Cerebras Systems (CBRS)

Executive Summary & Action Plan

Verdict

SELL — The business is real and scaling violently, but at $198.53 the stock trades above even our bull fair value ($117); every audited valuation path lands 41–71% lower, with a ~171M-share lock-up unlock looming around the Q3 print. Shares jumped +9.3% today on the Europe 200MW expansion and Flex 7x manufacturing news (verified) — strength to sell into, not chase.

Price $198.53 | Market cap $43.6B | Target $93 | Upside -53% IWANNAVY Fair Value · price as of 2026-07-10 · Street consensus $291.09 (n=11, reference) — we sit 68% below Street: post-IPO underwriter initiations price full $24.6B-backlog conversion at retained premium multiples, while our DCF already compounds revenue ~26x by FY2032

Abstract

Cerebras sells wafer-scale AI compute — CS-3 systems plus a fast-growing inference cloud — and is scaling violently: Q1'26 revenue of $193.4M grew 94% YoY, adjusted EBITDA turned positive for the first time, and a $24.6B backlog anchored by OpenAI's 750MW, >$20B commitment gives visibility few new listings ever show. Growth quality is the problem: OpenAI alone was 63% of Q1'26 revenue (two UAE entities were 86% of FY2025), guided core gross margin falls to 36–38% in Q2'26 as leased datacenter capacity runs ahead of utilization, and free cash flow stays deeply negative (consensus -$1.4B for FY2026). Triangulating a 13.5%-WACC DCF ($93), 30x FY2028E EV/EBITDA ($103), and 12x FY2027E EV/Sales ($91), fair value is $93 — 53% below the last close and 68% under the $291 Street mean. With ~171M lock-up shares unlocking around the Q3 report, 31.5% short float, and even the bull case ($117) below today's price, we rate CBRS SELL; re-engage below $93.

Action Plan

Exit longs into post-news strength; do not initiate above fair value — re-engage only below $93.

• Risk/Reward 0.4:1 for longs (+23% squeeze upside vs -53% to fair value) | Prob-weighted 12M return -57% (Bear 35% · Base 45% · Bull 20%) | Confidence: Medium

---

Last Four Quarters

Only two discrete quarters are public (May 2026 IPO); the prospectus discloses Q2+Q3'25 only in combination. Direction is unambiguous: revenue nearly doubled YoY, operating margin closed from -28.8% to -8.4%, and Q1'26 was the first positive adjusted-EBITDA quarter (+$12.7M core) — yet FCF stays deeply negative because capex ($132M in Q1'26) runs ahead of revenue by design. The June 23 selloff was not about these numbers but about guidance: Q2'26 core gross margin guided down to 36–38% (from 47%) on temporary rent-back of systems during the datacenter buildout (Q1'26 8-K, 2026-06-23).

---

Revenue & Profit Mix

Hardware (CS-3 wafer-scale systems) is still the larger line at 57% of Q1'26 revenue, but Cloud & Inference Services (+178% YoY) is closing fast and now drives the growth story — and the margin risk. Note: TTM GAAP net income of +$248M is an accounting artifact of a one-time, non-cash +$363M G42 forward-contract gain in FY2025; operations lose money in every quarter.

---

Business Lines

• Wafer-Scale Systems (~57% of Q1'26): CS-3 appliances on the WSE-3 chip — ~900k cores and 44GB on-chip SRAM eliminate the GPU memory-bandwidth wall; ~43% segment GM, lumpy mega-deployment sales. Must know — one UAE or hyperscaler order can swing a quarter; the Flex partnership lifts CS-3 output ~7x by end-2026, but recognition depends on a handful of buyers.
• Cloud & Inference Services (~43%): inference ~20x faster than GPU clouds per company benchmarks; +178% YoY. Fixed-cost heavy — FY2025 cloud GM whipsawed 68%→26%→16%→21% on utilization, with ~$2.3B non-cancelable leases. Must know — this line, not hardware, is where the margin inflection lives; new capacity earns nothing while accruing lease/power costs until tenants fill it.
• Strategic Capacity Agreements (backlog): OpenAI 750MW (>$20B, +1.25GW option), AWS inference partnership, Europe 200MW by end-2027 — underwriting the $24.6B RPO. Must know — RPO at ~48x trailing revenue is simultaneously the entire bull case and the concentration risk; conversion requires self-funding massive capex ahead of revenue.

---

IWANNAVY Fair Value

Fair value is the log-space median of three methods on 277M fully diluted shares (basic 219.6M + OpenAI ~$0-strike warrant 33.4M + options/RSUs): a 7-year FCFF DCF that already assumes revenue compounds from $604M TTM to ~$22B by FY2032, a 30x multiple on FY2028E EBITDA (anchored to NVDA 29x/AVGO 46x, not the inflated trailing prints of MRVL/AMD), and growth-adjusted EV/Sales — EV/FCF is explicitly infeasible because forward FCF stays negative through FY2028E. We differ from the $291 Street mean because we refuse to pay today for undiscounted, single-customer backlog conversion; even so, our bull ($117) sits below the current price.

• Thesis breaker: A Q2'26 print (est. 2026-09-02) with core GM at/above the 38% guide-top, cloud revenue still accelerating, and non-OpenAI/UAE mix rising would prove the rent-back margin dip temporary and force our ramp toward the full-conversion bull (~$185); OpenAI backlog slippage confirms the bear ($58).

---

Catalysts & Risks

---

References

• Q1'26 results 8-K — guidance shock (SEC, 2026-06-23) · 424B4 IPO prospectus — cap table (SEC, 2026-05-14)
• Cerebras pops 68% in Nasdaq debut (CNBC, 2026-05-14) · IPO lockup comes fast — ~171M shares (Seeking Alpha, 2026-06-26)
• Europe 200MW expansion (2026-07-09) · Analyst forecasts — FY26 consensus $863M (2026-07-10) · Why Cerebras is skyrocketing today (Motley Fool, 2026-07-09)