VASS

Another SAP #SAPPHIRE in the books. Could post about #AI , #A2A , #innovation — but the real takeaways were simpler for me... Relationships are everything: People work with people. #TRUST is the only currency that compounds. Takes years to build, seconds to lose. SAPPHIRE is a reunion: Some of my favorite people in one place. The fact that we only do this once a year is the only thing wrong with it. Partnerships are bi-directional: Real partnerships aren't built on asks — they were built on reciprocity and paying it forward. Strategic vs. Tactical: It's easy for me to only fixate on what's in front of me. The people who will win next year are planting seeds right now, not just harvesting. The badge is off. What a week. The work starts (continues) now. #VAMOS

https://lnkd.in/eSiyhr_5

SwissMartech is back. 🇨🇭 5 years after I stopped, I'm relaunching the project for 2026. Same principle. Bigger ambition. Back in 2018, I had one simple idea: List every Swiss Marketing Technology vendor in one place. I ran it until 2021. Met great people along the way, including Frans Riemersma , and contributed to the European Martech Landscape. Then I paused. Since then, the martech world has changed. AI Search. Agentic experience. Automation. New CX challenges. The Swiss market needs its own lens on all of this. Here's what's coming 👇 1️⃣ A refreshed Swiss SaaS and product listing 2️⃣ Original research with a Swiss focus 3️⃣ Webinars with Swiss market actors 4️⃣ A video series The content will cover key themes: → Brand visibility and AI search → AI and automation → Customer experience And I won't be doing this alone. New team members are joining soon. More on that shortly. 😉 The first original study is already live. 4,000 words on which bank is most visible in Swiss AI search. Link in the comments. 👇 PS: Subscribe to the newsletter if you want the next studies in your inbox.

We launched notape.ch — an alternative to hockey tape. And yes, we vibe coded the whole thing on Replit . Honestly? It was a great experience. But also more work than I expected. The cost side is almost absurdly low — no CMS license, no direct hosting fees, no dev agency. Just $20/month and a few extra dollars when the prompts got heavy. But don't mistake cheap for easy. Getting from "AI-generated draft" to something I was actually proud to launch took many hours and days of testing and fixing. The money is cheap. The time is not. AI also doesn't sweat the details the way you do. The 80% comes fast. The last 20% is still yours — a misaligned footer on mobile, a broken nav link, inconsistent spacing. You still need eyes, taste, and patience. What surprised me most though: domain knowledge matters more, not less. Things like client-side rendering affecting SEO, using the right formats for assets (WebP, WebM), running Lighthouse reports — none of that is visible to the user, but it makes or breaks the site. If you don't know what to ask for, AI might not help you get there. On the trust side — notape.ch version 1 takes orders through a simple form. Intentionally. I wouldn't trust a vibe-coded checkout to be secure and conversion-ready just yet. That's why we're migrating to @Shopify once we get some traffic going. There's a ceiling on where I'd trust AI-built tools right now, and payments sits above it. This is my second Replit build and I'm hooked. Next up: Claude Code. Would you build an e-commerce store with Replit , or leave payments to an established player like Shopify ? Sina Fröhlich , Tim Koller #VibeCoding #Ecommerce #Hockey

800 VDC Data Center Power Path — From Grid to GPU We often talk about efficiency in data centers, but rarely look at the full power journey end-to-end. Here’s what an 800 VDC architecture starts to look like when you trace it all the way down to the GPU: High voltage AC → step-down transformer → medium voltage switchgear → AC/DC rectification → 800 VDC distribution → DC switchgear → rack-level distribution → point-of-load conversion → low-voltage rails → GPU Compared to traditional AC designs, a few things stand out immediately: • Fewer conversion stages across the power path • High-voltage DC distribution closer to the load • Reduced current for the same power → lower I²R losses • Better alignment with batteries and renewable sources At the rack and server level, the model shifts as well: Instead of multiple AC → DC conversions, power is delivered as high-voltage DC and converted closer to the point of use. This allows more efficient regulation for CPUs and GPUs, especially in high-density AI environments. But the trade-offs are real: • Protection and fault isolation in DC systems is more complex • Arc management requires different design approaches • Standards and ecosystem are still evolving Still, as rack densities continue to rise and efficiency becomes critical, 800 VDC is moving from concept to serious consideration. The question is no longer “if” — but where it makes the most sense first. Curious how others are approaching DC distribution in real deployments.

I'm pleased to announce that I've earned the AWS Certified Solutions Architect - Associate certification, which I've been aiming for several months. For those preparing for this certification, the courses by Stéphane Maarek and Joan Amengual are invaluable, as are the practice exams by Jon Bonso .

Finance people often describe CUDA as a classic network-effects flywheel: huge install base, huge ecosystem, impossible to displace. I think that misses where AI developers actually live. Most of us do not care about kernels or low-level infrastructure. However, we do care about whether things work with PyTorch or TensorFlow, whether the tooling is mature, and whether the model stack is portable. So yes, compatibility matters. Yes, ecosystem matters. But those effects are increasingly higher up the stack and not where NVIDIA believes they are. That is why I do not view CUDA itself as a sustainable moat. It was a major lead-generating advantage, but that is different from permanent lock-in. If a competing platform can match performance and integrate cleanly with the frameworks developers already use, many users will not care what sits underneath. The market keeps assigning the moat to CUDA. I think the moat, to the extent it exists, is more fragmented and much less durable than NVIDIA bulls believe.

View my verified achievement from Amazon Web Services (AWS) .

Can't wait to kick off the next Hacking Finance session: AI Agents Are Here, Now What?, with Alex Lee, Matthew Paul , and Clifford Son ! See you all in ~1hr (9am PT)!