The Memory in Your Thumb Drive Could Fix AI’s Big Problem
THE SO WHAT
If high-bandwidth flash can sit closer to compute, the memory wall for LLMs becomes an engineering problem, not a hard limit. Infra teams should be tracking HBM, DRAM, and emerging flash architectures as a single design space, not separate procurement lines.
READ THE SOURCE
MORE FROM THE WIRE
Deep & Emerging TechAutonomous AI worms mark a new era of adaptive cyberattacks
Autonomous AI worms move cyber risk from static perimeter defense to live adversaries that learn and adapt mid-incident. Security teams need to treat AI-native malware as an inevitability and start testing response playbooks against agents that change tactics in real time.
Deep & Emerging TechKalshi Ramps Up Effort to Build Markets for AI Computing Power
Kalshi building prediction-style markets around future compute prices is an early move to financialize AI infrastructure risk. If your roadmap depends on cheap GPUs, you may soon have tools to hedge that exposure — but you’ll also be competing with speculators when capacity tightens.
Deep & Emerging TechMicrosoft promised to be carbon negative. Its emissions just jumped 25%
A 25% emissions jump tied to AI data centers underscores the carbon cost of scaling models, even for companies with aggressive climate targets. If your AI strategy leans on hyperscale, expect more scrutiny from boards and regulators on embodied and operational emissions — and start asking vendors for verifiable climate data.
Deep & Emerging TechThis new macOS infostealer poses as an Apple crash reporting tool to try and steal all your valuable data
Malware masquerading as an Apple crash reporter shows how attackers are leaning on familiar UX patterns to bypass user skepticism. Lock down install permissions on macOS fleets and train teams to treat any unexpected “system” prompt as suspect, not trusted by default.