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
$ANTHROPIC · Watts, Wafers, and the Future of AI Infra | Gavin Baker
now playing · $ANTHROPIC
$ANTHROPICbullish· high· posgavin baker
Anthropic adds $11B ARR in one month, outpacing decade-built SaaS giants
Anthropic achieved combined ARR of Palantir, Snowflake, and Databricks in a single month with 80% less capital burn than OpenAI, suggesting dramatically superior structural ROIC a…
$TSM···bullish· high· posgavin baker
TSMC wafer bottleneck prevents AI bubble; capacity decisions key to watch
TSMC's wafer supply constraint is the primary governor preventing an AI infrastructure bubble. If TSMC maintains lead over Intel/Samsung and expands capacity in a 'Goldilocks' zon…
$NVDA···bullish· high· posgavin baker
Nvidia could sell $2-3T GPUs if TSMC expanded capacity; GPU life extended to 10-15 years
Nvidia is the primary seller of compute shortage; prefill/decode disaggregation extends GPU useful life to 10-15 years, lowering financing costs from 7% to 5-6%. If TSMC met Jense…
$CBRS···bullish· high· posgavin baker
Cerebras wafer-scale computing is hard and different; three generations to get it right
Cerebras pursued wafer-scale computing — a fundamentally hard architectural decision that enables unique capabilities. Optical wafer bonding and hybrid DRAM bonding could solve IO…
$ALAB···bullish· high· posgavin baker
Astera mis-categorized in copper-loser baskets; biggest product is a switch, not copper
Astera is misplaced in 'copper loser' baskets because its biggest product is a switch that uses both copper and optics to connect to accelerators — definitionally not a copper los…
$SPCX···bullish· highgavin baker
SpaceX orbital compute (racks in space) solves watts shortage; TerraFab brings leading-edge fab to US
Orbital compute is racks in space connected by lasers, not Pentagon-sized stations. SpaceX has 20kW Starlink V3 today, scaling to 100-120kW; Starship reusability makes economics w…
$INTC···neutral· mediumgavin baker
Intel foundry discipline at risk; TerraFab partner but likely to break capacity restraint
Intel is a potential second source for leading-edge wafers but history suggests one of Intel or Samsung will break discipline and overbuild, forcing a bubble. TerraFab partnership…
$TERRA-FABbullish· mediumgavin baker
TerraFab (SpaceX/Tesla/Intel JV) to build largest US fab; attracts A-teams and top talent
TerraFab leverages Elon's hardware engineering reputation to recruit best engineers (Taiwan/Japan/Korea towns) and attract A-teams from ASML, KLA, Lam, Applied Materials. Intel pa…
$XAIbullish· mediumgavin baker
Grok 4.3 on Pareto frontier; best low-cost 500B parameter model
xAI's Grok 4.3 sits on the Pareto frontier (intelligence per cost) as the best low-cost 500B parameter model, alongside Anthropic and OpenAI. Google's Gemini 3.1 clings to frontie…
$GOOGL···neutral· mediumgavin baker
Google lost TPU cost leadership but retains largest compute installed base and YouTube data moat
Google dominated the Pareto frontier (intelligence per cost) 9 months ago but lost leadership due to conservative TPU v8 design decisions aimed at reducing Broadcom/Nvidia depende…
$MU···neutral· lowgavin baker
Google released Turbo Quad DRAM optimization during LTA negotiations with Micron/Samsung/Hynix
Google published Turbo Quad memory optimization paper during long-term agreement negotiations with memory suppliers to lock in high prices — a negotiating tactic. No AI engineer b…
$AMD···neutral· lowgavin baker
AMD MI450 unknown; zero startup engagement; trying to beat Nvidia at GPU game is hard
AMD's MI450 is unproven; zero engagement with startups vs Amazon/Nvidia/Google. Making a 'better GPU' is a losing game — Nvidia gets better TSMC pricing and optimizes across all m…
$ASML···neutral· mediumgavin baker
Semi-cap equipment at 40x earnings vs DRAM at mid-single digits — valuation disconnect
Semi-cap equipment (ASML, KLA, Lam, Applied) trades at 40x next quarter annualized earnings while DRAM trades at mid-single digits. At last cycle peak it was 5x vs 12x; 1000% mult…
$GEV···bearish· mediumgavin baker
GE Vernova valuation implies unfathomable Nvidia share loss
GE Vernova's current valuation builds in massive Nvidia share loss that seems inconsistent with Nvidia's cheap relative valuation and fundamental compute shortage.
$META···bullish· mediumgavin baker
Zuckerberg made Meta AI-first; Llama impressive for first model, improving competitive position
Meta is the only true internet giant to become AI-first internally. Llama was the first model from MSL and is close to the Pareto frontier, an impressive debut. Rate of change mat…
$AMZN···bullish· mediumgavin baker
Trainium best custom chip; Nova models underrated; robotics P&L lift; deep startup engagement
Trainium is the strongest custom silicon effort, with Trainium 3 needing to ramp switch scale-up network for MoE inference economics. Nova internal models are better than credited…
$MSFT···neutral· mediumgavin baker
Microsoft flinched on CapEx, now pivoting compute internally to own models — risky but courageous
Microsoft flinched in early 2025 when CapEx/return algorithm broke, losing allocation position. Satya is now courageously redirecting compute to internal products (Copilot, own mo…
$AVGO···bullish· mediumgavin baker
Broadcom is everyone's favorite ASIC supplier; startup engagement is a level-up signal
Broadcom engages with startups differently — considered a 'level up' for second-gen chips and 'mana from heaven' for first-gen. This engagement creates a structural advantage in t…