Cerebras Systems (CBRS)

Executive Summary & Action Plan

Verdict

SELL — Q2 proved the OpenAI conversion; $207 prepays it. 12M path $165 (-20%), intrinsic $92; supply, not fundamentals, sets the tape into 11/10.

Price $207.41 | Market cap $49.27B | Target $165.00 | Upside -20.4% 12M market-path target · intrinsic $48–$129 (blended $92) — see Fair Value · Street $291.64 (n=11, reference)

The Take

The tape says distribution: a +73% EPS surprise, two double-digit pops, and the stock still gave the pops back into a CEO-led Form 144 cluster. Consensus reads the $25.4B RPO as destiny and marks $292. It misses the price of proof: at $207 the committed 250MW path is already paid for — EV is 48x FY26E core sales, capex just tripled to $417M a quarter, and 60M+ lockup shares land 2026-11-10. We sell strength and let supply clear.

Action Plan

Holders reduce 50% into $245 strength and exit the rest above $260; new money waits for $129 (top intrinsic leg) or $92 (blended intrinsic); $75 is the severe-failure stop.

• R/R -0.32:1 (12M path target $165 vs 12M bear $75; bull-vs-bear cross-check 0.65:1 — BUY gated off) | Prob-weighted 12M -20.4% | Confidence: Medium | p_thesis_wrong 40%

Forecast Path

The cone falls through the supply window (Form 144 cadence into the 2026-11-10 lockup), bottoms at the 250MW year-end proof, and re-rates only on delivered megawatts. Probabilities are derived, not default: bear cut 40→30 because two July bear legs (core GM <35%, OpenAI stall) were falsified on 8/12; bull raised 12→22 on three-of-four falsifier legs fired plus revision breadth 1.0 (+35.5%/30d); base 48 is the supply-grind residual, and the blend cross-checks p_thesis_wrong 40%.

By the Charts

Core GM fell 46.5→40.6 but beat the 36–38 guide; FY26 guided 41–43 — the margin-collapse bear leg is retired.

GAAP net loss $450.5M is optics: ~$444M is one-time IPO comp and warrant contra-revenue; core net loss $6.9M.

Mix inverted in one quarter: cloud 43%→70% of GAAP revenue — the hardware vendor is becoming an inference utility.

Committed-capacity intrinsic $92 log-median; the other $115 of the price is the un-committed option book.

Fair Value — Two Layers

Layer 1 (intrinsic): DCF $48, 30x FY28E core EBITDA $92, 12x FY27E sales $129 — log-median $92, raised from $86 on the guide raise and OpenAI conversion, while Q2's revealed capex intensity cut the DCF leg. Layer 2 (12M path): $165 = 30% × $75 + 48% × $163 + 22% × $293. The layers differ by +80% because the market prices the un-committed option book — 1.25GW additional-capacity warrants and a 600MW+ contracted pipeline — that a committed-capacity DCF excludes; we assume that premium persists but compresses. Decomposition: growth -3.6% × re-rate +9.9% × cash/dilution -23.9% = -19.4% vs -20.4% stated (gap 1.0pp ≤ 5pp). Street $291.64 requires bull inputs ($5.8B FY28 revenue held at 15x) — reference, not base.

촉매 캘린더

단기 (0~6개월)

Supply and proof points collide inside six weeks: the earnings-triggered Form 144 cadence runs into the 11/10 lockup, the Q3 print lands roughly a day later, and the OpenAI 250MW contract deadline closes the year. This chain — not fundamentals — sets the path into December.

Risks & Thesis Breakers

• Supply: staged windows + 11/10 lockup vs 60.06% insider hold, short float 13.6% — ~65% odds of a -15% to -30% drawdown ($145–175). / Valuation vs conversion: EV 48x FY26E sales, Q2 FCF -$477M; an AI-capex wobble re-runs July — ~50% odds, -25% to -50%. / Concentration: MBZUAI 34% + OpenAI 32% + G42 9% of Q2 revenue; two customers hold 76% of receivables — ~35% odds, -30% to -50%.
• Breakers (4-of-5 kills the SELL): Q3 core GM ≥40% and revenue ≥$216M; OpenAI sustains ≥$60M/qtr with positive core gross profit; 250MW confirmed on time by 2026-12-31; 11/10 unlock absorbed above $205 without a Form 144 cascade; quarterly FCF burn better than -$50M by the Q4 print.

Appendix