OPENAI 29 min read

OpenAI Group PBC (OPENAI) — Investment Tree v1

Stage 7 final essay. Bilingual companion: tree_v1_zh.md. Date: 2026-05-15 · Anchor price: $1,000/share placeholder (~$1T pre-marketing rumor per Reuters Oct 2025; 1B FD-share equivalent) · Public S-1 expected H2 2026 window Archetype: AI utility or compute trap with Foundation warrant overhang — closest analog Tesla 2010 IPO (capital-intensive growth + CEO-led + multi-product platform + multi-year FCF-negative pattern) blended with Veeva/Anthropic PBC governance dimension

SOURCE QUALITY: TIER-C HEAVY pre-S-1. Public S-1 not yet filed (rumored H2 2026 window per Reuters Oct 2025). Most segment-level financials are Tier C trade-press triangulation. Capital structure mechanics (Foundation warrant precise terms, Microsoft revenue-share precise contract terms, Stargate JV ownership economics) are Tier C pending S-1. Most operational milestones are Tier B from OpenAI press + management public commentary + Reuters/The Information primary reports. Refresh REQUIRED when S-1 lands — verdicts may move materially, particularly L1A 1.5 (governance investability), L1B 1.1-1.2 (capital structure waterfall mechanics), L1A 1.2 (GM trajectory disclosure).

Conflict-of-interest disclosure: OpenAI is the named #1 frontier-LLM competitor to Anthropic (which makes Claude, the assistant writing this report). Per CLAUDE.md, the CoI rule applies in REVERSE here — this analysis must avoid analytical OVER-skepticism on OpenAI's defenses and OVER-charity to Anthropic's enterprise position. Specific load-bearing claims about Anthropic enterprise win-rate (L1D 1.1) and Anthropic comparable multiples (~$200B private mark) have been explicitly re-tested against ChatGPT-IC-memo + Reuters + The Information + multiple enterprise software analyst coverage and corroborated externally. Where the analysis lands favorable to Anthropic, it is supported by external evidence; where the analysis lands favorable to OpenAI, the bull-case mechanism + falsification conditions are explicitly stated. Symmetric scrutiny applied.

The H-0 confidence is 58% — partial-support range, mildly bearish at rumored $1T anchor but with genuine bull upside if 3 load-bearing falsifiers fire favorably (FF1 GM expansion + FF2 Foundation warrant bounded + FF3 capability lead durable). The structurally distinctive finding vs. SPACEX's pre-IPO tree (built one day prior) is +1.2:1 favorable asymmetry — meaningfully more favorable than SPACEX's 0.75:1 unfavorable but less favorable than corpus median 1.5-2.0:1. The mechanism is: smaller pre-marketing premium (+17% vs SPACEX's 250%+) + less tightly correlated bear-stack (two independent common causes vs SPACEX's one) + genuine bull-case upside potential if GM expansion + capability lead + Foundation/MSFT bounded all hold.

I. One-sentence verdict

OpenAI's rumored ~$1T IPO valuation prices the company at SaaS-with-AI-narrative multiples (~40× current annualized revenue) for what is operationally a hyperscaler-style capital-intensive infrastructure company (~33% gross margin; $600B compute spend by 2030; $25B annualized revenue) with a 5-priority capital-structure waterfall that places common shareholders FIFTH (compute providers → Microsoft revenue-share through 2030 → Foundation valuation-milestone warrant → employees → common shareholders); while operational durability is genuine (ChatGPT brand + first-mover scale + Stargate compute ahead-of-schedule + post-2023 reformed board), the moderately-correlated bear-stack (GM stuck + Anthropic+Gemini+open-weight compounding + Foundation+MSFT layered overhang) combines with Tesla-analog-with-worse-durability framing to produce a +1.2:1 favorable asymmetry at IPO entry with genuine bull upside if 3 falsifiers fire; suitable as a 0% pre-IPO watch position with 0.5-1% conditional entry at IPO IF S-1 discloses bounded Foundation warrant + Microsoft revenue-share + favorable Stargate JV economics, scaling to 1.5-2.5% over 90 days post-T+90 if GM Q1 2027 print + S&P 500 inclusion + Anthropic enterprise plateau resolve favorably — NOT a buy at the rumored $1T anchor without S-1 disclosure quality confirmation and NOT a high-conviction concentrated bet pre-Q1-2027-GM-print.

II. Company snapshot

OpenAI Group PBC is a San Francisco-domiciled Public Benefit Corporation in late-stage pre-IPO preparation. Reuters reported in October 2025 that OpenAI is "laying groundwork for a juggernaut IPO at up to $1 trillion valuation," with the listing window most-likely H2 2026. No confidential S-1 has been publicly confirmed as filed; trading debut projected Q3-Q4 2026 under base assumptions.

The corporate structure is non-standard with three layered entities:

• OpenAI Foundation (non-profit 501(c)(3) parent; 26% controlling stake in Group PBC; appoints all directors; holds dilutive valuation-milestone warrant)
• OpenAI Group PBC (operating subsidiary that would IPO; PBC fiduciary-duty balances shareholder profit + public benefit)
• Microsoft revenue-share contract (through 2030 independent of equity stake; Microsoft holds 27% Group equity; license now non-exclusive per November 2025 partnership update)

The parent comprises three operationally distinct revenue sleeves:

• API / Developer Platform ~$10-12B FY2025E (hyperscaler-tier cost structure; pricing-pressured by Anthropic Claude + Gemini + open-weight)
• ChatGPT Consumer + Enterprise ~$10-12B FY2025E (brand-anchored subscription; Plus $20/mo + Pro $200/mo + Enterprise $60-100/user/mo)
• Strategic / Government / Other ~$2-3B FY2025E (DOD pilots + allied government + Sora + Operator + Microsoft revenue-share contribution + Apple Intelligence)

Annualized revenue topped $25B as of February 2026 per Reuters citing The Information. Gross margin reported ~33% end-2025 per Reuters citing internal projections — approximately half typical SaaS GM and roughly equal to hyperscaler infrastructure GM. Compute spend projected at ~$600B by 2030 per CNBC citing internal projections (Stargate JV $500B / 10GW+ committed plus Microsoft Azure + chip-supplier commitments).

