TickerTain — Make Analysis Fun Again

TickerTain

←

▶ 3:48 · $NVDA · Cerebras CEO on the Future of Data Centres, Token Costs & Memory | Should US Companies Sell to China

episode shorts · 3

Cerebras CEO WARNS About Selling AI Chips to China

HR Is About to Be Automated

AI Didn’t Cause These Layoffs

now playing · $NVDA

$NVDA···neutral· highandrew feldman

Nvidia's GPU dominance creates HBM/CoWoS bottlenecks; neo clouds buy at 70-80% gross margins

Nvidia GPUs depend on HBM (only 3 suppliers) and CoWoS packaging at TSMC, creating structural supply constraints. Nvidia funds neo clouds (CoreWeave etc.) to compete with hypersca…

$OPENAIbullish· high· posandrew feldman

OpenAI's early compute foresight (Sam Altman) was a superpower; now a $20B+ Cerebras customer

Sam Altman believed exponential compute demand years ahead of others and contracted for power, data centers, and hardware early. This foresight is a superpower in exponential grow…

$MU···bullish· highandrew feldman

Micron earning 80-85% gross margins on HBM — software margins on memory

HBM is the #2 bottleneck after TSMC fab space. Only three companies (Samsung, Micron, Hynix) make HBM. They cannot add capacity quickly (fabs take 5 years, $40B). Micron is printi…

$CBRS···bullish· high· posandrew feldman

Cerebras CEO: Wafer-scale architecture avoids HBM/CoWoS bottlenecks, 15x faster inference

Cerebras uses on-chip SRAM instead of HBM, avoiding the severe memory shortage and CoWoS packaging constraints that limit GPU suppliers. At 5nm (vs oversubscribed 3nm), their supp…

$TSM···bullish· highandrew feldman

TSMC 3nm oversubscribed, 5nm advantage; CoWoS packaging a key bottleneck

TSMC's 3nm node is the most oversubscribed; Cerebras at 5nm has supply advantage. CoWoS advanced packaging is a critical constraint for GPU makers. Onshoring leading-edge fabs and…

$GOOGL···neutral· mediumandrew feldman

Google's full-stack TPU ownership may yield lowest token cost, but single-customer model limits scale

Google owns the full stack from TPUs to data centers, networking, power — potentially making them the lowest-cost token producer. However, historically selling chips only to yours…

$CRWV···bullish· mediumandrew feldman

CoreWeave praised for innovative financial engineering and rapid deployment

CoreWeave solved financial challenges with innovative structures (first to use GPU-backed financing) and excels at rapid deployment — a critical skill in the current supply-constr…

$KIMMY-K2bullish· mediumandrew feldman

Chinese open-source models (Kimmy K2, DeepSeek, Qwen, GLM) exceptionally good, near frontier

Chinese open-source models are exceptionally good — not quite at closed-source frontier but close. Cerebras demonstrated 6.7x faster inference on Kimmy K2 vs next fastest GPU clou…

$ASML···neutral· highandrew feldman

ASML/TSMC choke points enable US to keep China down-rev on leading-edge chips

The semiconductor supply chain has natural choke points: TSMC for fab, ASML for EUV lithography. These allow the US to restrict China's access to cutting-edge technology without n…

←