Victoria Levy

The Snowflake Resale Program is here. 🌐 Partners can now deliver the full AI Data Cloud experience—one partner, one platform, no complexity. See how this evolution accelerates customer innovation: https://bit.ly/4gm2pRA

204

2 Comments

Snowflake Openflow has been costing me $10/day to sit idle with all run times suspended. The Snowflake managed "deployment" compute pools can't be suspended and don't show with the `show compute pools` command. They bill under Cloud Services, not compute pools from what I'm seeing. $3,600 per year won't break the bank for an enterprise, but don't do some experimentation and just leave that thing on!

109

21 Comments

I’ve never seen a keynote pack in so many product updates. Snowflake Summit Day 2 delivered! Here’s the short version of what was announced: dbt Projects inside of Snowflake! This includes a partnership with dbt Labs enabling the new Fusion Engine in Snowflake. Workspaces: A new IDE that connects to Git providers, enabling end-to-end development in the Snowflake UI. Openflow: load data from sources, including realtime CDC for many sources. (Fivetran killer). End-to-End data pipeline development. The problem of the modern data stack is tool proliferation. Back in the day of SSIS, one tool did it all. Finally coming full circle, Snowflake will do it all. Snowconvert: AI enabled data migration services from legacy systems. FREE for all customers. Cortext AI SQL: new SQL syntax allowing natural language queries: ai_filter, ai_agg, as_summarize_agg, ai_classify ai_transcribe ai_embed, ai_similarity. Classify photos, audio, other unstructured data. Semantic Views: new object type that enables agents to work more accurately. Cortex Knowledge Extensions: Vectorized knowledge ready to be queried by Cortex Search. Cortex agents in MS Teams. Snowflake Intelligence: combination of best models, agentic, reasoning capable, enable business transformation and productivity, public preview very shortly. ai.snowflake.com. AI will inspect Tableau to help you make a semantic view in Snowflake. Tag based budgets Adaptive Compute - similar to BigQuery, you will no longer need to think about warehouse sizes. Just run your query. Snowpipe pricing simplified: volume based. 50% better economics. Snowflake Horizon Catalog: full data catalog that enables you to discover assets outside of Snowflake. Postgres, mysql, Oracle, Databricks, Tableau, PBI, dbt, Airflow. Marketplace improvements: workflow to request access to datasets. AI Agents in Marketplace.

236

26 Comments

We tested Snowflake's Gen2 warehouses on 2 billion query profiles across 14K warehouses here at Capital One. Here is where we saw improvement: 🍎 Approximately 30% of queries fell into categories that could benefit from Gen2—specifically those involving DELETE, UPDATE or MERGE. 🍎 Long-running DML operations: A significant subset of queries running longer than 30 minutes were performing complex DML tasks. These are prime candidates for Gen2 migration. 🍎 Large scans: Numerous queries scanned over 1 TB of data, where Gen 2’s improved scan performance can provide material runtime reduction. 🍎 Combined heavy patterns: A smaller but critical set of queries both scanned over 1 TB and ran for 30 minutes or more, often involving DML operations—these represent the highest potential gain from Gen2. These findings support a targeted migration strategy: focus on warehouses with a significant concentration of eligible workloads rather than migrating all warehouses indiscriminately. Check out the full blow written by our resident expert and all around awesome guy - Ganesh Bharathan https://lnkd.in/g8XufZaV

292

14 Comments

🔍 Had a great discussion with Jeff Dickey from NetApp at AppWorld 2025! Data is even more important than ever. AI wants lots of it. It wants it everywhere and it wants it fast. What’s struck me for a while now is that even though we have all sorts of new breakthroughs, we keep coming back to fundamentals and foundational technologies. 🧱 NetApp is one of those foundational technologies and so much of the world’s enterprise data resides on NetApp. They’ve solved for the problems that people who are from the ML/AI space that are just running into the issues with. Replication. Security policies. Data residency policies. All solved already. 💁‍♂️ But now, it’s time to reach out and help. Those AI projects that have been moving from test/dev and towards deployment - it’s time to put them on solid foundation so that not only do we deliver business value quickly, but it’s ready for F5 to help transport the data to where it needs to go in a secure way. NGINX F5 DevCentral Scott Laster Rusty Young Haiyan Song Yuichi Miyazaki Steve Gorman

17

3 Comments

I hate overpaying for things. It is why Costco is my favorite store and I never live more than 10 mins from a Costco. The same reason I love Costco is why I joined MemVerge. We develop software based solutions for technology leaders, HPC professionals, and developers who also hate overpaying for things. Specifically, these are the pain points we see and help customers deal with daily: 1. Stop running batch jobs using on demand instances unless you absolutely have to or there isn’t enough Spot inventory in your region. That’s like paying vending machine prices for the same bag of chips you can buy for 1/10 the price at Costco. 2. Don’t buy or rent another GPU unless you have figured out how to get higher utilitization. Users who reserved GPUs for projects and aren’t using them are clogging up your cluster’s resources. How much time would you save if you could burst and use reserved resources from all idle users in the cluster? Using 3rd party API providers for LLM inferencing in Prod will likely eat into your ROI and cost multiples vs. learning how to efficiently host and manage your own inferencing tech stack. It also seems like the smaller guys get throttled all the time as these providers prioritize their biggest users/customers. Utilize GPU sharing and fractionalization techniques so users who don’t need an entire H100 or H100 node (8x) aren’t overprovisioned. Did you know a single H100 node can support up to 56 users or Jupyter notebook sessions? Look up NVIDIA MIG or the Instaslice project h/t to Abhishek Malvankar — Don’t shrug these issues off. It won’t be a good convo when your CEO or board finds out how inefficiently you’ve been allocating precious AI project capex and opex. In Generative AI where are people overpaying today? What can they do about it? Please share a comment below!