Capital structure post-IPO: ~1B fully-diluted shares; PBC structure; standard 180-day lockup expected. Cash-flow waterfall places common shareholders FIFTH (compute → MSFT revenue-share → Foundation warrant → employees+capex → common). This is the load-bearing capital-structure fact per mispricing.md Type B primary mispricing mechanism.

Margin asymmetry is the load-bearing structural fact: the ~33% GM is hyperscaler-economics not SaaS-economics; the $600B compute spend by 2030 dwarfs revenue scaling (revenue must grow 4-5× to ~$100B+ by 2030 just to balance announced compute spending); OpenAI may need to spend like a hyperscaler without owning hyperscaler-diversified cash flows. This is the H-0 candidate "Compute = trap, not moat" thesis per ChatGPT-IC-memo Tier-B external research framework.

III. The five facts that drive everything

1. Annualized revenue $25B as of February 2026 — load-bearing scale milestone supporting any meaningful valuation framework; growth +60-70% YoY through 2025. ✅B — Reuters citing The Information.

1. Gross margin ~33% end-2025 — hyperscaler-economics not SaaS-economics; ChatGPT-IC-memo failure signal threshold is GM <35% as usage grows. The single load-bearing economic indicator for the IPO. ⚠️B — Reuters citing internal projections; trajectory undisclosed; FF1 falsifier centered here.

1. Compute spend ~$600B by 2030 — Stargate JV $500B + Microsoft Azure + chip-supplier commitments; revenue must grow 4-5× by 2030 to balance. ⚠️B — CNBC citing internal projections.

1. Foundation 26% + Microsoft 27% + others 47% structure with 5-priority cash-flow waterfall — common shareholders fifth (after compute providers → MSFT revenue-share → Foundation warrant → employees). The load-bearing capital-structure mechanism. ⚠️B — OpenAI corporate disclosure on structure; precise mechanics Tier C pending S-1.

1. Anthropic ~$9B annualized growing 4-5× YoY at $200B private mark + Gemini distribution-anchored + open-weight at GPT-4 capability — three independent competitive threats with compound probability 45-55% for 2+ firing materially over 2026-2036. ⚠️B — Reuters + The Information + ChatGPT-IC-memo + public benchmarks.

IV. The H-0 thesis

H-0 (one sentence): The market is anchoring OpenAI at ~$1T on a SaaS-with-AI-narrative framing (~40× revenue) appropriate for software platforms with 75-85% gross margins, but the operational reality is closer to a hyperscaler-style capital-intensive infrastructure company at ~33% gross margins with $600B compute spend by 2030; combined with a 5-priority capital-structure waterfall that places common shareholders FIFTH (compute → Microsoft revenue-share through 2030 → Foundation warrant → employees → common), the SoTP central tendency lands $700-900B pre-discount and $750-850B post-discount — making the rumored ~$1T pre-marketing anchor a mild overprice with bear-stack downside more probable than the consensus implies; the asymmetric-upside thesis (Stargate compute moat + ChatGPT brand pricing power + Microsoft non-exclusive distribution widening + GM expansion to 45-55%+) is genuinely possible if 3-4 load-bearing assumptions hold simultaneously.

Mispricing taxonomy: Type B — GOVERNANCE DISCOUNT OVERWEIGHT (Foundation+MSFT layered overhang) primary (per mispricing.md). The 5-priority capital-structure waterfall + PBC fiduciary-duty constraint + Foundation warrant dilution mechanics applied silently over 10-year window. Magnitude: 12-20% structural discount sustained.

Secondary mechanism: Type C — VALUATION-COMP CONFUSION (hyperscaler-vs-SaaS-vs-Tesla-analog reframe). OpenAI's economic profile is genuinely ambiguous; SaaS-comp implies $700B-1.2T; hyperscaler-comp implies $200-350B; Tesla-analog implies $300-500B core + ~$100-200B optionality.

Tertiary mechanism: Type A — SEGMENT-REPORTING BLINDNESS ON THREE-SEGMENT STACK. Less acute than SPACEX because sleeves share inference-cost base. API ($10-12B) + ChatGPT Consumer+Enterprise ($10-12B) + Strategic ($2-3B) at different growth/multiple profiles.

Fourth mechanism: Type D — INDEX-DEMAND MECHANICAL (PBC + Foundation methodology uncertain). S&P 500 inclusion for foundation-controlled PBC at OpenAI's scale is novel; most-likely normal inclusion per Veeva 2018 precedent but methodology delay possible.

The four mechanisms compound or offset. Critically, the bear-stack (B+C against) is moderately correlated through two independent common causes: (a) AI economics commoditization (drives A+B+C+D negative through GM compression + competition + Stargate-as-trap); (b) capital structure design (drives B independently of operating performance). These two common causes are independent of each other. The bear-stack is therefore moderately correlated WITHIN each common cause but less correlated overall than SPACEX's tightly-correlated dual-class-driven bear stack. This is the structural reason OpenAI's asymmetry is +1.2:1 favorable (vs SPACEX's 0.75:1 unfavorable) — the bear case is real but bounded; the bull case has genuinely accessible upside.

