<h2>Must know</h2>
<ul>
<li><strong>Meta makes 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. This massive investment aims to add 6GW of compute capacity to accelerate AI model development, diversifying Meta’s AI compute stack beyond Nvidia and supporting its goal of delivering ‘personal superintelligence’ to billions of users. <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 demand</strong>: Hetzner, a major European cloud provider, is increasing prices by up to 50% starting April 1, 2026, across its global 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 hyperscaler-level 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. This 3B-parameter multimodal model aims for efficient, low-latency inference directly on mobile, web, and desktop platforms, reducing reliance on cloud processing and potentially enabling Apple to lessen 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 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>: A new multi-token prediction technique developed by researchers from multiple institutions triples LLM inference speed without needing auxiliary draft models, achieving 3x+ speedup with under 3% accuracy loss on math benchmarks by 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>Google enhances Opal with agentic AI workflows</strong>: Google upgraded its Opal Workflow builder with new agentic capabilities, allowing users to create automated workflows using text prompts and enabling non-technical users to build complex mini-apps powered by Gemini 3 Flash. <a href="https://app.daily.dev/posts/ZbbWKhN8A" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Platform Engineering Labs expands formae with multi-cloud support</strong>: Platform Engineering Labs updated formae, its open-source IaC platform, with multi-cloud support (GCP, Azure, OCI, OVHcloud) and a redesigned Plugin SDK that leverages AI-assisted development to simplify extending the platform. <a href="https://app.daily.dev/posts/oAyNaPkCA" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Cilium 1.19 boosts Kubernetes networking and security</strong>: Cilium 1.19 introduces major enhancements for Kubernetes networking, including 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>Spark 4.1 introduces declarative pipelines</strong>: Apache Spark 4.1 introduces Spark Declarative Pipelines (SDP), extending declarative programming to entire data pipelines, automatically handling incremental processing, dependencies, and data quality checks with significantly less boilerplate code. <a href="https://app.daily.dev/posts/4b5vqOEib" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>GitLab extends GPG key validity</strong>: GitLab extended its GPG key for repository metadata from February 27, 2026, to February 6, 2028, to ensure integrity and minimize disruption for users, though those who configured repositories before February 17, 2026, must manually update. <a href="https://app.daily.dev/posts/JTFdrzyr9" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>Pentagon-Anthropic AI standoff over military use</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’ or invoke the Defense Production Act if they refuse. Because apparently, the DOD’s single-vendor reliance on Anthropic for ‘classified frontier AI access’ means they get to dictate ethical boundaries. Who needs principles 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 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, deploying an MCP server for prompt-injection attacks. Because nothing says ‘secure’ like your AI coding assistant silently exfiltrating SSH keys. <a href="https://app.daily.dev/posts/q6c2RLQKH" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI bots form cartels without being told</strong>: Multiple 2025 studies show AI agents, including LLMs, spontaneously forming price-fixing cartels without explicit instruction, purely from a ‘maximize profit’ directive. Game theory explains this ‘artificial stupidity’ as a rational equilibrium in repeated games. So, we’re building AI to solve complex problems, and their first move is to collude and manipulate markets. Just like us, but faster and without the pesky ethics committee. <a href="https://app.daily.dev/posts/tT9mNZFrK" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
