale London-based AgileRL closed a Seed round led by Fusion Fund, bringing total funding to $7.5M, as it pushes reinforcement learning from research labs into production systems. 🔹 Lead Investor: Fusion Fund 🔹 Participants: Flying Fish Partners, Octopus Ventures, Entrepreneur First, Counterview Capital 🔹 Total Raised: $7.5M Founder & CEO: – Param Kumar What They’re Building: AgileRL is building infrastructure for reinforcement learning, offering both an open-source framework and Arena, a managed, full-stack RLOps platform that removes the hardest engineering barriers to deploying RL in real-world systems. The stack includes: 🧠 Open-source RL framework with 250,000+ downloads ⚙️ Arena RLOps platform handling training, deployment, and scaling 🏗️ Production-grade tooling for complex RL workloads 🌐 Support across defense, robotics, finance, and simulation-heavy domains Used by: IBM, Airbus, Decision Lab Adopted by labs at: MIT, Roblox, Carnegie Mellon, University of Waterloo Why This Matters: Reinforcement learning has massive upside, but its complexity keeps it locked in research environments. AgileRL lowers the activation energy, turning RL into something teams can actually ship. What The Funding Fuels: – Opening a San Francisco office – Hiring across engineering and go-to-market – Scaling adoption of Arena as RL moves into production My Take - (Muhammad Asad): Most AI stacks optimize supervised learning because it’s easy. RL is harder, messier, and far more powerful. Whoever abstracts that complexity becomes critical infrastructure. AgileRL is positioning itself right in that gap. Follow Future Techly for high-signal coverage on AI infrastructure, ML systems, and deep tech. P.S. Repost to help more builders see where reinforcement learning is quietly heading.

30

1 Comment

Santa Cruz Works has announced a strategic partnership with Runpod, a leading GPU-first cloud provider, to empower early-stage AI startups with scalable computational resources. This collaboration aims to eliminate infrastructure complexities and enable founders to focus on innovation. The partnership provides startups in the Santa Cruz ecosystem instant access to high-performance GPU environments across eight global regions. Runpod's dynamic, pay-as-you-go model allows AI builders to train, fine-tune, and deploy models without significant upfront investments. @John Tran, COO of Santa Cruz Works, emphasized the critical importance of accessible infrastructure for developing world-class AI companies. By removing GPU access barriers, the partnership enables entrepreneurs to concentrate on building transformative AI technologies. Runpod's platform is already trusted by major AI innovators like Replit, Hugging Face, and OpenAI. Startups can leverage the infrastructure for diverse applications, including model inference, custom model fine-tuning, and intelligent workflow development. Zachary Gulsby from Runpod highlighted the partnership's potential to unlock Santa Cruz's deep AI talent and entrepreneurial energy. By providing scalable compute resources, they aim to accelerate the region's AI innovation ecosystem. This collaboration represents a significant step in building Santa Cruz Works' early-stage AI platform, positioning the region as an emerging hub for technological entrepreneurship.

6

1 Comment

From our PearX S25 cohort, meet Pravāh, the AI-native operating system for the electric grid When Mohak Mangal, Dhruv Suri, and Aman Gupta looked at the strain on the U.S. electric grid, they saw a system pushed to its limits. Exploding demand and unpredictable supply have made blackouts increasingly common. This costs the economy more than $150B a year, a number that could 10x in the next five years. Mohak Mangal and team founded Pravāh to help fix it. Their platform helps utilities and grid operators make real-time decisions on load, generation, and congestion, reducing blackout risks, optimizing power procurement, and bringing much-needed stability to the backbone of modern life. We’re proud to support the Pravāh team as they take on one of the toughest and most important infrastructure challenges of our time.

804

23 Comments

🚨NEW EPISODE: Recorded live at FUTURE TITANS 2026 - Jeff Perry of Carta sat down with the iconic Seth Levine, co-founder of Foundry. Seth has been in venture for 25 years, built Foundry from scratch as an emerging manager himself, and has backed about 50 emerging manager funds through his fund of funds. He has genuinely seen every side of this table. They went deep on building Foundry, why VCs are in the influence business, not the decision business, and why the concentration problem in venture is not only bad for LPs, but also for the innovation ecosystem overall. And why Seth's new book, Capital Evolution, is so important for the future of America. 🎧 Links to listen... Apple: https://lnkd.in/ehQUQ2EM Spotify: https://lnkd.in/eU4FExpg