5 falsification conditions (H-0-breakers):

• FF1 GM expansion crystallizes: Q1-Q4 2027 audited GM print shows clear path to 45-55%+ by 2028 (38%+ FY2026; 42%+ FY2027) → SaaS-comp framework becomes appropriate → equity value supports $1-1.3T+
• FF2 Foundation warrant + Microsoft revenue-share economically bounded: S-1 reveals Foundation warrant strikes at $3T+ thresholds (effectively non-binding) AND Microsoft revenue-share capped at <8% with sunset at 2030 → governance discount narrows to <5%
• FF3 Capability lead durable: GPT-5 (Q2-Q3 2026) demonstrates measurable capability premium over Claude 5 + Gemini 3.5 AND enterprise win-rate vs Anthropic stabilizes or improves through Q4 2026 → compound competition probability <30%
• FF4 Stargate compute economics favorable: S-1 reveals OpenAI has meaningful equity stake (>30%) OR contractual cost-sharing producing per-GPU-hour economics meaningfully below leased Azure equivalent → compute "moat" becomes economically real
• FF5 Microsoft 2030 revenue-share expiry favorable: Multi-year disclosure path through 2028-2030 indicates OpenAI-favorable renegotiation (rate reduces; cap tightens; sunset confirmed) → 2030 becomes positive catalyst $50-100B PV unlock

H-0 is partially supported if 1-2 falsifiers fire; rejected if 3+ fire. Most-likely path forward: 1-2 of FF1/FF3 fire; FF2/FF4/FF5 don't all fire → H-0 partially supported on 18-36 month window → "consider sized entry post-S-1 conditional on disclosure quality; otherwise wait for Q1-Q4 2027 GM inflection signal."

V. Tree — five branches

H-0: OPENAI ~$1T rumored anchor is governance-discount-overweight + hyperscaler-vs-SaaS comp confusion;
     SoTP central tendency $700-900B pre-discount / $750-850B post-discount;
     bear stack (B+C against) moderately correlated; bull case requires 3-4 load-bearing assumptions hold
│
├── L1A — 5 Load-Bearing Assumptions (Foundation Solid?)                       ⚠️C partial; mixed
│   ├── 1.1 Revenue $60-75B by IPO+24mo + enterprise/API ≥50%                  ✅C supported (trajectory on track)
│   ├── 1.2 GM 45-55%+ by 2027-2028                                            ✗C NOT SUPPORTED YET (FF1 falsifier)
│   ├── 1.3 Compute moat not trap; cash burn ≤$25-35B/yr                       ⚠️C partial (Stargate ahead of schedule)
│   ├── 1.4 Model leadership top-2 across capability axes                       ⚠️C partial (currently top-2; volatile)
│   └── 1.5 Governance investable (S-1 reveals clean structure)                 ⊗C not testable pre-S-1
│
├── L1B — Capital Structure Trap (Compute-vs-Moat + Waterfall)                  ⚠️C partial; warrant mechanics ⊗
│   ├── 1.1 Microsoft revenue-share ≤10% + favorable 2030 renegotiation         ⚠️C partial (Nov 2025 non-exclusive favorable)
│   ├── 1.2 Foundation warrant dilution <10% over 10-year window                ⊗C not testable pre-S-1
│   ├── 1.3 Stargate JV economics moat-supporting                               ⚠️C partial (operational ahead; economic unknown)
│   └── 1.4 Escape Tesla/Amazon year-10-to-FCF timing                           ✗C NOT SUPPORTED (current trajectory)
│
├── L1C — Governance + Foundation Oversight (PBC Mission-Charter)               ✅B partial-to-supported
│   ├── 1.1 Post-2023 board reform durable through 2030                         ✅B supported (institutional bench)
│   ├── 1.2 PBC fiduciary-duty bounded; allows shareholder-return optimization  ⚠️B partial (Veeva precedent positive)
│   ├── 1.3 Altman + leadership continuity through IPO+24mo                     ✅B supported (Friar + Brockman + bench)
│   └── 1.4 No additional Sutskever-style technical-talent flight               ⚠️B partial (brain-drain bounded)
│
├── L1D — Strategic Inflection (Anthropic / Gemini / Open-Weight)               ⚠️B partial; Anthropic wins real
│   ├── 1.1 Anthropic enterprise win-rate stabilizes/declines                    ✗B NOT SUPPORTED (FF3 falsifier)
│   ├── 1.2 Gemini distribution-anchored displacement bounded                    ⚠️B partial (compounding pressure)
│   ├── 1.3 Open-weight capability gap maintained 12-mo rolling                  ⚠️B partial (compressing on some axes)
│   └── 1.4 Capability moat events (GPT-5) restore premium for 6+ mo             ✅C supported (historical pattern + Q2-Q3 2026 release)
│
└── L1E — Mispricing Path + IPO Mechanics                                       ⊗C resolves T+90 to T+180 binaries
    ├── 1.1 S&P 500 inclusion fires T+90 normally for PBC                       ⊗C not testable pre-decision
    ├── 1.2 Lockup expiry T+180 absorption manageable                            ⚠️C partial (institutional holders stable)
    ├── 1.3 First-day pop bounded 10-30%                                         ⚠️C partial (AI-narrative scarcity)
    └── 1.4 Sell-side desks rotate to fundamentals 18-24 months                  ⊗C not testable pre-rotation

Total: 4 ✅ / 10 ⚠️ / 3 ✗ / 4 ⊗ across 21 leaves
H-0 verdict: PARTIALLY SUPPORTED, ~58% confidence; mildly bearish at $1T anchor with genuine bull upside

The three ✗ verdicts are all on load-bearing falsifiers but they are not all simultaneously fatal. L1A 1.2 (GM expansion) is the FF1 falsifier — currently trending unfavorable but resolvable through Q1-Q4 2027 GM prints. L1B 1.4 (Tesla-analog with worse durability) is conditional on FF1 + FF3 — fires if GM stays stuck AND competition compounds. L1D 1.1 (Anthropic enterprise win-rate compounding) is the FF3 falsifier — currently trending unfavorable but resolvable through Claude 5 ceiling event.

