Expert-curated analysis on the hardware, software, and security shifts that actually matter. Updated weekly.
NVIDIA's Vera Rubin platform leaks show a radical co-designed rack architecture with NVLink 6 Switch, ConnectX-9, and a seamless cableless compute tray. This isn't just a GPU — it's an entire AI datacenter building block.
The Iran-affiliated Handala hacktivist group claims a devastating wiper attack on medical technology giant Stryker. Over 200,000 systems, servers, and mobile devices reportedly wiped. Stryker's offices in 79 countries affected.
The viral open-source AI agent OpenClaw has over 21,000 exposed instances on the public internet. Researchers found misconfigured servers leaking API keys, OAuth tokens, and enabling remote code execution.
Threat actor UNC6426 leveraged stolen keys from the nx npm package supply chain compromise to breach a victim's cloud environment within 72 hours — stealing a GitHub token, abusing OIDC trust to create AWS admin roles, and exfiltrating S3 data.
HBM memory prices have doubled again in Q1 2026. SK Hynix, Samsung, and Micron are all capacity-constrained. Every NVIDIA Blackwell and AMD MI400 GPU needs 6-8 HBM stacks — and there aren't enough.
A new phishing kit called Starkiller dynamically loads real login pages via headless Chrome containers, proxies all user inputs including MFA codes, and offers real-time session monitoring with SaaS-grade analytics dashboards.
Broadcom's custom compute engine business is growing faster than expected. Google TPU, Meta MTIA, and other hyperscaler ASICs are positioning Broadcom as the only company that can challenge NVIDIA's AI datacenter dominance.
CISA added a critical n8n expression injection vulnerability (CVE-2025-68613, CVSS 9.9) to the KEV catalog. Over 24,700 n8n instances are internet-accessible, and attackers are exploiting sandbox escapes for remote code execution.
AMD says its rack-scale 'Helios' AI infrastructure and MI400 series GPUs (CDNA 4, HBM4) remain on track for second half 2026. Oracle confirmed as first customer for MI450 'Altair' GPUs.
Startup Eridu launches a high-radix switch system targeting AI datacenter fabrics. With 102.4 Tb/s aggregate bandwidth, it aims to replace multi-tier switch topologies with single-tier architectures for AI training clusters.
NVIDIA has begun shipping Blackwell Ultra B300 GPUs to major cloud providers. The B300 delivers 1.5x inference throughput over the B200 with 288 GB HBM3e memory, targeting the largest foundation model workloads.
CrowdStrike's Next-Gen SIEM platform has crossed 1,000 enterprise deployments, with 40% of new customers migrating from Splunk. Charlotte AI now triages 85% of alerts autonomously.
TSMC has officially started mass production on its N2 (2nm) process using gate-all-around (GAA) nanosheet transistors. First customers: Apple and NVIDIA. Yields reportedly above 80%.
Apple's M5 Pro has taped out on TSMC's N3P process node. Leaked benchmarks show a 14-core CPU, 20-core GPU, and a 40 TOPS Neural Engine. Expected in MacBook Pro late 2026.
Google's Sec-Gemini v2 achieved a 94% detection rate in DARPA's AI Cyber Challenge benchmarks, outperforming the median human red team. Now integrated into Google Cloud Security Command Center.
AMD's next-gen MI400 AI accelerator specs have leaked: CDNA 4 architecture, 256 GB HBM4, and claimed 2x inference performance over MI300X. Set to ship Q3 2026.