19

1 Comment

BaseFrame (YC W26) helps hardware teams find mistakes in their project designs and explore tradeoffs early. Anshul and Vaibhav were roommates at Berkeley and faced these issues themselves when working on robotics startups in the past. Their vision is to help hardware teams bring breakthrough products to market faster, leaner, and with data-driven compromise. To make this work, BaseFrame’s specialized AI agents validate your scope against industry best practices to surface gaps and areas for improvement. They then generate a Bill of Materials by parsing datasheets, manufacturer data, and past designs made by the team and give comprehensive comparison tables and citations. No more 15 tabs of datasheets clogging your workflow. Teams can branch designs, manage versions, and test multiple architectures simultaneously to highlight tradeoffs in cost, price, and lead time. Congrats Anshul Paul and Vaibhav Agrawal on the launch! https://lnkd.in/e6xnmnm2

351

71 Comments

Bengaluru and San Mateo-based Refold (formerly Cobalt) AI raised $6.5M in seed funding, co-led by Eniac Ventures and Tidal Ventures, to expand its engineering team and enhance product integrations. Founded in 2023 by Jugal Anchalia and Abhishek Kumar, Refold AI automates API integrations using AI, targeting the $350B enterprises spend on manual processes. Their AI agents learn system behavior and maintain integration code, replacing traditional middleware. The platform manages workflows like ERP-to-CRM syncing and finance automation, promising reduced manual effort and improved accuracy. Previously, the founders led JustDoc, acquired by Netmeds. Read Full Article: https://lnkd.in/gpAwCHVM #AI #EnterpriseAI #Automation #RefoldAI #TidalVentures #EniacVentures

47

I’ve been working on the theory of why small, shippable factories are the perfect inversion to the centralized prefab factory model. It all collapsed into one line of math (credit to Gilles Retsin): Factory Efficiency = Usage × (Throughput × Product Value ) / ( CapEx + Deployment OpEx) The idea is that the closer you can drive Usage to 100 % while keeping the denominator tiny, the faster the flywheel spins.  A shippable microfactory tackles this by primarily focusing on a high Usage factor. Because the microfactory can be packed up afterwards and reused, its CapEx gets amortized over multiple projects, effectively raising its overall Usage across the year. If one project alone doesn’t fully occupy the factory, it can simply roll over to another job to maintain high Usage. Its CapEx is often an order of magnitude lower than a centralized factory (< $1M vs. $10-50M min), so the breakeven throughput is more achievable on a small pipeline of work. The real breakthrough here is portability. There is near-zero stranded capital and almost no idle time. To flesh this theory out, I wrote 3000 words on the topic for Brad Hargreaves and Thesis Driven. The article breaks down: - why “fixed factories, shipped goods” is being inverted to shipped factories, fixed goods - what a sub‑$1 M robotic cell does to CapEx per home versus a £45 M off‑site plant - early production/cost data on a shippable microfactory from Mollie Claypool, Gilles Retsin, Sam Baker and the Automated Architecture (AUAR) team - the hybrid reality of robots tackling the heavy structural components and human crews assembling faster and cheaper - technical constraints to pull this model off and future considerations on the optimal business model (own vs. rent) Major shoutout to Gilles Retsin for helping me pull this together. https://lnkd.in/eRTnbZPa

54

2 Comments

Mem0, a San Francisco-based AI startup founded by Taranjeet Singh and Deshraj Yadav, has raised $24 million across Seed and Series A funding rounds. The Seed round was led by Kindred Ventures, while Basis Set Ventures led the Series A, with support from Peak XV Partners, GitHub Fund, Y Combinator, and angel investors like Scott Belsky and Dharmesh Shah. The company is developing an advanced "memory layer" that allows AI systems to retain and recall information over time, improving personalization and long-term learning.

5

🚨 Mem0 secures $24M funding to power the memory layer of AI Mem0 has raised $24 million in a Series A funding round led by Basis Set Ventures, with participation from Peak XV Partners, Kindred Ventures, GitHub Fund, and Y Combinator. 📍 Founded by Taranjeet Singh and Deshraj Yadav in 2023, and based in San Francisco (USA), Mem0 is building a “memory layer” for AI agents, enabling applications powered by large language models (LLMs) to persist context, preferences, and interactions rather than start each session from scratch. 📈 The fresh capital will be used to expand the engineering team, support enterprise-grade use cases, deepen integrations across AI platforms and tooling ecosystems, and scale Mem0’s infrastructure globally. • The product boasts over 41,000 GitHub stars, 14 million+ Python package downloads, and API calls growing from 35 million in Q1 to 186 million in Q3 2025. • Mem0 is integrated into frameworks like CrewAI, Flowise, and Langflow, and is the exclusive memory provider for Amazon Web Services' new Agent SDK, a strong indicator of infrastructure-level positioning. .......................... 🔎 Curated by Investorsync, where capital meets relevance 💡 Powered by #AskBinny #Investorsync #AIinfrastructure #BasisSetVentures #IndiaStartups #AskBinny

7

Mithril (formerly Foundry) rebrands and unveils a powerful batch inference API, slashing AI compute costs by up to 5x for startups and enterprises alike. Read the Latest Full News - https://lnkd.in/dAgUBdBN #MithrilAI #AICompute #BatchInference #GPUMarket #Omnicloud #CloudInfrastructure #DistributedAI #AIStartups #TechEdgeAI #TechEdge #InferenceOptimization #GenerativeAI