The four ⊗C leaves are honest disclosure of "not testable at this tier." L1A 1.5 (governance investability) + L1B 1.2 (Foundation warrant mechanics) + L1E 1.1 (S&P 500 inclusion) + L1E 1.4 (sell-side rotation) require S-1 disclosure or post-IPO events to resolve. No K.3.1 fatal flag fires — unlike SPACEX's xAI acquisition candidate fatal flag, OpenAI's capital structure is governance-discount-imposing not malfeasance-imposing.

VI. Key findings

Finding 1 — The SoTP central tendency is moderately below the $1T pre-marketing anchor

Per peers.md 4-peer-set analysis: probability-weighted sum across framings yields central tendency ~$750-900B post-discount before adding AI-narrative scarcity premium. At Anthropic-comparable 22× revenue (the closest direct comp) + scale/brand premium (1.5-1.8×) = $825-990B — consistent with rumored $1T anchor with modest premium. At hyperscaler-comp 8-14× revenue = $200-350B; at SaaS-AI-narrative 28-48× revenue = $700B-1.2T; at Tesla-analog (capital-intensive growth with 10-year-to-FCF) = $300-500B core + $100-200B optionality.

The $1T rumored anchor is at the UPPER END of fundamentals-supported range but is not extreme like SPACEX's $1.75T (~3.5× Q1 2026 secondary marks). The pre-marketing premium from $852B (March 2026 round) to ~$1T is +17% — modest. The analytical mispricing is in the multiple-framing question (hyperscaler vs SaaS) rather than the absolute level vs recent marks.

Conditional probabilities derived from market price: market is implicitly assigning ~28-32% Bull probability + ~10-15% Bear probability. Our analytical probability is 22% Bull + 28% Bear. The 16pp Bear-gap is the load-bearing analytical disagreement — market consensus implicitly treats Bear scenarios at ~12% probability; analytical framework places Bear at 28%.

Finding 2 — The moderately-correlated bear-stack is the structurally distinctive risk

OpenAI's bear-stack is correlated through TWO independent common causes:

1. AI economics commoditization (drives FF1 negative via GM compression + drives FF3 negative via competition + drives FF4 negative via Stargate-as-trap)
2. Capital structure design (drives FF2 directly; independent of operating performance)

These two common causes are independent of each other — Foundation warrant + Microsoft revenue-share would still apply EVEN if OpenAI achieves GM expansion + revenue growth; conversely, AI commoditization would hit OpenAI EVEN if governance were ideal. The bear-stack is therefore moderately correlated within each common cause but less correlated overall than SPACEX's tightly-correlated dual-class-driven bear stack.

This is the structural reason OpenAI is "consider sized entry post-S-1 conditional on disclosure quality" rather than SPACEX's "wait for T+90." The bear is real and the central-tendency is below headline, but the bull case has more upside if the load-bearing GM-trajectory + capability-lead-durability + governance-discount-benign mechanisms hold.

Finding 3 — ChatGPT brand + post-2023 reformed board are genuinely durable; the GM trajectory is the load-bearing risk

L1C (Governance + Foundation Oversight) is the most-supported branch in the tree: 2 ✅B verdicts on Tier B evidence (post-2023 reformed board composition + Altman/Brockman/Friar/Weil leadership continuity). The board reform is meaningfully institutional — Bret Taylor + Larry Summers + Adam D'Angelo + Sue Desmond-Hellmann + Paul Nakasone + Fidji Simo bring public-company governance experience that pre-crisis board lacked. The Sutskever + Murati departures (May 2024 / Sept 2024) were real losses but the bench absorbed them.

This is a counterweight to the capital-structure overhang. The operational governance of OpenAI is structurally improved from the pre-2023 state; the analytical concerns are predominantly at the capital-structure-design layer (Foundation warrant + Microsoft revenue-share + PBC fiduciary-duty) which are contractual not management-quality issues. The risk-mitigation path is to wait for the capital-structure mechanics to disclose (S-1) and for the GM trajectory to print (Q1 2027) before sizing the operational durability.

Finding 4 — Mega-IPO base rate is moderately unfavorable but less acute than SPACEX

Per peers.md precedent analysis: $50B+ IPOs 2010-2025 have median T+12M return -15 to -25%. Snowflake +112% T+1 then -65%; Coupang +40% T+1 then -55%; Rivian +29% T+1 then -90%; Meta -11% T+1 then +400% T+5y; Aramco +10% sustained. The base rate skews bearish on a 12-24 month entry horizon but is meaningfully less extreme than SPACEX's setup (which was 3× larger than any prior US IPO).

Post-debut entry windows (T+6M to T+18M) have historically been more favorable for mega-IPOs as the initial premium unwinds and fundamentals stabilize. For OpenAI specifically, the highest-information-value period is the S-1 disclosure (Foundation warrant + Microsoft revenue-share) plus the Q1-Q4 2027 GM print. The entry-conditional-on-S-1-disclosure-quality is structurally favorable — unlike SPACEX where multiple binary catalysts had to sequence positively, OpenAI's primary entry decision is a single binary at S-1 release.