13

2 Comments

“5 minutes could save you 10% or more — on your Snowflake costs.” No, its not Geico ad, but its my latest blog on how you can build Snowflake Cost Savings and Performance Agent in 5 minutes. https://lnkd.in/gm9XbEbR It highlights how you can quickly deploy an AI-powered agent using Snowflake Cortex, monitor warehouse performance and cost, apply governance controls automatically, and deliver actionable recommendations using Snowflake Documentation and best practices, Query History and Snowflake Intelligence! What stands out is how AI and governance now work hand-in-hand — reducing operational overhead while improving visibility and compliance across data platforms. #Snowflake #AI #DataGovernance #DataEngineering #cortex #costoptimization

80

3 Comments

Going through three months of Databricks product updates after paternity leave is no joke: Here are my top 5: Databricks One -- if you didn’t know by now, Databricks has a full blown BI tool that competes with the best of them. We now offer a single pane of glass where you can see all of your BI content, as well as ask questions about all the data in your ecosystem and get answers to those questions. Data Science Agent -- ML is still by far and away being used in more production applications than LLMs today. Now you can build and deploy these models using Natural Language. Reliable ML models built with AI. Embedded Dashboards for external users -- my personal favorite. If you are a SaaS company that wants to embed dashboards… you can now use Databricks (and we don’t charge p/user) Unity Catalog + Apache Iceberg -- full interoperability with open-table formats. Every enterprise I’m talking to is moving to a Lakehouse architecture (which Databricks invented). Databricks makes it easy. Integrated OpenAI models -- GPT now available across the Databricks platform for all customers. Use your production data + OpenAI models all governed by Unity Catalog. Oh, and while I was out, Databricks crossed $4B ARR and is now valued at $100B. This is not even a tenth of what was released… not bad for a few months work. 

168

11 Comments

TL;DR: Context Matters. This summer at NetApp, I set out to learn not just what people know, but why they know it—how tiny decisions ripple out into game-changing breakthroughs. Imagine spinning up a workload across "64,000 individual CPUs" (64K threads actually). That’s the rough equivalent of wrangling 8,000 laptops writing and reading from disks…all to ask: Did we really need to go that big? As a lifelong tinkerer (I’m proudly a homelabber), I love experimenting with switches, servers, and every bit of infrastructure I can get my hands on. Yet when I dove into understanding Proof-of-Concept (PoC), I discovered it’s not just code and hardware—it’s a delicate dance of hardware, creativity, and context. That’s where neto from Brazil, PhD came in. A true NetApp legend and one of the brightest minds I’ve met, he took a “strange request” and turned it into a PhD exploration of contextual models and knowledge networks in PoC. Watching his sessions at NetApp Insights, it’s clear: he’s the GOAT. Over coffee, Neto shared how reframing a single requirement unlocked whole new dimensions of thinking—and reminded me that our understanding is shaped by both our exposure and our interpretation framework (yes, like expert systems vs. neural nets!). Key takeaway: Before jumping into solutions, map out the multidimensional “why” and “what” you actually have on your hands—because context isn’t just background, it is the solution. For a deeper dive into how context has transformed my own approach (and why it could transform yours), check out my full reflection on the blog: https://lnkd.in/dyc3Pp2G Lesson learned; Context matters no matter what. neto from Brazil, PhD thank you again for everything and you got one more fan to your endless list! #ContextMatters #ProofOfConcept #LearningJourney #NetAppInsights #TechLeadership #Homelab #Innovation

35

9 Comments

12 years ago, if you walked into an Oracle or Teradata product meeting with a smarter optimization idea, it was an easy decision: faster query execution meant happier customers, and happier customers meant the vendor would sell more software licenses, which was the main revenue model. A win-win situation. That dynamic began to change about 12 years ago as data warehousing started moving to the cloud. Cloud databases now earn bulk of their revenue from renting you servers (compute). They don't charge you separately for their software as it's all built into their hourly rent! But that has misaligned the vendor and customer incentives. New optimizations and performance gains are still somewhat helpful in winning and keeping customers from the competing vendors but CDW vendors are very careful to ensure when they release new optimizations they don't reduce their own margin! Snowflake Gen 2 warehouses are the most recent example of this: roughly 25 % faster runtimes arrive with a comparable increase in the unit costs (35% more expensive per hour in AWS and GCP and 25% in Azure). So please don't assume switching to Gen2 is going to automatically reduce your Snowflake bill overnight or give you a turn-key magic that speeds up your queries overnight for free. Whether Gen 2 is a net benefit to your organization depends on your queries' arrival patterns, concurrency, and which operators truly exploit the new silicon. Our team at Keebo has spent the past few weeks validating Gen 2 on real production traces and shipping those insights into our workload-intelligence algorithms. In my latest blog post, I explore: * What Graviton 3 (the hardware behind Gen 2) actually changes—and which query patterns feel it most * How to know when faster genuinely translates to cheaper * Why reinforcement-learning and AI agents beat static rules as tiers like Gen 3 emerge If you’d like early access to Keebo’s Gen 2-aware recommendations—customer or not—reach out to me or the team. We’re happy to explore early access for you to analyze your workload and if/how to best leverage these new developments. 🔗 Read the full analysis here → https://lnkd.in/eymFUisq #CloudDataWarehouse #Snowflake #Gen2 #CostOptimization #DataEngineering

62

2 Comments