Cerebras Systems (CBRS)

Executive Summary & Action Plan

Verdict

SELL / AVOID NEW MONEY — the OpenAI contract validates Cerebras as a differentiated inference platform, but $172.86 still values the company at roughly twice our diluted intrinsic anchor before the 250MW ramp, normalized margins, and cash conversion are proven.

The stock is now below its $185 IPO price, yet all three intrinsic methods remain below spot. The $280 market Bull case is a scarcity-premium path, not fair value.

Price $172.86 | Market cap $38.522B | Target $86.00 | Upside -50.25% IWANNAVY Fair Value · price as of 2026-07-17 regular close / 2026-07-18 05:00 KST · 시장 데이터/Yahoo $172.86 / $172.86 · Yahoo all-class market cap; filing shares × price $39.158B cross-check · Street $291.09 (n=11, reference only)

Abstract

Cerebras has a real moat in latency-first inference: OpenAI uses WSE-3 for Codex-Spark and committed to 750MW worth more than $20B. The investment question is conversion, not contract size. Q1 OpenAI revenue was only $16.9M, Q2 core revenue guidance is approximately $194M, and core gross margin is guided down to 36–38% from 46.5% as capacity costs arrive before utilization. CBRS fell 19.6% in five sessions while SOXX lost 10.2%; no new issuer operating release followed July 9, so the excess decline is attributed—cautiously—to valuation, IPO-price failure, and staged share supply. Our diluted intrinsic median remains $86. The 12M market Base is $144 and probability-weighted target $128.72. A reflex rebound can begin near the post-IPO low, but a durable turn requires $185 recovery, Q2 margin defense, and unlock absorption. Use rallies to reduce; reconsider only below $108 or after operational proof.

Forecast Path

The expected path falls through the Q2/supply window, bottoms near the year-end 250MW proof point, and recovers modestly; the 70% bands measure volatility, not Bear/Bull endpoints.

Scenario probabilities are uncalibrated judgmental overlays. A 44-close issuer bootstrap supplies dispersion only; it removes the sample mean and does not predict direction.

Action Plan

Existing holders should sell into $185–205 rebounds; new capital waits for a filing-backed evidence turn or a price below the $108 intrinsic Bull ceiling.

• 12M Bull/Bear reward-to-risk 0.99:1 (+62.0% to $280 vs -62.4% to $65) | Prob-weighted 12M return -25.5% | P(thesis wrong) 30%, judgmental | Confidence Medium (forecast-model confidence: Low)

Forecast Drivers & Expectations

• Why it fell: CBRS lost 19.6% over five sessions versus SOXX -10.2% and SPY -1.5%. Reuters documented a global AI/semiconductor de-rating around rich valuations, capex durability, and China's Kimi K3; the additional CBRS weakness is an inference from its Q2 margin step-down, IPO-price break, and supply overhang—not a claimed single-cause event study.
• What the market expected: the $20B-plus OpenAI commitment and $25B gross RPO encouraged rapid-conversion assumptions. Q2 guidance instead implies core revenue near-flat sequentially and core gross margin down about nine points.
• What OpenAI has validated: 750MW of committed inference capacity, staged to 250MW by end-2026, 500MW by end-2027, and 750MW by end-2028. The extra 1.25GW is an option, not committed backlog.
• What remains unproven: $25B RPO includes low-margin pass-through economics; Q1 pass-through revenue/cost of $4.111M/$3.991M produced only $0.120M gross profit. OpenAI warrants also reduce revenue and dilute shareholders.
• Rebound sequence: profitable AI leaders and SOXX should stabilize first; CBRS then needs a multi-day $185 recovery, Q2 core GM at the upper end or better, and high-volume unlock absorption. Until then, rallies are technical rather than fundamental.

Last Four Quarters

Only two discrete comparative quarters are public; the prospectus provides Q2+Q3'25 in aggregate and Q4 revenue only. “—” means the issuer has not supplied a comparable standalone figure.

Revenue nearly doubled YoY and accounting profitability improved, but Q1 free cash flow remained deeply negative because cloud capacity is built before it is fully utilized.

Revenue & Profit Mix

Hardware is still larger; cloud/services is the faster-growing and potentially recurring lane, but it carries front-loaded data-center costs.

• Q1 hardware revenue grew 59% YoY; cloud and other services grew 178%. MBZUAI generated 63% of revenue and G42 11%, while the OpenAI MRA generated $16.9M—current concentration is still UAE-heavy even as future RPO shifts toward OpenAI.
• Q1 core cloud gross margin was 52.9%, but total Q2 core GM guidance of 36–38% shows utilization and build timing can overwhelm mix benefits.

Business Lines

• CS-3 / WSE-3 systems: wafer-scale compute removes much inter-chip communication from decode and supports exceptionally fast inference, but hardware revenue is lumpy and dependent on manufacturing and site readiness.
• Cerebras Cloud: the recurring-revenue path for OpenAI and other users. Its moat strengthens with utilization, software compatibility, and token economics—not simply installed MW.
• Strategic distribution: AWS and planned European capacity broaden reach. Europe targets first capacity by end-2026 and 200MW by end-2027; capacity plans are not revenue until energized and used.
• Capital structure: 226.533M reported basic shares rise to about 301.0M in the committed-750MW planning case and 315.6M in the 2GW maximum case before 9M performance RSUs. At spot, those cases imply $52.0B and $54.6B equity values.

IWANNAVY Fair Value

The operating model is unchanged because no material filing followed Q1; the per-share values use the 301M committed-capacity dilution case. Negative forward FCF makes an EV/FCF method unusable, so EV/Sales is the third method.

• Robust intrinsic median: $86; Bear/Base/Bull $53/$86/$108. At 315.6M maximum dilution, the same $25.9B equity value becomes about $82 per share.
• Separate 12M market states: Bear $65 assumes FY28 revenue $2.4B at 6x; Base $144 assumes $3.8B at 10x; Bull $280 assumes $5.5B at 15x. Net cash and diluted shares are scenario-specific, so these are market-price paths rather than additional intrinsic methods.
• Expectations gap: the $291.09 Street mean is 238% above IWANNAVY fair value. It requires faster RPO conversion, sustained scarcity multiples, and cleaner margin/FCF economics than the filing evidence currently supports.

Catalysts, Leading Indicators & Risks

Falsification & Review Rules

• SELL-thesis falsifier: Q2 core GM ≥38%, clear OpenAI revenue acceleration, on-time 250MW by 2026-12-31, and supply absorption above $205. If the bundle occurs, rebuild DCF and raise Bull weight; one datapoint is insufficient.
• Downside confirmation: Q2 core revenue below $190M, core GM below 35%, a committed-date delay, or staged unlocks repeatedly breaking $160.81. Exit residual speculative positions and raise Bear weight.
• Price rule: $185 is the first trend checkpoint, not fair value. A close above it without margin or capacity evidence is a tradeable rebound; below $108 operational evidence still controls entry size. A sustained close below $53 is the severe-failure stop.
• Review cadence: weekly price/supply check; immediate refresh on Q2 date announcement, Q2 results, Form 4/144 activity, or capacity disclosure; formal review by 2026-10-20.

References

• OpenAI–Cerebras partnership · GPT-5.3-Codex-Spark
• OpenAI MRA exhibit · Final IPO prospectus
• Q1'26 results and Q2/FY26 outlook · Q1'26 Form 10-Q
• Europe 200MW plan · Issuer news index
• Reuters on the July AI/semiconductor de-rating · Yahoo price history · Street target distribution