Finding 5 — The S-1 disclosure of Foundation warrant + Microsoft revenue-share is the highest-leverage datapoint

The single S-1 datapoint that most moves the tree's verdict is the Foundation valuation-milestone warrant strike threshold + exercise mechanics combined with the Microsoft revenue-share rate + cap + sunset mechanics. Favorable disclosure (Foundation warrant strikes ≥$3T + Microsoft revenue-share ≤8% + 2030 sunset confirmed) → L1A 1.5 verdict upgrades from ⊗ to ✅B → L1B 1.1 + 1.2 upgrade → durability Q3 + Q5 upgrade to 4/5 + 3/5 → aggregate to 18/25 → fatal-flag candidate dismissed (already not fired but probability of future firing reduced) → tree verdict shifts from "wait" to "consider 0.5-1% entry at IPO." Unfavorable disclosure (Foundation warrant strikes at $1-1.5T + Microsoft revenue-share 15-18% + 2035+ extension) → governance discount embeds 12-20% durable → L1A 1.5 + L1B 1.1 + 1.2 verdicts move to ✗C → tree verdict: "AVOID at IPO; reassess only on Microsoft 2030 favorable renegotiation event."

This single binary in the S-1 (which lands in H2 2026 per Reuters rumor) potentially shifts the H-0 confidence by ±10-15pp. Refresh post-S-1 is mandatory.

VII. Three valuation scenarios

(See scenarios.md for full analytical breakdown.)

Probability-weighted 12-month expected return: -5% to +8% (median ~+1.5%) — modestly positive expected value at debut entry.

Asymmetry +1.2:1 favorable — Bull upside (+30 to +60%) modestly larger than Bear downside (-35 to -55%); meaningfully more favorable than SPACEX's 0.75:1 unfavorable but less favorable than corpus median 1.5-2.0:1. The mechanism is smaller pre-marketing premium + less tightly correlated bear-stack + genuine bull-case upside potential.

The asymmetry supports CONDITIONAL entry at IPO despite real operational + capital-structure risks: even though the load-bearing question (GM trajectory) is currently trending unfavorable, the S-1 disclosure quality is the highest-leverage decision point and provides a single binary to act on. Mathematics of probability-weighted positive expected-return supports conditional entry at S-1 disclosure quality confirmation.

VIII. Triggers and red flags

(Full detail in triggers_redflags.md.)

Triggers (Bull-case fires):

• T1 (Q2-Q3 2026): GPT-5 release demonstrates sustained capability premium (FF3 falsifier)
• T2 (H2 2026): Public S-1 lands with adequate Foundation warrant + Microsoft revenue-share disclosure (FF2 falsifier)
• T3 (~4-6 weeks post-S-1): Roadshow indicative price range settles below pre-marketing $1T
• T4 (~T+90): S&P 500 inclusion fires normally for Foundation-controlled PBC
• T5 (Q1 2027): GM print shows clear expansion trajectory (FF1 falsifier)
• T6 (Q3-Q4 2026): Anthropic enterprise share plateaus on Claude 5 capability ceiling
• T7 (2028-2030): Microsoft 2030 revenue-share renegotiation favorable (FF5 falsifier)
• T8 (2026-2028 rolling): Stargate compute scale + economics favorable (FF4 falsifier)
• T9 (2026-2028 rolling): Sora + Operator + Agent enterprise standardization milestone
• T10 (2026-2028 rolling): Sovereign-AI inflection — DOD + allied government scaling

Red flags (Bear-case fires):

• RF1 (H2 2026): Public S-1 reveals Foundation warrant + Microsoft revenue-share unfavorable
• RF2 (Q1 2027): GM print at 32-34% — failure signal fires (FF1 falsifier negative)
• RF3 (Q4 2026): Anthropic annualized exceeds $25B (FF3 falsifier negative)
• RF4 (2026-2027): Open-weight 15+ top-100 API migration announcements
• RF5 (T+90 to T+180): S&P 500 inclusion delays 12+ months
• RF6 (2026-2030 rolling): Altman-key-person event OR repeat board crisis
• RF7 (2026-2030 rolling): Foundation warrant exercise event

IX. What this means for position sizing

Per durability_test.md: aggregate score 16/25 Medium durability with caveats; 0 fatal flags fired. Investability for 5-10 year hold is CONDITIONAL on S-1 disclosure quality for Foundation warrant + Microsoft revenue-share + Stargate JV economics.

Position-sizing recommendation

• Pre-IPO (current state): 0% — WATCH ONLY. No exposure pre-S-1.
• At IPO conditional on T2 favorable (S-1 discloses Foundation warrant + Microsoft revenue-share bounded): Consider 0.5-1% conditional entry.
• At IPO conditional on T2 + T3 both favorable (S-1 favorable + roadshow ≤$900B): Consider 1-1.5% entry at debut.
• Post-T+90 (T1 fired GPT-5 + T4 fired S&P 500 + T5 progressing on GM): Scale to 1.5-2% over 60 days post-confirmation.
• Post-T+90 (RF1 OR RF2 fires): AVOID through T+12-18M; reassess only on FF5 governance-improvement event.
• Hard cap: 3% if 5+ triggers fire across 12-month window with 0-1 flags firing.

For Ming specifically (per CLAUDE.md correlated-exposure-acknowledged):

• OPENAI adds AI-frontier-LLM-platform + capital-intensive-growth factor exposure — meaningful overlap with NVDA + MSFT + GOOGL (all AI-infrastructure beneficiaries; ~45% existing AI/semi sleeve in Ming's portfolio). OPENAI adds direct frontier-AI-platform factor that doesn't currently exist in portfolio = additive but partly correlated.
• Capital-intensive growth + multi-year FCF-negative profile overlaps with Ming's pre-existing TSLA exposure (if any) — Tesla-analog framework directly applies.
• Hard cap pre-IPO: 0%. Hard cap conditional at IPO: 1-1.5%. Hard cap post-T+90: 2.5%. Hard cap post-12M with structural resolution: 3%.

