<h2>Must know</h2>
<ul>
<li><strong>Hetzner announces significant price hikes</strong>: Hetzner, a major European cloud provider, is increasing prices by up to 50% starting April 1, 2026, across all regions. This is attributed to a dramatic surge in hardware costs, particularly for DRAM, NAND flash, and server CPUs, driven by the AI industry’s demand. Unlike hyperscalers with long-term vendor contracts, smaller providers like Hetzner are directly passing these increased costs to users, with RAM prices alone rising 80-400%. <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 aims to add 6GW of compute capacity for 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, 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>Quesma releases OTelBench for OpenTelemetry and AI</strong>: Quesma released OTelBench, an open-source benchmarking suite to evaluate OpenTelemetry pipeline performance and AI agent effectiveness in SRE tasks, finding 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>Google enhances Opal with agentic AI workflows</strong>: Google upgraded its Opal Workflow builder with new agentic capabilities, enhancing “Super Gems on Gemini” with an “Agent step” that analyzes goals, determines approaches, and calls tools automatically, powered by Gemini 3 Flash. <a href="https://app.daily.dev/posts/ZbbWKhN8A" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Spark declarative pipelines for end-to-end data engineering</strong>: Spark Declarative Pipelines (SDP), introduced in Apache Spark 4.1, extends declarative programming to entire data pipelines, handling incremental processing, dependency ordering, and data quality checks automatically. <a href="https://app.daily.dev/posts/4b5vqOEib" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>JFrog offers 99.99% uptime SLA for software supply chain</strong>: JFrog announced Premium Availability, offering a contractual 99.99% in-region uptime SLA for its software supply chain platform, reducing allowable annual downtime to just 52 minutes for core products. <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 SQLite updates</strong>: Node.js 25.7.0 (Current) has been released, featuring HTTP/2 <code>http1Options</code>, ESM entry points 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>Project Leyden improves Quarkus startup performance</strong>: Project Leyden’s class caching approach helped Quarkus identify and address hidden startup bottlenecks like runtime reflection and expensive static initializers, leading to significant performance improvements. <a href="https://app.daily.dev/posts/q0c622lXB" 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 up to 4x performance boost for vector search compared to v8.18.3, with 2x throughput and 50% less CPU usage under high concurrency. <a href="https://app.daily.dev/posts/Iw8wLbMFe" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>Pentagon demands Anthropic remove AI guardrails</strong>: The Pentagon issued a Friday deadline to Anthropic to remove AI guardrails on Claude for “unrestricted military use,” threatening to label them a “supply chain risk” if they don’t comply. Because apparently, the DOD, despite a Biden-era directive to avoid single-vendor reliance, is now completely dependent on Anthropic for classified frontier AI. Who needs ethical AI when you have a deadline? <a href="https://app.daily.dev/posts/vbdfUwGoN" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>NPM supply chain attack uses ‘Shai-Hulud-style’ worm</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 article title uses “Shai-Hulud-style worm,” because nothing says serious security threat like a Dune reference. The malware harvests tokens and credentials, but don’t worry, the “dormant polymorphic engine” and “dead switch home-directory wiper” are inactive by default. Phew. <a href="https://app.daily.dev/posts/q6c2RLQKH" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Microsoft and Salesforce confirm AI suspicions</strong>: Microsoft Research and Salesforce AI “analyzed over 200,000 AI conversations” and reportedly “confirmed industry suspicions regarding AI.” What suspicions? What findings? Who knows! The information comes from a vague retweet, making it prime clickbait. It’s like saying, “We looked at a lot of data and, yep, things are still things.” Groundbreaking. <a href="https://app.daily.dev/posts/FAzFsVs8B" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
