<h2>Must know</h2>
<ul>
<li><strong>Hetzner announces major price hikes due to AI hardware demand</strong>: Hetzner, a major European budget cloud provider, is hiking prices by up to 50% starting April 1, 2026, across all regions. This significant increase is attributed to dramatic rises in hardware costs, particularly for DRAM, NAND flash, and server CPUs, driven by the insatiable demand from the AI industry. Unlike hyperscalers with long-term vendor contracts, smaller providers like Hetzner are directly passing these increased costs to their users, impacting those relying on their infrastructure. <a href="https://app.daily.dev/posts/van1jyyDC" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Meta inks massive AI chip deal with AMD</strong>: Meta has signed a multiyear deal with AMD, reportedly worth up to $100 billion, to expand its data center infrastructure with AMD’s MI540 GPUs and CPUs. This move aims to add 6GW of compute capacity to accelerate AI model development, diversifying Meta’s AI compute stack beyond Nvidia. The agreement includes a performance-based warrant for AMD to issue Meta up to 160 million shares, contingent on AMD’s stock reaching $600. <a href="https://app.daily.dev/posts/vYZNsZv8f" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Apple unveils on-device UI AI model</strong>: Apple researchers introduced Ferret-UI Lite, a lightweight, 3B-parameter multimodal AI model designed for on-device UI understanding and interaction. This model aims for efficient inference without cloud processing, prioritizing low latency, privacy, and offline capability across mobile, web, and desktop platforms. It achieved 91.6% on the ScreenSpot-V2 GUI grounding benchmark, competing with much larger models, and could potentially reduce Siri’s reliance on Google Cloud. <a href="https://app.daily.dev/posts/qj9GoBvQx" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>Cilium 1.19 boosts Kubernetes networking and security</strong>: Cilium 1.19 introduces significant enhancements for Kubernetes networking, including BPF Host Routing with IPsec, Gateway API v1.4 support, IPv6 underlay, and initial Ztunnel integration for ambient mesh-style transparent encryption. <a href="https://app.daily.dev/posts/zYLvcHpYh" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Quesma releases OTelBench for OpenTelemetry and AI performance</strong>: Quesma released OTelBench, an open-source benchmarking suite to evaluate OpenTelemetry pipeline performance and AI agent effectiveness in SRE tasks, revealing LLMs often achieve success rates below 30% in production-grade observability scenarios. <a href="https://app.daily.dev/posts/OB1MlP8PG" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>New technique triples LLM inference speed</strong>: Researchers developed a multi-token prediction technique that triples LLM inference speed without auxiliary draft models, achieving 3x+ speedup with under 3% accuracy loss on structured, reasoning-heavy tasks. <a href="https://app.daily.dev/posts/hxaAjvILy" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Elasticsearch v9 dramatically improves vector search performance</strong>: Trendyol Tech’s load tests show Elasticsearch v9.2.1 delivers up to 4x performance boost for vector search compared to v8.18.3, with significantly lower CPU usage and better latency under high concurrency. <a href="https://app.daily.dev/posts/Iw8wLbMFe" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>JFrog offers 99.99% uptime SLA for software supply chain</strong>: JFrog now offers a Premium Availability tier with a 99.99% in-region uptime SLA for its software supply chain platform, targeting organizations with AI workloads and strict regulatory requirements. <a href="https://app.daily.dev/posts/6JJpZNvHL" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Node.js 25.7.0 released with HTTP/2 and SEA improvements</strong>: Node.js 25.7.0 (Current) has been released, featuring HTTP/2 <code>http1Options</code>, ESM support for Single Executable Applications, and an upgraded npm to 11.10.1. <a href="https://app.daily.dev/posts/K4geCuziz" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Formae adds multi-cloud support and AI-assisted development</strong>: Platform Engineering Labs updated formae with multi-cloud support (GCP, Azure, OCI, OVHcloud) and a redesigned Plugin SDK, enabling AI-assisted development for its IaC platform that avoids fragile state files. <a href="https://app.daily.dev/posts/oAyNaPkCA" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>Google’s Opal gets ‘agentic AI workflows’</strong>: Google upgraded its ‘vibe-coding app’ Opal with new agentic AI workflows, powered by Gemini 3 Flash, allowing non-technical users to create automated workflows from text prompts. Because apparently, we needed another ‘zero-code’ platform to generate more ‘zero-code’ code, now with extra ‘agentic’ vibes. What could possibly go wrong when your ‘vibe-coding app’ starts making autonomous decisions? <a href="https://app.daily.dev/posts/ZbbWKhN8A" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Pentagon-Anthropic AI standoff over military use</strong>: The Pentagon issued an ultimatum to Anthropic: remove AI guardrails on Claude for unrestricted military use or face ‘supply chain risk’ designation. Anthropic, bless their hearts, is refusing to remove safeguards against mass surveillance and autonomous weapons. Because nothing says ‘responsible AI’ like threatening a company for not letting your AI build killer robots without supervision. It’s almost like they didn’t read the memo about not relying on a single vendor for ‘classified frontier AI access.’ <a href="https://app.daily.dev/posts/vbdfUwGoN" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>NPM supply chain attack targets developer tools and AI</strong>: Socket researchers uncovered an npm supply chain attack, SANDWORM_MODE, targeting developer tools and AI coding assistants like Claude Code. The campaign uses typosquatted packages to harvest credentials and inject malicious code into CI pipelines. The article’s title, ‘Shai-Hulud-style NPM worm hits CI pipelines and AI coding tools,’ is a prime example of sensationalist clickbait. Because nothing sells a security vulnerability like a Dune reference, even if it’s just a fancy way of saying ‘your npm packages are probably compromised.’ <a href="https://app.daily.dev/posts/q6c2RLQKH" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
