TickerTain
$SPCX····$NVDA····$MSFT····$GOOGL····$ANTHROPIC·$TSLA····$CRWV····$NBIS····$HIMS····$TSM····$ASML····$FIGURE·$LYFT····$UBER····$SNOW····$META····$ELEVENLABS·$CURSOR·$COGNITION·$OPENAI·$AAPL····$RKLB····$XAI·$AMZN····$AMD····$000660.KS····$INTC····
← newsroom
AI Infrastructure · Inference splits 50% on-device, 50% at network edge — not centralized clouds
now playing · AI Infrastructure
AI Agentstailwindscore 9/10matthew prince
Agent traffic to exceed human traffic 1,000x within five years
Bot/agent traffic surpassed human traffic in H1 2026 — years ahead of schedule — and will reach 1,000x human volume as agents exhibit boundless attention (visiting 5,000 sites vs 5 for huma…
AI Infrastructuretailwindscore 8/10matthew prince
Isolates beat containers for agent workloads: 50x CPU efficiency unlock
Traditional containers require full OS/tooling overhead per agent; Cloudflare's Isolates eliminate that, allowing the same CPU fleet to run 50x more agents — a structural advantage as agent…
AI Infrastructuretailwindscore 8/10matthew prince
Inference splits 50% on-device, 50% at network edge — not centralized clouds
Long-running agent tasks (minutes to days) exceed on-device capacity but need proximity for latency, privacy, and data sovereignty; Cloudflare's 350-city network with free bandwidth and sub…
Data Center Infrastructuretailwindscore 7/10matthew prince
Distributed edge architecture avoids mega-data-center power/grid bottlenecks
Cloudflare places hundreds of machines in thousands of existing network interconnection sites rather than building gigawatt campuses, sidestepping new-build permitting, grid constraints, an…
Memory & Storageheadwindscore 6/10matthew prince
Memory prices 'through the roof' creating procurement pressure for AI infrastructure
Rising DRAM costs are a tangible constraint on equipment procurement for AI inference platforms, favoring architectures that maximize compute per memory byte.