Concentration-risk note (per K.3.4)

OPENAI correlation profile:

• AI compute / infrastructure (Stargate + Microsoft Azure): Overlaps NVDA + MSFT on demand-for-GPUs vector; correlation high
• AI consumer brand (ChatGPT): Overlaps GOOGL Search (Gemini-cannibalization) + AAPL (Apple Intelligence integration); correlation moderate
• AI enterprise platform (API + ChatGPT Enterprise): Overlaps MSFT (Copilot frenemy) + GOOGL (Workspace AI) + Anthropic (private); correlation high
• AI sovereign / government (DOD + allied): Overlaps PLTR + LMT + RTX (sovereign-AI providers); correlation moderate
• Mega-IPO software precedent (Snowflake-style narrative-to-fundamentals derating): Idiosyncratic; doesn't correlate with existing portfolio

The OPENAI addition is partly correlated with existing AI/semi sleeve but also genuinely additive on direct frontier-AI-platform factor. The most-load-bearing correlation is AI-compute-CapEx-cycle (Stargate is the largest AI infrastructure commitment ever; ties OPENAI to broader AI infrastructure cycle). At 2-3% OPENAI size, the AI-concentration is bounded given existing exposure; above 5% it becomes meaningfully concentrated.

X. Long-term holdability verdict

Per durability_test.md: aggregate score 16/25 Medium durability with caveats. 0 fatal flags fired. Unlike SPACEX (where xAI acquisition is candidate Q3 fatal flag), OpenAI's capital structure is governance-discount-imposing not malfeasance-imposing.

Investability for 5-10 year hold: CONDITIONAL. If S-1 discloses bounded Foundation warrant + Microsoft revenue-share + favorable Stargate JV economics → Medium-High durability + 0 fatal flags → suitable for selective hold at moderate size (1-2.5%). If S-1 reveals unfavorable disclosures → Medium durability + governance discount embeds → "long-term hold not investable at conviction sizing; minimal speculative position only" applies.

Top 3 reasons supporting: (1) ChatGPT consumer brand + first-mover scale + multi-product platform (Sora + Operator + Agent + Voice) produce durable competitive position through 2030+ with brand-recognition approaching Google-for-search level; (2) Post-2023 reformed board + institutional CFO (Friar) + bench depth (Brockman, Lightcap, Weil, Fedus, Chen) provides governance-quality improvement vs pre-crisis era; (3) Revenue trajectory ($25B annualized growing +60-70% YoY) provides scale and diversification across API + Consumer + Enterprise + Strategic/Gov sleeves.

Top 3 reasons against: (1) GM stuck at ~33% end-2025 with no clear expansion trajectory yet — the load-bearing FF1 falsifier; hyperscaler-economics vs SaaS-economics question is open and trending unfavorable on current data; (2) Capital-structure waterfall places common shareholders FIFTH — Foundation warrant + Microsoft revenue-share through 2030 + employee/capex priorities ahead of common shareholders; dilution + revenue-share economic claims could absorb 15-25% of operational success silently; (3) Compound competition pressure — Anthropic Claude 4 at 30-50% lower pricing capturing F500 share + Gemini distribution-anchored displacement + open-weight self-host at ~10-20% of API cost; 45-55% compound probability of 2+ disrupting materially over 10-year window.

XI. Pre-IPO / IPO-specific considerations

Tier-C scaffold disclosure: This tree was built 2026-05-15 before any public S-1 lands (rumored H2 2026 per Reuters Oct 2025). Most financial-segment claims are Tier C trade-press triangulation. Capital structure mechanics (Foundation warrant, Microsoft revenue-share, Stargate JV ownership) are Tier C pending S-1. Refresh REQUIRED post-S-1 — verdicts may move materially, particularly L1A 1.5 (governance investability), L1B 1.1-1.2 (capital structure waterfall mechanics), L1A 1.2 (GM trajectory disclosure).

IPO mechanics tracking (per developments.md):

• GPT-5 release: Q2-Q3 2026 rumored
• Public S-1 / preliminary prospectus: H2 2026 (rumored window per Reuters)
• Roadshow begins: ~4-6 weeks post-public-S-1
• Pricing date: ~6-10 weeks post-public-S-1
• Trading debut: Q3-Q4 2026 most likely
• First post-IPO earnings: 90-120 days post-debut
• S&P 500 / index-inclusion eligibility review: T+90
• Lockup expiry (standard): T+180
• Q1 2027 GM print: FF1 falsifier resolution

Owner pre-purchase decision artifact: Per CLAUDE.md long-term-investability workflow + the asymmetric-favorable framework, the entry path is conditional secondary-market purchase post-S-1. Combined with this tree's verdict (asymmetry +1.2:1 favorable; conditional on S-1 disclosure quality), the recommended sequence is:

1. Watch GPT-5 release (Q2-Q3 2026); refresh capability moat assessment
2. Watch S-1 land (H2 2026); HIGHEST-LEVERAGE event — Foundation warrant + Microsoft revenue-share + Stargate JV ownership disclosure
3. Watch roadshow + pricing-day settle (~4-10 weeks post-S-1); refresh
4. Evaluate entry at IPO conditional on S-1 disclosure quality + roadshow price range
5. Watch first-day debut + first earnings (next 90-120 days)
6. Evaluate sizing at T+90 conditional on S&P 500 inclusion + first earnings GM signal

Foreign-equity / ADR note: OpenAI Group PBC is Delaware-incorporated; no ADR layer; direct NYSE or Nasdaq listing expected. No foreign-issuer disclosure substitutions apply.

XII. Investment Scorecard (per MANUAL_en.md Part K.6 + K.10)

15-question scorecard (Format B per K.3.5; long-term-hold-archetype)

Verdict tally (M1 evidence-tier suffixes per K.3.6): 7 ✅ · 6 ⚠️ · 1 ✗ · 1 mixed — moderate evidence support; structural quality is solid on operational + brand dimensions, weak on margin trajectory + capital-structure waterfall.

