Jun. 6 at 12:26 AM
$CBRS = Weak models, absurd valuation.
This deserves deeper analysis.
Cerebras CS-3 needs proprietary liquid cooling, ~23kW per 15U rack, and full data center retrofits.
Nothing like plug-and-play NVIDIA DGX/HGX.Real bottlenecks: HBM memory (SK Hynix 2026 sold out, mostly to
$NVDA) and TSMC CoWoS packaging.
Any big ramp fights the same constrained supply.AWS term sheet + OpenAI backlog exist, but scaling means rewriting 15+ years of CUDA software + building specialized liquid-cooled infra.
That’s a multi-year, high-risk lift.Financials: ~
$510M 2025 revenue, -
$345M GAAP op loss, -
$623M FCF burn, heavy customer concentration. At ~
$44B market cap that’s 87x sales while deeply unprofitable.
More realistic:Semi median 4.5x P/S → ~
$10/share
NVDA-like 20x → ~
$46/share
deep dive = https://x.com/ravagelaserbeak/status/2063049012308283422?s=20
or = https://www.reddit.com/user/constructiontimeagnn/comments/1ty1tht/cbrs_speed_means_nothing_if_the_models_cbrs_is/?utm_source=share&utm_medium=web3x&utm_name=web3xcss&utm_term=1&utm_content=share_button
Extreme hype vs. execution & supply reality. SEC filing: https://www.sec.gov/Archives/edgar/data/0002021728/000162828026035214/cerebras-424b4.htm
$SPY $IWM $QQQ
