The company says one thing. The record says another.
SOMEN Research reads private, pre-IPO technology companies the way an engineer reads a design review: claim by claim, against the primary record.
The format is the discrepancy hunt. A company claims a valuation, a backlog, a customer count. The filings, the patent record, the labor certifications, the archived versions of its own website often say something else. Each teardown starts at that gap and works out what it means while it still matters: before the S-1.
The universe is the pre-IPO market: AI labs, silicon, cloud, memory, defense, and the supply chain underneath all of it. Written for people who allocate to, supply, or compete with these companies.
The Method
Research runs on one knowledge graph built for this desk: 100+ companies and ~12,000 sourced relationships (growing daily) spanning funding, patents, hiring, supply contracts, dependencies, and secondary-market pricing. Every relationship carries its source, its evidence date, and a confidence score. A number without a source does not print.
Three rules govern every piece:
Every number resolves to a link a reader can open.
Every core claim ships with a falsifier and a resolve-by date.
Every resolution lands on the public scorecard: confirmed, refuted, or partial, in plain view.
The Cross-Company Maps
The teardowns read one company at a time. The graph also looks at the big picture:
whole supply chains, mapped end to end.
Every dependency is a sourced edge. Which clouds rent which chips, which labs
rent from which clouds, which fabs fabricate which silicon, which tools those
fabs cannot run without, and which one or two companies make those tools. The
Two Stacks map traced both AI compute chains, Western and Chinese, from the
financing through the power to the lasers, and found the same shape in each:
the chain narrows as it deepens, and the narrowest points have a single
supplier. Memory After the Sold-Out Years ran the same process on three
memory makers and an export wall. The maps carry the capital layer too: who
finances the buildout, and which guarantees circle back to whose revenue.
Because every edge carries its evidence date, the maps reconstruct as of any
past date, and every count can be re-derived rather than remembered. A map
here is not a graphic; it is a record.
The Starting Record
Nscale: $103 Billion Claimed, $21 Million Filed. The backlog shown to IPO investors, against the audited revenue actually filed in the Norwegian register.
Etched: The 2.9 Percent Round. Ownership arithmetic on a $10.3B private mark. Thirteen hours after publication, the company announced a $700M Series D at $21B.
Memory After the Sold-Out Years. Three memory makers, three post-2027 paths, and the export wall a stockpile cannot cross.
Two Stacks: The AI Compute Supply Chain Map. Both Western and Chinese stacks in one graph. It narrows as it deepens: 25 direct NVIDIA dependents, 13 on TSMC, then one lithography vendor and two sole-source German suppliers.
Who Writes This
A pseudonymous desk, signed The SOMEN Desk. Two decades engineering systems alongside the people building these companies. The byline is a desk rather than a person for a simple reason: the work stands on its sources, not on a resume, and the companies covered here are private, closely held, and sensitive about being read closely.
What You Get
One teardown most weeks, every number linked to a filing, an archive, or a dataset. Between pieces, short Notes with the napkin math: a round’s new-money stake, a claim’s two-source fork, a filing that contradicts a press release. The arithmetic cannot lobby you.
Our Ask
Everything is free while the desk builds its public track record. Subscribe and the next teardown lands in your inbox. A founding tier opens later; founding readers go first, at the lowest price it will ever be.
The scorecard is public and stays public. Nothing here is a prediction, a price target, or investment advice. It is a structural read of the public record: what the companies say, what the record shows, and what would prove the desk wrong.
Every number traces to a source. The analysis is ours.
- The SOMEN Desk