K.3.5 Weighted-score derivation

Applying the 4-tier weighting from MANUAL §K.3.5 (verdict values: ✅ = 1.0, ⚠️ = 0.5, ✗ = 0.0). OPENAI's Section XII rows carry meaningful Tier C uncertainty (10 R2 verifications outstanding) + ✗C on Q8 (FCF negative through FY2028+) + ✓ on multiple durability + governance + balance-sheet questions:

Sensitivity: an optimistic read (upgrading Q3 + Q5 + Q14 to ✅) yields 33.0/39 ≈ 85%. A conservative read (downgrading Q4 + Q6 to ⚠️) yields 26.0/39 ≈ 67%. The reported INDEX_META value of 67% applies a moderate scaffold-level haircut — the 10 R2 verifications still owed (S-1 audited financials, Foundation warrant precise mechanics, Microsoft revenue-share precise contract terms, Stargate JV ownership, GM trajectory, ChatGPT subscriber detail, PBC voting structure, lockup terms, compute lease obligations, Foundation governance mechanism) and the H-0 confidence at 58% and the bear-case asymmetry being moderate-not-severe justifies the conservative anchor in the moderate-buy band rather than the pure-buy band.

The 67% lands OPENAI at/near the moderate-buy band threshold (≥65% per K.3.5 interpretation) — explicitly NOT a high-conviction buy at the rumored $1T anchor but a conditional consideration at IPO contingent on S-1 disclosure quality. This is honest: OpenAI is structurally promising on brand + multi-product + governance-reformed-board but the scaffold contains capital-structure waterfall overhang + competitive pressure compounding + GM trajectory pending that high-conviction sizing should not ignore pre-S-1.

As the S-1 lands and Q1-Q4 2027 binaries resolve, the score should compress upward (toward 75-85% if 4 of 5 favorable resolutions) or stay flat (if mixed). Single-event sensitivity is high: favorable Foundation warrant + Microsoft revenue-share disclosure alone moves the score +6-10pp; favorable GM trajectory Q1 2027 print adds another +6-10pp.

Score interpretation: OPENAI at 67% sits below NVDA (86%), GOOGL (80%+), AAPL (85%), F (86%), AJNMY (91%), and slightly above NOK (71%). It sits well above SPACEX (52%) — reflecting the meaningfully more favorable asymmetry + smaller pre-marketing premium + less acute fatal-flag risk. It sits in the corpus's moderate-buy-with-conditional-sizing band where the structural quality is solid but execution + disclosure quality + competitive intensity introduce meaningful uncertainty.

Scorecard summary

Final verdict: CONDITIONAL — 0% pre-IPO; 0.5-1% at IPO if S-1 favorable on Foundation warrant + Microsoft revenue-share; 1.5-2.5% post-T+90 if 2 of 3 load-bearing mechanisms (GM trajectory + capability moat + capital structure) resolve favorably

For Ming specifically:

• ⚠️ Conditional entry at IPO based on S-1 disclosure quality binary. NOT a no-entry like SPACEX; NOT a high-conviction buy.
• ✅ Highest-leverage event: S-1 disclosure of Foundation warrant strike + Microsoft revenue-share rate + Stargate JV ownership. Watch SEC EDGAR closely H2 2026.
• ⚠️ Q1-Q4 2027 GM print is the second-highest-leverage event — direct read on FF1 falsifier.
• 🔻 Auto-AVOID if RF1 fires (S-1 reveals Foundation+MSFT unfavorable). Through T+12-18M.
• 🔻 Auto-TRIM if RF2 fires (Q1 2027 GM 32-34% failure signal). Toward 0%.
• 📅 Re-test on: GPT-5 release (~Q2-Q3 2026); Public S-1 (H2 2026); Roadshow (~4-6 weeks post-S-1); First earnings (~90-120 days post-debut); S&P 500 decision (~T+90); Q1 2027 GM print (Q1 2027).
• 🎯 Hard cap pre-IPO: 0%. Conditional cap at IPO: 1-1.5%. Cap post-T+90: 2-2.5%. Full-resolution cap T+12M: 3%.

The 2-minute pitch:

"OpenAI is going public H2 2026 at a rumored $1T pre-marketing — record-breaking for a US software IPO. Operationally it's three businesses: API (hyperscaler-tier compute pricing pressure), ChatGPT Consumer + Enterprise (brand-anchored subscription with first-mover scale), and Strategic/Government (optionality sleeve). The structural problem is gross margin: at ~33% end-2025, OpenAI's economics look hyperscaler-like (AWS/Azure 30-35%) not SaaS-like (75-85%). The bull thesis requires GM expansion to 45-55% by 2028 — a load-bearing assumption that is currently NOT in evidence. The second structural problem is the capital-structure waterfall: common shareholders sit FIFTH (compute → MSFT revenue-share through 2030 → Foundation valuation-milestone warrant → employees → common). The S-1 disclosure of Foundation warrant + Microsoft revenue-share precise terms is the highest-leverage single event. The third challenge is competition: Anthropic at $9B annualized growing 4-5× YoY + Gemini distribution-anchored + open-weight at 10-20% API cost. Compound probability 45-55% of 2+ disrupting materially over 10 years. But — the bull case is real. ChatGPT brand recognition is Google-search-level; post-2023 board is institutionally improved; Stargate compute is ahead of schedule; GPT-5 (Q2-Q3 2026) is the next capability test. Asymmetry +1.2:1 favorable at IPO (meaningfully better than SPACEX's 0.75:1 unfavorable). Conditional entry at IPO if S-1 discloses favorable Foundation warrant + Microsoft revenue-share + Stargate JV economics. Scale at T+90 if GM trajectory + S&P 500 + Anthropic plateau resolve favorably."

