Culper's CBRS short report is straight up bullshit and here's the receipts
retained text
Culper dropped their "**hot garbage**" report this morning and paper hands dumped the stock -10%. I read all 16 pages so you don't have to. **Their whole thesis is one bad model.** **The “wafers die in 1.5-2.5 years**" number comes from an Arrhenius calc where they scale aging by **POWER DRAW.** chip aging depends on **temperature**. **not watts.** **temperature** and their own footnote admits Cerebras cools with \~21C water vs up to 45C on Nvidia's NVL72, with \~3x the coolant flow. So they literally show the chip gets cooled way harder, then assume it runs just as hot anyway and call that "**conservative**." **that's not conservative, that's making shit up with extra steps.** **they're mad about the thing that makes it last longer.** constant temp = no thermal cycling. heat up, cool down, repeat, that's what cracks solder and kills big packages. Cerebras keeps the wafer evenly warm on purpose. Culper calls this a "jerry-rigged workaround." any hardware engineer would call it a **design choice.** costs more power, **sure**. a death sentence? lol no😆 **the "evidence" is 6 anonymous guys.** mostly university and research on-prem setups, some obviously talking about the old CS-2. first gen hardware needed some warranty swaps. call the SEC. still zero actual fleet failure rate in the entire report. not one fcking number. **the connector "scandal" is embarrassing.** "Cerebras didn't invent the connector, they bought it from a vendor!!" yeah bro, that's **called a supply chain.** They scaled a part to a size nobody's done before and qualified it with the vendor. their own anonymous source literally says that. this is a full section of a short report. **think about that.** **the OpenAI math is a random number generator.** NPV goes from +$4.1B to -$13.8B depending on whether a box costs $1M or $3.5M. pick a number, get a conclusion. their discount figures don't even match between the summary and their own table. who proofread this. **and look at the timing.** same day 19.4M shares unlock. bigger unlock Nov 9. earnings Nov 12. their disclaimer straight up says they may trade right after publishing. this isn't research, it's a big short position with a PDF stapled to it. **THE ONE fair point:** Cerebras bumped equipment useful life from 1.5-3 years to 3-5 years right before IPO and never explained it. ok, mgmt needs to answer that on the Q3 call. but one unexplained accounting estimate ≠ "the chip dies in 2 years." **meanwhile Q2 core rev +103%, \~$25B in RPO.** **tldr:** one legit question about depreciation, wrapped in a garbage model, anonymous quotes and a gotcha about a connector. Nov 12 earnings decides this. Culper doesn't. positions: 600+ shares @ \~$115 avg. holding through earnings. not financial advice, squeeze the shit out of them CRBS \#Edit also nobody's talking about the timing here. yesterday Oracle announced at AI World they're putting Cerebras systems in their data centers and selling it as a cloud service starting 2027. stock ripped +10% on it. then the very next morning Culper drops a 16 page "it's garbage" report on unlock day. funny how that works. Also oracle up today 15% for strong revenue Sources: https://thenextweb.com/news/oracle-cerebras-ai-inference-cloud-2027 https://www.tikr.com/blog/cerebras-gains-10-on-oracle-cloud-deal-as-ai-inference-demands-choice https://www.sec.gov/Archives/edgar/data/0002021728/000162828026056186/cbrsannouncesfinancialresu.htm https://www.sec.gov/Archives/edgar/data/0001341439/000119312526435853/d75538dex991.htm **Here is the PDF for the report:** **https://culperresearch.com/wp-content/uploads/2026/10/Culper\_CBRS\_10-28-2026.pdf**
document fetched 2026-10-29 00:36 · canonical https://reddit.com/r/wallstreetbets/comments/1xhyr19/culpers_cbrs_short_report_is_straight_up_bullshit