晚期公司研究：前沿AI首次公开定价 SpaceX定价；Anthropic和OpenAI在下一轮

LATE -STAGE CO M PAN Y RE SE ARCHThe First Public Pricefor Frontier AI Institutional Research Group Harrison RolfesSenior Research Analyst,Late-Stage Company Researchharrison.rolfes@pitchbook.com pbinstitutionalresearch@pitchbook.com SpaceX priced it; Anthropic and OpenAI are next Published on June 26, 2026 PitchBook is a Morningstar company providing the most comprehensive, mostaccurate, and hard-to-find data for professionals doing business in the private markets. Contents Key takeaways1AIBQ: Where the three sit on business quality2Reading SpaceX: The cross-subsidy testand the sum of the parts3What the frontier AI labs are implicitly worth:The per-quality-point bridge6The revenue-recognition swing and thepublic tape7The market and the four-month window9What would change the call11References12 Key takeaways •Business quality runs inverse to valuation.Anthropic scores highest on the AIBQframework (8.20) at the lowest value per quality point for frontier AI while OpenAI(4.53) and the xAI segment (4.49) score lowest and cost the most. The IPOs decidewhether the market pays for fundamentals or narrative, a verdict that sets themultiple for every subsequent listing. •SpaceX is not a clean comp.A profitable Starlink absorbs the AI losses, a cushionthe labs do not have, so the lab listings test appetite for pure burn far more directly.Expect both to lean on scarcity and brand to make up for the missing cash engine. •The live comp implies a wide range.Pricing each lab off the per-quality-pointladder spans roughly $1 trillion to $2.8 trillion for Anthropic and $0.5 trillion to $1.6trillion for OpenAI, weighting to about $1.4 trillion and $0.95 trillion. The roadshowranges are the first read on which regime is clearing. •Revenue recognition is the swing factor.Anthropic’s roughly 20.5x run-ratemultiple could be 50x to 60x on recognized revenue, near the richest level a publicAI name has sustained. The public S-1 is where a gross-to-net restatement wouldsurface, resetting every multiple at once. •The constraint is appetite, not capacity.Roughly $135 billion to $165 billion of newsupply sits against about $8 trillion of money market cash, so the deals can befunded. The catch is that the natural seller is the same megacap AI book buyersalready own. Watch for rotation pressure on incumbents as the deals price. •The call is falsifiable at dated levels.The SpaceX volume-weighted average price,whether OpenAI prices first on brand, a gross-to-net restatement, and roadshowranges versus the private marks are the tests. Anthropic likely prices first and setsthe comp OpenAI is measured against. AIBQ: Where the three sit on business quality Before reading SpaceX into the frontier AI companies, it is important to anchor onfundamentals rather than headlines.We score each name on the AIBQ framework,a 0-10 composite across capital efficiency, revenue quality, compute independence,governance optionality, and moat durability. Together, they measure how muchdurable business sits beneath the valuation, which is exactly what a headline marketcap obscures. SpaceX does not receive a clean AIBQ score. AIBQ measures AI business quality, andSpaceX is a launch and connectivity company that houses an AI segment (xAI, X)acquired in an all-stock deal in February 2026. The AI-relevant entity inside SpaceXis xAI, and that is what the table reflects. That Developing score is corroborated bySpaceX’s own actions: Its $60 billion Cursor acquisition reads,in our SpaceX coverage,as a tacit concession that xAI’s frontier-model effort lags. The fact that the AI is buriedinside a profitable parent is the central observation of this note, because it meansSpaceX flatters the category: Its print shows what a frontier AI lab looks like with acash engine underneath it, which is precisely what Anthropic and OpenAI lack. We read the pattern two ways: •Constructive read:Quality is on sale at Anthropic. If public markets eventuallyreward fundamentals, a strong-tier business at the lowest multiple in the group isthe most defensible name in the frontier AI cohort. •Cautionary read:The market is pricing narrative, scarcity, and optionality ratherthan fundamentals. If that is what clears, the AIBQ score measures something thebuyer is not paying for, and the cheap name stays cheap. The listings settle which read is right, and the first clean print is the signal. Afundamentals-led Anthropic pricing would validate the framework and pull capitaltoward quality across private AI. A scarcity-led print across the cohort would signalto allocators that access and narrative are being prioritized over business quality andwould anchor the multiple grid that OpenAI and every later AI IPO inherit. Reading SpaceX: The cross-subsidy test and thesum of the parts The common shorthand since the June 12, 2026, debut is that SpaceX proves publicmarkets will fund frontier moonshots. That is half right. What SpaceX demonstrates isthat public markets will pay an extreme