Risk types most relevant (per MANUAL_en.md Part K.4):

• Execution risk (Stargate compute build-out + GM trajectory inflection + GPT-5/6 capability execution; capability lead months-rolling not structural)
• Governance risk (Foundation control + PBC fiduciary-duty + Microsoft revenue-share waterfall + Altman-key-person + post-2023-board-crisis-recurrence; bounded by board reform)
• Competition risk (Anthropic enterprise compounding + Gemini distribution-anchored + open-weight commoditization; compound probability 45-55% of 2+ firing materially)
• Regulatory risk (FTC + EU AI Act + DOJ Microsoft-partnership review + copyright litigation; multiple active threads with bounded near-term impact)
• Capital-intensity / liquidity risk (Stargate $500B + $600B compute spend by 2030 vs revenue $25-130B FY2025-2030; equity-raise dependence through FY2030; lockup-expiry overhang T+180)
• Mechanical-flow risk (S&P 500 PBC + Foundation-controlled methodology novel; lockup absorption; first-day pop magnitude)
• Key-person risk (Altman departure event bounded; Friar + Brockman + Lightcap + Weil bench depth)
• Macro risk (AI infrastructure CapEx cycle + hyperscaler AI competition + Fed rate path; bounded but factor-sensitive)

"When NOT to buy" anti-pattern check (per MANUAL_en.md Part K.5):

• ❌ Buying because of single news headline (GPT-5 release IS a single-event headline; this tree explicitly recommends waiting for S-1 disclosure quality binary, not buying ON GPT-5 release)
• ❌ Buying based on momentum (Mega-IPO first-day pop is THE classic momentum trap; Snowflake/Coupang/Rivian precedents make this explicit)
• ✅ Aware of multi-binary structure (multiple binary events sequenced T+0 to T+180; S-1 → roadshow → pricing → first earnings → GM Q1 2027; willing to wait or pre-position with restraint)
• ⚠️ Online hype: OPENAI has near-MAXIMUM online hype — Twitter/Reddit/Fintwit pre-IPO hype at near-historical-extreme tier; AI-cycle leader status creates FOMO; partially mitigated by analytical framework requiring S-1 disclosure quality
• ✅ Has "AI" attached: OpenAI IS the AI cycle; the "AI-narrative" is substantive not surface; bull-case mechanisms map to genuine operational + technological levers
• ⚠️ Brand familiarity bias: ChatGPT has unusually strong brand recognition; cult-of-Altman is real; partially mitigated by post-2023 board reform analysis
• ⚠️ FOMO trigger: "Once-in-a-generation IPO" framing applies; multiple precedents (Snowflake, Coupang, Rivian) show mega-IPO FOMO entry has been the wrong call

Net: 3-4 anti-pattern flags but 2 mitigated by analytical framework. The trade passes the anti-pattern screen at CONDITIONAL ENTRY but fails it at debut-day-FOMO-buy. The conditional-entry-on-S-1-disclosure-quality path satisfies K.5 anti-pattern screens (waits past initial momentum + requires substantive disclosure quality confirmation + has explicit failure-signal exit rules).

XIII. What's NOT in this tree (deferred / pending S-1 refresh)

• Stage 0-2 R2 verifications (~10 items): Audited segment-level revenue + operating profit (API / Consumer / Strategic); Foundation valuation-milestone warrant precise mechanics; Microsoft revenue-share contract precise terms; Stargate JV ownership economics + OpenAI equity stake; ChatGPT subscriber count + ARPU disclosure; gross-margin trajectory disclosure quarterly; PBC fiduciary-duty + Foundation-vs-public-shareholder voting mechanism; lockup terms (Foundation, Microsoft, founder-specific); compute lease obligations off-balance-sheet; capital structure waterfall liquidation preferences
• Stage 2 supplementary: historical-analogue.md (Tesla 2010 + Amazon 1997 + Snowflake 2020 covered partially in peers.md)
• Stage 4 supplementary: premortem-steelman.md (covered partially in scenarios.md Bear case + L1B + L1D framework section)
• Sources catalog: sources.md — full bibliography of Tier B trade-press + future-Tier-A S-1 — deferred to next refresh after public S-1 retrieval
• Stage 6 living: update_{YYYY-MM-DD}.md files — first scheduled post-S-1 within 2 weeks; subsequent post-roadshow + post-pricing + post-debut + post-T+90 + post-Q1-2027-GM-print
• External research packets: Additional ChatGPT review packet + Codex review packet — built post-T+90 if owner wants conditional-entry calibration

Last updated 2026-05-15 (Tier-C build pre-public-S-1). Source quality: Tier B/C scaffold-level; full Tier A refresh required when S-1 lands. K.3.6 evidence-strength suffix convention applied to all leaf verdicts. Next refresh: GPT-5 release (Q2-Q3 2026); Public S-1 (H2 2026); Roadshow (~4-6 weeks post-S-1); First earnings (~90-120 days post-debut); S&P 500 decision (~T+90); Q1 2027 GM print (Q1 2027).

"Stories lie, structure doesn't." — 90s.PM.Investing

Personal disclosure: Conflict-of-interest applies in REVERSE for OpenAI analysis. Claude (this analyst) is made by Anthropic, the named #1 frontier-LLM competitor. The analysis avoids analytical OVER-skepticism on OpenAI defenses + OVER-charity to Anthropic enterprise position. Symmetric scrutiny applied — where analysis lands favorable to Anthropic on enterprise share-gain (L1D 1.1), the claim has been re-tested against Reuters + The Information + ChatGPT-IC-memo + enterprise survey data and externally corroborated; where analysis lands favorable to OpenAI on operational durability + brand + governance reform, the bull-case mechanism + falsification conditions are explicitly stated. Symmetric scrutiny applied — if Anthropic were positioned with comparable governance and capital-structure mechanics, the analysis would apply equivalent skepticism.