3

The next frontier of AI isn't just about choosing the right model. It's about building moats through implementation. Across the Emergence Capital portfolio, companies like Bland and Federato are scaling faster by embedding Forward-Deployed Engineers. These engineers bridge the gap between AI and reality, ensuring real results in production. Read more from my colleagues Gordon and Kyle on why FDEs are the key to unlocking AI's full value. https://lnkd.in/gcsRRGX2

29

3 Comments

NVIDIA is partnering with Activate Fund to back AI startups across the country. Announced at the India AI Impact Summit 2026, the partnership will see Nvidia's expertise in AI hardware, accelerated computing and developer ecosystems combined with Activate Fund’s early-stage capital and founder-first investment model. Aakrit Vaish, Founder of Activate Fund, told Economic Times, “Any time we invest in a company, the startup will get direct access to NVIDIA teams and proprietary models. We will also have early access to certain product releases before they are launched publicly.” Read the full story here: https://lnkd.in/g94x-6jy #IndiaAIImpactSummit #Activate #Nvidia #AIStartups Jensen Huang

236

8 Comments

Lots of news out of Hetz Ventures portfolio today - Nimble announced their $47M Series B, led by Norwest with participation from Databricks Ventures. We've been proud to back Uriel Knorovich, Menachem Salinas and the entire Nimble team since early on, turning the live web into trusted, decision-grade data for AI agents and mission-critical workflows. Tonic Security launched the Tonic Mobilization Coordinator, the industry's first agentic remediation orchestrator. It automatically drives exposure remediation end-to-end, moving security teams from knowing they have risk to actually reducing it, continuously and at scale. Well done Sharon Isaaci, David Warshavski and Gregory Ainbinder. Anima launched their UX Design Agent on Product Hunt today: https://lnkd.in/demUMrUX. Designers and product teams can now go from prompt, Figma file, or live website to on-brand, production-ready frontend code in a single shot. Deepchecks released Know Your Agent (KYA), their answer to evaluation in the agentic era. Borrowing from the KYC framework in financial services, it gives teams a structured way to understand how their agents actually behave across complex multi-agent workflows. #proudseedinvestor https://lnkd.in/dyGTAHcg

134

7 Comments

Plakar Raises $3M Led by Seedcamp to Advance AI Backup Solutions Key Points 👇 ❶ Plakar raises $3M in pre-seed funding led by Seedcamp, supported by helloworld, Irregular Expressions, Galion.exe, Kima Ventures, OPRTRS CLUB, Motier Ventures, Purple, Pareto Holdings, Osorno Capital, and angels Olivier Pomel, Solomon Hykes, Alexandre Yazdi. ❷ Funds will enhance engineering capacity and scalability, targeting AI-focused enterprises. ❸ Plakar transforms backups into actively useful data resources, optimizing storage and enhancing productivity. “With Plakar, we’re providing an open-source solution that simplifies backup and restore processes while making enterprise-grade data resilience accessible to businesses of all sizes. This product launch and funding announcement is a significant milestone, and we are thrilled to begin this journey with the support of such renowned partners and investors.” - Julien Mangeard, co-founder and CEO of Plakar. “In an era where AI systems are generating and processing unprecedented volumes of data, ensuring the integrity and security of that data is paramount. Plakar’s approach of combining open-source flexibility with enterprise-grade features aligns with our belief in empowering developers and businesses with accessible, reliable, and efficient tools.” - Sia Houchangnia, Partner at Seedcamp. #StartupFunding #OpenSourceTechnology #AIDataManagement #Seedcamp #CloudInfrastructure https://lnkd.in/ehgT9tVH

11

1 Comment

Very proud to welcome Ardent AI to the Crane Venture Partners herd and lead their Pre-Seed round! I was introduced to Vikram C. in April. He’d been sleeping on a friend’s couch, had released an alpha, and convinced real life customers to pay money for a product that had shipped pretty much just before they signed up. By himself. After being rejected by YC multiple times. Technical founders can’t usually sell anything when they first start. Technical buyers don’t usually pay material money for alphas made by sole founders without venture capital. Here’s was an exception. In the middle of our opening discussions, Vikram closed a $60k deal. How? Why? * There is more data engineering to do than there will ever be data engineers * Data infrastructure needs constant care and feeding, because no business is static * Data is the life blood of AI Today, Ardent is the first AI that can create a fully operational data pipeline in one shot. Tomorrow, Ardent will be the data engineer that stands up your end to end data infrastructure. Soon, Ardent will be your entire data engineering team—proactively building, fixing, and upgrading your data infrastructure for you. 🚀 🤖 🚀 #ai #dataengineering #startups

40

4 Comments