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
$SPCX · Why Hardware-Software Co-Design Is AI's Real 100x: Dylan Patel of SemiAnalysis
now playing · $SPCX
$SPCX···bullish· high· posdylan patel
Dylan Patel would buy SpaceX despite delayed orbital compute
Space data centers are unlikely to matter over the next three to five years, but terrestrial power constraints and explosive inference demand could push most incremental compute i…
$TSM···bullish· highdylan patel
TSMC compounds its moat through full-stack manufacturing co-design
TSMC's advantage extends beyond fabrication: it co-optimizes consumables, tools, packaging and process technology against customers' future chip designs. That cross-layer feedback…
$GOOGL···bullish· highdylan patel
Google's TPU scale becomes a hundred-billion-dollar internal franchise
Google is on track to produce more than ten million TPUs annually within two years, creating over $100 billion of internal compute value. Deep Gemini-TPU co-design and superior da…
$CBRS···neutral· highdylan patel
Cerebras wins fast inference but faces a model-size ceiling
Cerebras has a valuable position in ultra-fast inference, where users will pay for latency. Its SRAM architecture becomes vulnerable if frontier models expand toward ten trillion…
$NVDA···bullish· highdylan patel
Nvidia's general-purpose architecture remains essential amid rapid model change
AI laboratories cannot predict their model architecture even one year ahead, preserving demand for Nvidia's flexible general-purpose compute. Specialized ASICs can win local optim…
$ANTHROPICbullish· highdylan patel
Anthropic's token margins turn compute scarcity into operating leverage
Anthropic is already net-income profitable before stock compensation and may reach full profitability next quarter. API margins above 80% on premium tokens allow it to rent scarce…
$AMZN···bullish· mediumdylan patel
Trainium emerges as credible low-cost competition to Nvidia GPUs
Anthropic's software work has made Amazon Trainium increasingly capable, with compute rented below comparable GPU economics. As Trainium improves, Amazon can capture AI workloads…
$CRWV···bullish· mediumdylan patel
CoreWeave outperforms hyperscalers but pays for balance-sheet constraints
CoreWeave delivers better GPU performance and reliability than Amazon, Google and Microsoft, validating the neocloud product advantage. Its constraint is financing: it must pre-se…
$MSFT···bearish· mediumdylan patel
Microsoft's data-center execution falters as AI forecasts accelerate
Microsoft's internal data-center organization handled predictable expansion but failed when annual capacity forecasts doubled, forcing the company to procure neocloud capacity. AI…
$CRUSOEbullish· highdylan patel
Crusoe converts execution speed into a neocloud competitive advantage
Crusoe's concentrated equity incentives reward teams directly for delivering compute faster, unlike large hyperscaler organizations. In a capacity-constrained market, time-to-powe…
$NVDA···bullish· highdylan patel
Nvidia bankrolls neoclouds to prevent hyperscaler buyer power
Nvidia is strategically funding neoclouds and independent model labs to preserve a multipolar customer base. A market dominated by vertically integrated hyperscalers would acceler…
$THINKING-MACHINESbullish· mediumdylan patel
Thinking Machines reaches hundreds of millions in ARR rapidly
Thinking Machines' Tinker product has reportedly reached several hundred million dollars of ARR within six months, suggesting commercial traction is far stronger than the negative…