TickerTain — Make Analysis Fun Again

▶ 0:05 · $ANTHROPIC · Why compute prices might 10x as AI gets smarter

now playing · $ANTHROPIC

Anthropic revenue projected to hit $100-150B as inference margins surge to 80%+

Anthropic's revenue has 10x'd YoY for three straight years while compute only grows 3x, driving inference margins from 40% to over 80% and creating massive economies of scale in m…

$OPENAIneutral· mediumdwarkesh patel

OpenAI inference compute share doubled from 25% to ~50% in 2024

OpenAI's compute allocation to inference has risen from 25% in 2024 to approximately 50% now, reflecting the structural shift toward inference-heavy workloads as models mature and…

$GOOGL···neutral· mediumdwarkesh patel

Google paying 2x spot price for 110K GPUs from SpaceX as compute scarcity intensifies

Google is paying $900M/month for 110,000 GPUs (GB200/GB300 blend) from SpaceX at 2x the spot price, which itself is 40% above February troughs, illustrating frontier labs' willing…

$SPCX···bullish· mediumdwarkesh patel

SpaceX emerging as major AI compute landlord renting GB200/GB300 clusters to Google and Anthropic

SpaceX is supplying large-scale GPU clusters (110K GPUs to Google alone) to frontier labs at 2x spot pricing, positioning itself as a critical compute infrastructure provider in t…

$NVDA···bullish· mediumdwarkesh patel

Compute prices could rise 15x as AI approaches human-level software engineering

As AI models become more capable, they monetize compute far more effectively — an H100 running a human-level software engineer would be worth $250K/year (15x current spot), and st…

$TSM···bullish· mediumdwarkesh patel

TSMC N3 wafer capacity bottleneck as AI absorbs 86% of leading-edge allocation

AI's share of TSMC N3 wafer capacity will jump from 60% to 86% by end of next year, hitting a hard wall where no more leading-edge capacity can be reallocated from smartphones/PCs…

$ASML···neutral· mediumdwarkesh patel

ASML EUV machine production bottlenecking new fab capacity through 2030

New fab construction is bottlenecked up to 2030 and beyond by ASML EUV machine production rates, limiting the 1.2x compute scaling contribution from new fabs.

←