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
$CBRS · Two 24-Year-Olds Just Beat NVIDIA
now playing · $CBRS
$CBRS···neutral· lowjosh kale
Cerebras public but down 35% YTD despite inference tailwind; hosts question pricing
Cerebras plays the same accelerated inference mode as Etched and Groq but stock has slumped post-IPO; hosts debate whether market underappreciates inference demand trajectory or i…
$GROQneutral· lowjosh kale
Nvidia acquired Groq for ~$20B as first step toward inference-optimized architecture
Hosts state Nvidia bought Groq (likely meaning Groq) for $20B to integrate inference accelerator tech; Groq's LPU architecture similar to Etched/Cerebras but now inside Nvidia.
$NVDA···neutral· mediumejaaz
Nvidia's 75% GPU share masks inference underutilization as demand shifts 80% to serving
Nvidia GPUs only achieve 30-40% utilization on inference workloads while inference demand has flipped from 1/3 to 2/3 of compute and heads toward 80%; Nvidia acquired Groq for ~$2…
$ETCHEDbullish· highejaaz
Etched builds transformer-hardcoded inference ASICs targeting 10-50x efficiency gains
Etched's full-stack inference rack achieves 80-90% utilization vs Nvidia's 30-40% by hardcoding the transformer architecture into silicon, halving voltage for 75% less power, and…
$GOOGL···neutral· lowejaaz
Google TPUs and Ironwood chips lack startup urgency; not sleeping in factories like Etched
Google builds custom accelerators but lacks existential urgency of startups; TPU team doesn't have 'people sleeping in the factory' because Google survives if TPUs fail.
$TSM···bullish· mediumejaaz
TSMC manufactures and invests in Etched via venture fund; critical enabler for custom AI silicon
TSMC's advanced packaging and process technology enables startups like Etched to tape out quickly; TSMC's venture fund investment signals confidence in transformer-hardcoded ASIC…
$OPENAIbullish· mediumejaaz
OpenAI vertically integrates with Broadcom on Jalapeno inference chip hyper-optimized for ChatGPT
OpenAI's custom chip avoids hardcoding transformer architecture, instead specializing for GPT workloads using proprietary model knowledge; vertical integration of model + silicon…
$MDTKF···bullish· lowejaaz
MediaTek up 180% YTD as custom AI chip design partner for hyperscalers
MediaTek benefits from hyperscaler custom silicon wave; hosts highlight 180% YTD gain as evidence market rewards inference chip design exposure.
$AVGO···bullish· mediumejaaz
Broadcom emerges as key custom silicon partner for OpenAI and other frontier labs
Broadcom designs purpose-built inference chips (OpenAI's Jalapeno) and enables rapid tape-outs (9 months vs 1.5-2 years norm); stock up ~10% YTD but hosts see asymmetric upside as…
$ANTHROPICbullish· lowejaaz
Anthropic rumored profitable this quarter on inference margins; potential Etched acquirer
Anthropic's inference profit margins may drive Q1 profitability; as a frontier lab without custom silicon, it's a natural acquirer for Etched to achieve vertical integration like…
$AMZN···neutral· lowejaaz
Amazon Trainium chips present but not pace-setters in inference specialization race
Amazon's Trainium joins Google TPUs as hyperscaler custom silicon but hosts see slower velocity versus nimble startups like Etched and Cerebras.