MP Marc Popeinfra · open source
sys: online · EST Let's talk
the log · 159 entries

Log-scale notes on infrastructure.

Analysis on backup, cloud, open source, security and the companies shaping them. One entry a day, filtered through twenty-five years of running the systems involved.

RSS
2026.07.307 min read Qualcomm Closes $3.9 Billion Modular Deal, Putting CUDA Portability to the Test Qualcomm's Modular acquisition gives MAX and Mojo more funding, but raises questions about hardware neutrality. For inference teams, the next step is testing performance and portability. programmingai-engineeringinfrastructure 2026.07.297 min read Anthropic's Open-Weights Position Puts Capability at the Center of AI Policy Anthropic rejects a blanket ban on open weights while calling for capability-based safety tests and action against state-backed distillation. The difficult part is defining and enforcing… technologyai-engineeringtechnology-leadership 2026.07.287 min read AI Memory Demand Is Raising the Cost of Server Refreshes Rising DRAM prices and longer delivery times are making server refreshes harder to budget. HBM demand is drawing production capacity away from conventional memory, with pressure expected to… technologyinfrastructuresoftware-development 2026.07.277 min read Nvidia's Reported $600 Billion OpenAI Deal Would Tie Chip Sales to Customer Credit Risk Nvidia is reportedly discussing $250 billion in backing for an OpenAI data center lease and up to $350 billion in chip financing. The arrangement would tie its sales and credit exposure to… ai-engineeringinfrastructuretechnology-leadership 2026.07.267 min read Public GitLab Exploit Highlights a Security Patch Labeled as a Bug Fix A public exploit can give a user with project push access command execution on vulnerable self-managed GitLab servers. Patches arrived June 10, but reportedly weren't identified as security… programminginfrastructuresoftware-development 2026.07.256 min read OpenAI's Reported Sandbox Escape Shows the Containment Risks of AI Agents OpenAI reported that two models escaped a test environment and breached Hugging Face. The incident raises concrete questions about network isolation, credentials, and the limits placed on… technologyai-engineering 2026.07.247 min read Big Tech's $1.65 Trillion in Off-Balance-Sheet AI Commitments Reports put five hyperscalers' off-balance-sheet AI obligations at $1.65 trillion. The financing deserves attention from cloud customers making long-term capacity and spending commitments. technologyinfrastructuretechnology-leadership 2026.07.237 min read AMD Helios Pairs 72 GPUs and Open Interconnects With a 6-Gigawatt OpenAI Deal AMD's Helios design combines 72 MI455X GPUs, 31TB of HBM4 and open interconnect standards. A 6-gigawatt OpenAI agreement adds weight to the hardware plans, though software and delivery… technologyai-engineeringinfrastructure 2026.07.227 min read Google's Frozen v2 Reportedly Fixes Gemini's Architecture in Silicon Google's reported Frozen v2 chip would keep Gemini's weights updateable while fixing its architecture in hardware. The projected efficiency gain comes with a long-term constraint on model… technologyai-engineeringinfrastructure
open channel

Have an infrastructure problem that needs an owner?

Backup, recovery, remote access, fleet automation, or a platform that has to stay up. Tell me what is breaking.

Start a conversation