<h2>Must know</h2>
<ul>
<li><strong>Cloud hardware costs skyrocket due to AI demand</strong>: Hetzner, a major European cloud provider, is hiking prices by up to 50% starting April 1, 2026, across all regions. This is due to a dramatic increase in hardware costs, particularly DRAM, NAND flash, and server CPUs, driven by the AI industry’s insatiable demand. Smaller providers like Hetzner and OVHcloud, lacking the long-term vendor contracts of hyperscalers like AWS and Google, are forced to pass these costs directly to customers. <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 acquire MI540 GPUs and CPUs for its data centers. This massive investment aims to add 6GW of compute capacity to accelerate AI model development, diversifying Meta’s AI compute stack beyond Nvidia. The deal includes a performance-based warrant for AMD to issue Meta up to 160 million shares if AMD’s stock hits $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>Node.js 25.7.0 released</strong>: Node.js 25.7.0 is out, bringing HTTP/2 <code>http1Options</code>, ESM entry point support for Single Executable Applications, and a release candidate for the built-in SQLite module. <a href="https://app.daily.dev/posts/K4geCuziz" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>OpenTelemetry and AI performance evaluation tool released</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. <a href="https://app.daily.dev/posts/OB1MlP8PG" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>GitLab GPG key extended</strong>: GitLab extended its GPG key for repository metadata from February 27, 2026, to February 6, 2028, to minimize disruption for users, though those who configured repositories before February 17, 2026, need to manually update. <a href="https://app.daily.dev/posts/JTFdrzyr9" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Cilium 1.19 enhances Kubernetes networking</strong>: Cilium 1.19 introduces BPF Host Routing with IPsec, Gateway API v1.4 support, stable Multi-Pool IPAM, 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>Erlang and Elixir ecosystem updates</strong>: The Thinking Elixir podcast discusses Hackney 3.1’s pure Erlang HTTP3 support, Hornbeam 1.0 for hosting Python web apps on BEAM, and Pyex, a pure Elixir Python interpreter. <a href="https://app.daily.dev/posts/rNCq4cuuf" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Elasticsearch v9 boosts vector search performance</strong>: Trendyol’s load tests show Elasticsearch v9.2.1 delivers a 4x performance boost for vector search compared to v8.18.3, with significantly higher throughput and lower latency. <a href="https://app.daily.dev/posts/Iw8wLbMFe" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Spark 4.1 introduces declarative pipelines</strong>: Apache Spark 4.1 introduces Spark Declarative Pipelines (SDP), extending declarative programming to entire data pipelines to reduce boilerplate and automate incremental processing, dependency management, and data quality checks. <a href="https://app.daily.dev/posts/4b5vqOEib" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>Google’s new vibe-coding app with agentic AI</strong>: Google upgraded its Opal Workflow builder with “agentic capabilities” and “Super Gems on Gemini,” allowing users to create automated workflows with text prompts. It’s powered by Gemini 3 Flash and described as a “vibe-coding app.” So, Google’s latest attempt to make coding “vibey” involves an AI agent that autonomously selects tools and plans tasks. Because nothing says “developer productivity” like a “vibe-coding app” that lets non-technical users build “complex mini apps.” What could possibly go wrong? <a href="https://app.daily.dev/posts/ZbbWKhN8A" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>NPM supply chain attack gets a sci-fi movie title</strong>: Socket researchers uncovered a new npm supply chain attack, SANDWORM_MODE, involving 19 typosquatted packages impersonating developer utilities and AI coding tools like Claude Code. The malware harvests tokens and credentials, injecting malicious code into CI pipelines and deploying an MCP server for prompt-injection attacks. The article title, “Shai-Hulud-style NPM worm hits CI pipelines and AI coding tools,” is a prime example of sensationalist clickbait. Because apparently, a regular old supply chain attack isn’t dramatic enough without invoking Dune references. <a href="https://app.daily.dev/posts/q6c2RLQKH" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Pentagon demands Anthropic remove AI guardrails</strong>: The Pentagon issued an ultimatum to Anthropic: remove AI guardrails on Claude for unrestricted military use or face designation as a “supply chain risk.” Anthropic, apparently still clinging to quaint notions of ethics, refuses to allow mass surveillance and autonomous weapons. The DOD, having put all its eggs in one AI basket, is now threatening to invoke the Defense Production Act. Because when you can’t get your way, just threaten to nationalize the AI company. Classic. <a href="https://app.daily.dev/posts/vbdfUwGoN" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
