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 · GPU useful life 5-6 years, not 16-18 months; older gens (A100) see price appreciation from new use cases
now playing · AI Infrastructure
AI Infrastructuretailwindscore 9/10michael intrator
GPU compute demand relentless for 4+ years, overwhelming global capacity; no idle GPUs exist
Both CoreWeave and IREN CEOs confirm structural AI compute shortage persisting for years; unconstrained supply would be instantly absorbed, supporting continued capex intensity.
Energy & Power Generationtailwindscore 9/10daniel roberts
Stranded renewable energy arbitrage: data centers follow wind/solar to monetize curtailed power at source
IREN locates at stranded renewable sites (West Texas 45-50 GW vs 12 GW transmission) to convert cheap green electrons into high-value compute tokens; grid connection guarantees 24/7 reliabi…
AI Infrastructuretailwindscore 9/10daniel roberts
Jevons paradox in compute: efficiency gains (lower token cost, faster inference) induce exponentially more usage
IREN and CoreWeave both observe that 10x compute efficiency → many more inferences (images, agents, robotics), not less spend; demand curve bends up with supply.
AI Infrastructuretailwindscore 8/10michael intrator
GPU useful life 5-6 years, not 16-18 months; older gens (A100) see price appreciation from new use cases
Long-term customer contracts (5-6 years) and rising secondary-market prices for Ampere GPUs disprove rapid obsolescence thesis; depreciation schedules should reflect 6-year life.
AI Infrastructuretailwindscore 8/10michael intrator
Innovative GPU-backed 'box' financing drops cost of capital 600bps toward hyperscaler levels
CoreWeave's bankruptcy-remote SPV structure — isolating each customer contract with GPU collateral — enables progressive cost-of-capital reduction, creating a structural financing advantage…
Semiconductorstailwindscore 8/10michael intrator
Memory (HBM) is the current bottleneck in AI infrastructure, not just GPUs
CoreWeave identifies memory as the key throttle: AI-driven demand surge collided with missed 2023 fab investment cycle, creating structural HBM shortage that will take years to resolve.
AI Applicationstailwindscore 8/10aravind srinivas
Multi-model orchestration ('Switzerland' model) enables positive gross margins and best-of-breed routing
Perplexity's model-agnostic orchestration — auto-routing to GPT, Claude, Gemini, open-source — minimizes token spend via RAG and avoids context-window bloat, yielding positive gross margins…
Frontier AI Modelstailwindscore 8/10arthur mensch
Open-source models + portable platform win enterprise customization; data never leaves customer infrastructure
Mistral's portable platform deploys training tools on customer hardware, enabling deep vertical specialization (finance, manufacturing) on proprietary IP without data egress — a key differe…
AI Applicationstailwindscore 7/10aravind srinivas
Enterprise AI shifting to local+cloud hybrid (Perplexity Personal Computer) for privacy and cost savings
Perplexity's Mac Mini-based local orchestration layer keeps sensitive data on-prem while bursting to cloud for heavy tasks; enterprises save >$100M via Computer automation, driving faster e…
Semiconductorsmixedscore 7/10daniel roberts
Custom silicon (Google TPU, Amazon Trainium, Meta MTIA) emerging but Nvidia ecosystem head start makes its roadmap safest for near-term scale
Hyperscaler custom chips are arriving in data centers but Nvidia's incubated software stack, standards, and ecosystem create massive switching costs; following Nvidia roadmaps remains lowes…
Energy & Power Generationtailwindscore 7/10daniel roberts
Nuclear SMRs inevitable for AI power but 10+ year timeline; now is time to mobilize capital and policy
IREN sees nuclear as structural correlation between human progress and energy consumption; SMRs will unlock distributed clean power for compute but require decade-long lead time for deploym…
Data Center Infrastructuretailwindscore 7/10daniel roberts
Fiber abundance and low latency (6ms West Texas to Dallas) debunk metro-proximity requirement for AI data centers
IREN proves AI training/inference clusters can sit at remote renewable sites with negligible latency impact; fiber is ubiquitous in Texas, removing a historic siting constraint.