ARR doesn't describe the active private market anymore

Last week's two biggest private rounds didn't go to AI companies.

Anduril raised $5 billion at $61 billion post-money. Isomorphic Labs raised $2.1 billion at an undisclosed post-money valuation — lbxpro's secondary market mark on the name is $17.1B.

Both oversubscribed. Both in the top 25 most valuable private companies. And neither can be honestly underwritten on the framework venture capital was built to use.

At $61 billion post-money against $2.2 billion in 2025 revenue, Anduril is priced at roughly 28x trailing revenue. For reference, Lockheed Martin trades at roughly 1.6x revenue and Palantir trades at multiples that are themselves contested as a software-comp anomaly. But the 28x number doesn't actually describe Anduril. Anduril's revenue is not a recurring base that compounds — it is the realized portion of contracts the government may or may not continue to fund, against ceiling vehicles like the $20 billion Army IDIQ that authorize spending without obligating any of it.

Isomorphic is an even harder case. The company's revenue line in PitchBook is blank. Its commercial agreements with Eli Lilly ($45M upfront, up to $1.7B in milestones) and Novartis ($37.5M upfront, up to $1.2B in milestones) are option contracts, not subscriptions. Realized revenue is roughly 4% of contingent value. The other 96% is a real option whose strike is FDA approval. You cannot put a multiple on zero.

This is not unique to last week. It is the shape of the modern private market.

Six out of twenty-five

Look at the top 25 active companies on the secondary market. Six are recognizably SaaS-shaped — Stripe, Ramp, Databricks, Mercury, ClickHouse, Glean. ARR multiples work. The other nineteen are not that.

Anthropic, OpenAI, Perplexity, ElevenLabs, Ayar Labs sit in AI, where ARR exists but is so far below burn that the multiple is uninformative. Anduril, Shield AI, Saronic are defense, where revenue is the output of a political-bureaucratic process rather than a customer adoption curve. Ripple and Kraken are crypto, valued on regulatory regime and exchange-volume cycles. Polymarket and Kalshi are prediction markets, valued on CFTC posture and network liquidity. Lambda and Crusoe are AI infrastructure — capex-and-PPA economics, closer to a utility than a software company. SpaceX is space, where dual-use revenue is starting to allow partial demand-driven valuation but the launch business is still sovereign-driven. Neuralink is FDA-gated. PsiQuantum is pre-revenue, milestone-driven, government-grant-supported. Figure is industrial robotics, valued on physical deployment.

Roughly 75% of the most active private companies in the world today sit in categories where the venture industry's standard valuation framework does not produce a defensible number.

And here is what we are actually seeing on our side.

Since we launched Leo, our AI analyst on lbx.pro, we have had exactly one user query about Anduril fundamentals. One. We do not think this is because secondaries buyers don't care. We think it's because the work is hard, the methodology isn't standardized, and the path of least resistance is to anchor on the primary mark and call it a day.

Why the standard toolkit fails

ARR/revenue underwriting assumes four things: revenue recurs by contract structure, customer churn is the main downside risk, growth extrapolates from cohort behavior, and multiple expansion follows when net revenue retention compounds.

Each of those fails differently across the non-SaaS categories.

In defense, the $20 billion Army IDIQ that Anduril sits on is a ceiling, not a backlog — a contract vehicle that allows the government to place multiple task or delivery orders against a maximum without obligating any particular order. Task orders convert at the government's discretion, against appropriations that move with administrations. There is no churn rate because there is no contractual continuity to churn from. The customer is a sovereign with policy goals, not a buyer with usage growth.

In AI drug discovery, Isomorphic's commercial structure is even further from SaaS. The Lilly and Novartis deals are explicit option contracts: small upfront payments giving partners the right but not the obligation to advance specific molecules, with milestone payments contingent on clinical and regulatory events. There is no ARR to grow. There is a probability-weighted scenario tree.

So what does fit?

We are not going to claim we have the full answer. Neither does anyone else. The honest position is that the private market is solving a methodology problem in public.

Based on cases like Anduril or Isomorphic Labs we reverse-engineer the checklist — but it will not cover all the non-revenue-based valuation cases.

Any checklist will require more than understanding several numbers (ARR, rule of 40, and the rest). It requires hunting for specific signals across DoD award filings, FDA databases, allied procurement announcements, clinical trial registries, satellite imagery of factories, regulatory rulings, sovereign capital flows. And this is what we do — we collect news, awards, filings, leaks, policy moves, and sovereign capital flows. The good news: with current AI capabilities, this is a tractable workflow. Three years ago it was not.

Leo is here to share the signals we are gathering. Start by asking Leo.