How we calculate this

We identify 3–5 publicly traded comparable companies for each unicorn. Peer choice matters enormously: anchoring a 200%-growth private AI lab to 15%-growth mega-caps under-prices it badly. We use sector-appropriate, growth-matched peer baskets — high-growth (NVDA, PLTR, CRWD, MDB, NET) for AI labs; data/infra (SNOW, DDOG, MDB, NET) for Databricks; payments (ADYEN, V, MA, MELI, PYPL) for Stripe; defense-tech (PLTR, KTOS, RTX, NOC) for Anduril; and so on.

For each peer we use the EV/Revenue multiple, weighted by that peer's revenue growth rate so the fastest-growing peers carry more influence. We then apply a per-company illiquidity discount (10–25%) to reflect lack of liquidity, audited financials, and preference-stack risk. Category leaders with deep secondary markets and a clear IPO path get a smaller discount (~10–13%); mid-tier names get 18–20%; smaller / less-liquid names get 22%+.

We multiply the adjusted multiple by the company's estimated annual revenue. For pre-revenue or opaque companies we substitute EV/Employee as a proxy.

Weight: 25%. Confidence: 30–75%, depending on peer fit and revenue-estimate freshness.

Secondary markets (Hiive, Forge, Caplight) publish bid/ask indications for private company shares. These prices already incorporate a discount to primary — they reflect what investors have recently indicated willingness to pay in private secondary transactions.

We collect the 3–5 most recent indications per company and apply exponential recency weighting: a transaction from 30 days ago has roughly 5× the weight of one from 180 days ago (e⁻ᵈ/¹⁸⁰ decay function). This prevents stale data from dominating the estimate.

Weight: 45%. Confidence: high when recent trades exist (up to 80%); drops quickly if data is older than 6 months.

The simplest anchor: take the last disclosed primary-round valuation and adjust forward using the company's estimated revenue growth and a small multiple-decay tax. Three corrections apply:

1. Per-company revenue growth rate — set explicitly from reported or estimated ARR trajectory (e.g. ~120% YoY for OpenAI, ~50% for SpaceX, ~27% for Stripe). Compounded for elapsed time since the primary. No more headcount proxy: for AI/SaaS companies, revenue decouples from headcount thanks to compute leverage.
2. Multiple decay — a 5%/year haircut on the multiple, floored at 0.85×, to reflect that markets re-rate downward by default unless the company grows into its valuation.
3. Secondary cap — if a secondary trade exists in the last 180 days, Method C's output is capped at 1.3× that secondary-implied valuation. Stops runaway extrapolation when modeled revenue growth outpaces what the market is actually paying.

Weight: 20%. Most reliable for recently funded companies (within 12 months). Confidence decays 20 pp/year from the primary date, flooring at 20%.

Method D uses the performance of a sector ETF as a market-proxy forward adjustment to the last primary round. It answers the question: "if this sector kept pace with its public benchmarks since the last funding round, what would that imply?"

Each company is mapped to the most relevant sector ETF:

We compound the ETF's annualised return from the last primary date to today to derive an adjusted valuation. Confidence is fixed at 50% — Method D is a tiebreaker, not a primary signal.

Weight: 10%. Used as a tiebreaker and for companies with no secondary market data.

We compute a confidence-weighted average of the four implied valuations using weights A=25%, B=45%, C=20%, D=10%.

Outlier auto-downweight. Any method whose output strays more than 50% from the median of the other three has its confidence weight halved before the blend. This is a robust-statistics safety net — it prevents a single bad extrapolation (most often Method C overshooting for high-growth companies) from dragging the headline figure.

Spread guardrail. Even after downweighting, if the ratio between the highest and lowest method output exceeds 3×, the result is flagged as "wide spread". The displayed range widens (±1.8 σ instead of ±1.2 σ), overall confidence is capped at 35%, and the UI surfaces an explicit "methods disagree" notice. This is the single most important credibility safeguard on the site — a 13× spread shown without warning would tank user trust.

The spread (low–high range) is set to ±1.2 σ of the four method outputs (±1.8 σ when wide), with a minimum floor of ±15%.

Overall confidence = weighted average of the four method confidence levels. Above 60% = high; 35–60% = medium; below 35% = low.

The three scenarios apply fixed shocks to the mid-point valuation and model dilution:

The calculator correctly handles five equity types:

PrivatePulse data is based on public sources — company press releases, SEC filings, reputable financial press (Bloomberg, Reuters, TechCrunch, WSJ, FT), and verified secondary market publications. It is not a live data feed.

Secondary indications are used in Method B only when publicly verifiable (Forge/Hiive public summaries, SEC tender filings, or named press reports). Unverified bids and internal board marks are never used.

Reported signals (unconfirmed fundraising talks, IPO targets, or strategic deals) are shown in a separate "Reported signals" section and never overwrite the confirmed primary valuation.

Data is reviewed periodically — high-priority companies every 30 days, others on a 90–180 day cycle. Data is not continuously updated and does not reflect intraday market moves. A structured refresh workflow was introduced in June 2026 to make this process systematic.

• We don't model liquidation preferences in detail — this requires cap table access.
• We don't predict IPO timing or probability.
• We don't account for clawbacks, transfer restrictions, or company-specific plan rules.
• We don't provide state tax estimates — consult a local tax advisor.
• We don't incorporate bid-ask spreads as separate secondary inputs yet.
• PPU holders: we model economic value only, not the profit-sharing mechanics of OpenAI's PPA. Consult your plan document.