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 Agents · Token value per watt per user is the single most important metric; orchestration layer captures value as models commoditize
now playing · AI Agents
AI Agentstailwindscore 9/10aravind srinivas
Token value per watt per user is the single most important metric; orchestration layer captures value as models commoditize
Models are becoming utilities; the durable moat is orchestrating across models, tools, local and server compute to maximize valuable output tokens per unit of energy per user. The orchestra…
AI Infrastructuretailwindscore 9/10aravind srinivas
Power and permitting remain the structural bottleneck for AI data center build-out
Physical constraints — power generation, grid interconnection, cooling, and local permitting — are far exceed digital infrastructure. Resistance to data center build-outs will continue and…
Frontier AI Modelsheadwindscore 8/10aravind srinivas
Frontier models are becoming interchangeable utilities; application-layer orchestration captures the economic value
Reselling raw model tokens is a commodity business with no moat. Value accrues to the layer that grounds models in context, orchestrates across multiple models and tools, and delivers a uni…
Memory & Storagetailwindscore 8/10aravind srinivas
HBM supply bottleneck gives memory suppliers like Micron outsized pricing power and valuation upside
High-bandwidth memory is the binding constraint for GPU cluster scaling; the supplier of the bottleneck component captures disproportionate economic value, making memory vendors more valuab…
AI Geopolitics & Export Controlsmixedscore 8/10aravind srinivas
Export controls forced China to build vertically integrated, memory-efficient AI stack that may leapfrog US architecture
Denied access to NVIDIA GPUs and HBM, Chinese labs (DeepSeek/Huawei) innovated on KV cache compression, SSD-based inference, and low-interconnect training — a different architectural bet th…
AI Infrastructuretailwindscore 8/10aravind srinivas
Hybrid local/server inference is essential to make 24/7 agents affordable; unmetered local compute unlocks persistent autonomy
Running continuous agent loops purely on server-side tokens is economically infeasible. The winning architecture routes workloads to local models on device for routine tasks, calling fronti…
Semiconductorstailwindscore 7/10aravind srinivas
AI agents shift compute demand back to CPUs, creating renewed growth for AMD and Intel
Agent harnesses run orchestration, data munging, and tool execution on CPUs, not GPUs. This reverses the GPU-only narrative and creates a structural tailwind for enterprise CPU vendors.
Geopolitics & Tradetailwindscore 7/10aravind srinivas
US competitiveness requires treating AI infrastructure as industrial policy: streamline permits, power, and domestic fab capacity
Physical infrastructure build-out — fabs, power, permitting — is the new strategic frontier. The US must accelerate approvals, invest in domestic manufacturing (TSMC Arizona, Intel), and co…
Advertisingheadwindscore 7/10aravind srinivas
Agent-based commerce will disrupt objective-decision categories (travel, commodities); subjective categories (fashion, aesthetics) remain ad-supported
When purchase decisions are objective (price, specs, availability), agents will execute autonomously, bypassing ad-driven discovery. Subjective, vibe-based decisions still require human bro…