Skip the demo. Build it yourself. 🛠️ In these technical workshops, you'll get hands-on experience with AIStor + Databricks, connecting on-premises data to analytics without moving or replicating it. Two hours. Your own lab environment. A working, secure table share. 📅 November 12 or 18 | Limited to 40 participants per session Register now: bit.ly/3WBy0IR

Agentic AI is changing what GPUs need from memory infrastructure. Using @NVIDIA Nemotron 3 Ultra on NVIDIA H200 GPUs, our engineering team tested what happens when context can persist beyond GPU memory, and what that means as concurrent agentic workloads scale. The results point to a new way to think about context memory for enterprise AI. bit.ly/3W2NzZV

What if security agents could reason across relationships without moving data? @puppyquery + AIStor enables graph queries and SQL analytics on the same data, giving AI agents grounded context without a separate graph database. ➡️ Here's a tutorial: bit.ly/4zvDlj7

3️⃣ steps. No data movement. AIStor connects on-premises data directly to @databricks with native Table Sharing, keeping a single copy of your data in place and ready to query. Go deeper in our upcoming training webinar to move from setup to query. 🔗: bit.ly/4rCLRKx

Congratulations to AB and all the innovators recognized by H2O.ai for advancing the future of AI! lnkd.in/d4QeUpgY

Tomorrow in Dallas, get hands-on with AIStor + @databricks. We're at #DataAIWorldTour to demo native Delta Lake and Iceberg support, production-ready authentication, zero-downtime administration, and data protection, all from a single AIStor binary. 📅 Oct. 14 | 1:00–1:40 PM CDT 🔗: bit.ly/4z6TUle

Amazon’s acquisition of DuckLabs reinforces where the lakehouse is headed: query data where it lives. That’s the architecture MinIO is built for—a high-performance data foundation for open analytics engines and table formats like DuckDB and Iceberg. bit.ly/4eeVsBF

Prompt caching works until reusable KV state is trapped on one GPU. At production scale, that state needs to move across workers and nodes. This is the architecture behind shared context memory. Keep reading: bit.ly/4i5kYeZ

What does the data architecture for production AI agents actually look like? Join MinIO and @lakeFS to dive into agent memory, data isolation, reproducibility, and governance + how to build those capabilities into the infrastructure layer without slowing development down. 📅 10/22 | 12 PM ET 🔗: bit.ly/3WbJB15

Every GPU dollar should produce useful work. When GPUs repeatedly process context they’ve already seen, you’re paying for duplicate compute. Prompt caching changes the economics: reuse context, reduce recompute, and get more from every GPU dollar. bit.ly/47Br1Sz

Faster queries. Single-copy cost savings. Stronger access controls. Data under your control. Get the blueprint at @databricks #DataAIWorldTour Dallas with product experts Dwight Evers and Denis Debeau. bit.ly/4z6TUle 📅 Oct. 14 | 1:00–1:40 PM CDT Not in Dallas? Stay tuned for insights from the session.

A cache should be faster than what’s behind it. CoreWeave LOTA warm cache: 18.4 GiB/s per node. MinIO AIStor, no cache: ~33.5 GiB/s per node from durable, erasure-coded storage over plain TCP. This is what CoreWeave’s benchmark leaves out—and the number every cache benchmark should publish: bit.ly/4jyqhEY

What does it take to build AI at scale? Start with data. As AI moves from models to production, the underlying infrastructure matters more than ever. More from co-CEO Garima Kapoor on the data foundation powering enterprise and sovereign AI—and what it takes to build for what comes next. bit.ly/4iWRfGf

Your data can stay on-prem. Your analytics don’t have to. Join us on 10/22 for a hands-on look at AIStor Table Sharing and how enterprises can securely connect on-prem Delta and Iceberg tables to @Databricks. ✍️ Register to see it in action: bit.ly/4rCLRKx

From Saudi Arabia to the world. #LEAP26 brought together the people building the future of AI. For MinIO and @HUMAIN, it marked a major step in building the data and memory foundation for sovereign and enterprise AI at a global scale with HUMAIN Fabric. Thank you to our team and everyone who joined us in Riyadh. See you next year.

Agentic sessions build context, go idle, then return after KV cache may have been evicted. We benchmarked MemKV around that real behavior: human-paced pauses, GPU caching left on, and context offloaded when needed. bit.ly/46wQLPI

Three days of great conversations at #AIInfraSummit with @NVIDIAAIInfra. The takeaway: GPUs shouldn’t be waiting on data. The data and memory layer needs to keep them productive as AI workloads scale. Thanks to everyone who joined us last week.

Repeated prefill burns GPU cycles on context the model has already processed. Prompt caching changes that equation. As context windows grow, how you cache and reuse KV directly affects inference performance and GPU efficiency. ➡️ We’re starting with the fundamentals. bit.ly/46BAA3w