KubeSimplify Diaries - Tuesday April 7, 2026
Your daily dose of what matters in AI, Cloud Native & Tech.
TOP STORY
DeepSeek V4 Confirmed to Train on Huawei Ascend Chips: The AI Chip War Just Got Real
Reuters confirmed on April 4 that DeepSeek’s upcoming V4 model → a ~1 trillion parameter MoE monster with 1M context window → will train on Huawei’s Ascend 950PR chips instead of Nvidia GPUs. This is the first frontier-class model to break the Nvidia dependency entirely. With Apache 2.0 open weights planned, 81% SWE-bench score, and $0.30/MTok pricing, DeepSeek continues to punch way above its weight. Expected to drop in the last two weeks of April.
Why it matters: If a trillion-parameter model can train competitively on non-Nvidia silicon, every assumption about the GPU moat needs revisiting. This has massive implications for AI infrastructure strategy globally.
Source: Reuters via FindSkill.ai | NxCode Specs
AI & MODELS
GPT-5.4 Ships: 272K default context (up to 1M experimental), native computer use, 75% on OSWorld-V benchmark. OpenAI is pushing hard on agents-that-act, not just agents-that-chat. Source
MCP Crosses 97 Million Installs: Anthropic’s Model Context Protocol hit 97M installs in March. Every major AI provider now ships MCP-compatible tooling. It’s becoming the USB-C of AI agent connectivity.
Google Gemma 4 Goes Open Source: Google DeepMind dropped Gemma 4 on April 2 with Apache 2.0 licensing. Four variants (2B, 4B, 26B MoE, 31B Dense), 256K context, native vision + audio, 140+ languages. The 31B model hit #3 on Arena AI’s text leaderboard, beating models 20x its size. Runs fully offline on phones, Raspberry Pi, and Jetson Orin Nano. Source
Broadcom Expands AI Chip Deals: New agreements with Google and Anthropic for custom silicon. Quietly becoming the second most important AI chip company after Nvidia.
VOICES & IDEAS
Andrej Karpathy: “LLM Knowledge Bases → more tokens into knowledge, less into code”
Karpathy shared his workflow for using LLMs to build personal knowledge bases. He indexes source docs (papers, repos, articles) into a raw/ directory, then has an LLM “compile” a wiki → a collection of .md files with summaries, backlinks, and concept articles. He uses Obsidian as the frontend, and the LLM writes and maintains all the wiki content. His research wiki has grown to ~100 articles and ~400K words. Once big enough, you can ask complex questions against it and the LLM researches answers across the whole wiki. No fancy RAG needed, just auto-maintained index files and summaries.
His take: “I think there is room here for an incredible new product instead of a hacky collection of scripts.”
CLOUD NATIVE & INFRA
Dapr Agents v1.0 GA: Announced at KubeCon EU Amsterdam. Production-ready framework for resilient AI agents on Kubernetes: durable workflows, state management across 30+ databases, SPIFFE-based identity, multi-agent coordination. ZEISS Vision Care already running it in production. Source
Kubernetes 1.36 Drops April 22: Big one: Ingress-Nginx officially retires in favor of Gateway API. Enhanced Linux User Namespaces and DRA improvements for hardware maintenance. Start planning your migration. Source
Velero Joins CNCF: The backup/restore project for Kubernetes officially enters the CNCF ecosystem. Announced at KubeCon EU. Source
WORTH READING
Nvidia Physical AI Data Factory Blueprint: Open blueprint on GitHub (this month) for accelerating robotics, vision AI agents, and autonomous vehicle development.
AI Data Centers Now Use More Power Than 30 Countries: Data centers account for 70%+ of new large-load grid requests. The energy bottleneck is real.
Bain: GTC 2026 — AI Becomes the Operating Layer: Solid analysis of Nvidia’s shift from selling chips to selling infrastructure stacks.
EVENTS
KubeCon EU 2026: Just wrapped in Amsterdam. Catch the recorded sessions if you missed it.
Kubernetes 1.36 Release: April 22. Mark your calendars.
SAIYAM’S TAKE
DeepSeek 4 on the Huawei chips is an interesting bet and seeing the model is what I am looking forward to. Also removing Nvidia dependency would be crazy to see happening in 2026 or early 2026 I should say. If this happens, does this impact your AI strategy?
Subscribe to KubeSimplify Diaries for daily updates on AI, Kubernetes & Cloud Native. Share with your network if you found this useful!