<h2>Must know</h2>
<ul>
<li><strong>Meta goes big on AMD for AI infrastructure</strong>: Meta has inked a multiyear deal with AMD, reportedly worth up to $100 billion, to acquire MI540 GPUs and CPUs. This massive investment 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>Hetzner announces significant price hikes due to AI hardware demand</strong>: Hetzner, a major European budget cloud provider, is increasing prices by up to 50% starting April 1, 2026, across all regions. The company attributes these hikes to dramatic increases in hardware costs, particularly for DRAM, NAND flash, and server CPUs, driven by the AI industry’s demand straining global supply chains. This highlights how smaller providers, lacking hyperscalers’ long-term vendor contracts, are directly passing these costs to users. <a href="https://app.daily.dev/posts/van1jyyDC" 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, on-device AI model designed to understand and interact with user interfaces directly on mobile, web, and desktop platforms. This 3B-parameter multimodal model prioritizes low latency, privacy, and offline capability, aiming for efficient inference without relying on cloud processing. It achieved 91.6% on the ScreenSpot-V2 GUI grounding benchmark, competitive with much larger models, and could reduce Siri’s dependence 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>Quesma releases OTelBench for OpenTelemetry and AI evaluation</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 instrumentation. <a href="https://app.daily.dev/posts/OB1MlP8PG" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Cilium 1.19 enhances Kubernetes networking and security</strong>: Cilium 1.19 introduces BPF Host Routing with IPsec compatibility, Gateway API v1.4 support with GRPCRoute, and stable Multi-Pool IPAM, enhancing secure and observable Kubernetes networking. <a href="https://app.daily.dev/posts/zYLvcHpYh" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>New technique triples LLM inference speed</strong>: A new multi-token prediction technique triples LLM inference speed without auxiliary draft models by fine-tuning pretrained models with a student-teacher RL-inspired paradigm, embedding acceleration directly into model weights. <a href="https://app.daily.dev/posts/hxaAjvILy" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Debezium joins Google Summer of Code 2026</strong>: Debezium is joining Google Summer of Code 2026, offering projects including source connectors for Milvus and SQLite, a CLI tool, and PyDebeziumAI for AI integration via Python. <a href="https://app.daily.dev/posts/XH8DOZHBb" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Formae IaC platform expands multi-cloud support</strong>: Platform Engineering Labs updated formae, its open-source IaC platform, with beta support for GCP, Azure, OCI, and OVHcloud, featuring a redesigned Plugin SDK and an AI-assisted toolkit for infrastructure builders. <a href="https://app.daily.dev/posts/oAyNaPkCA" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Babyshark: a new PCAP TUI in Rust</strong>: Babyshark is a new open-source Rust-based terminal UI for analyzing PCAP network captures, offering a flows-first approach with anomaly detection and plain-English flow explanations. <a href="https://app.daily.dev/posts/v12mxt3cf" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>SurrealDB aims to replace entire backend stacks</strong>: SurrealDB offers a multi-model database engine combining authentication, authorization, API endpoints, graph relationships, full-text search, vector embeddings, live queries, and file storage into a single system. <a href="https://app.daily.dev/posts/CsZNoAuHp" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>Google enhances Opal with agentic AI workflows</strong>: Google upgraded its ‘vibe-coding app’ Opal with new agentic capabilities, allowing users to create automated workflows using text prompts powered by Gemini 3 Flash. They’re calling it ‘Super Gems on Gemini’ and claiming it lets non-technical users build complex mini-apps. Because what the world really needs is more ‘vibe-coding’ and ‘super gems’ to solve all our problems, right? <a href="https://app.daily.dev/posts/ZbbWKhN8A" 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, involving 19 typosquatted packages impersonating developer utilities and AI coding tools like Claude Code. The malware harvests credentials and injects malicious code into CI pipelines. The article title uses sensationalist ‘Shai-Hulud-style worm’ language, because nothing says ‘serious security threat’ like a Dune reference and a healthy dose of clickbait. <a href="https://app.daily.dev/posts/q6c2RLQKH" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Pentagon-Anthropic AI standoff over military use</strong>: The Pentagon issued a deadline to Anthropic to remove AI guardrails on Claude for unrestricted military use, threatening to label them a ‘supply chain risk’ if they refuse. Anthropic, apparently, still believes in ethical AI, even when Uncle Sam comes knocking. The Pentagon, meanwhile, is shocked—shocked!—to find that relying on a single vendor for ‘classified frontier AI access’ might have consequences. <a href="https://app.daily.dev/posts/vbdfUwGoN" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
