Cosmic against the
peer syndicate

Cosmic’s overlap book is, first and foremost, a late-entering crossover book layered on top of a small, high-quality angel spine. Two-thirds of Cosmic’s total invested capital sits in these 62 shared names, and where entry valuations are comparable Cosmic enters at a median 13.4x the first mover’s post-money. The consequence shows up in the MOIC spread: Cosmic’s median pooled MOIC when first-in is 2.1x against 1.8x when later — a +0.3x differential, in line with Greenoaks and DST and well below Spark (+7.7x), 8VC (+8.2x) and Khosla (+4.6x). Notably, a16z’s differential on fully-blended marks is also just +0.3x: once its growth-fund capital is marked rather than excluded, the platform managers that both seed and repeatedly re-up in the same companies — a16z and Cosmic alike — converge to the same pattern, in which pooled company economics are set by the largest, latest cheques rather than the first one.

Mechanically this is the averaging-down effect of stacked follow-on capital rather than a selection failure, and the Deel case is the cleanest illustration on both sides: Cosmic’s 204.6x Yeda seed pools to 5.1x once the $32m Bet 2 cheque at $12.1bn is included, and a16z’s 49.4x Fund VI entry pools to 8.9x once its $162m of growth capital is marked. Cosmic frequently was early — Deel, Middesk at 30.5x, Perplexity at 82.5x, Harvey at 65.1x, Braintrust at 53.8x, Cognition at 40.1x on the Aleph side — but the pooled per-company MOIC folds in far larger Bet cheques written at five to a hundred times the original entry, pulling company-level economics toward the growth basis. Read properly, the data supports the Aleph/Bet bifurcation thesis rather than refuting it: Aleph behaves like a genuine early-mover selection engine, while Bet behaves like a valuation-taking crossover allocator into the same consensus fairway as Kleiner Perkins Select, Greenoaks, DST and the a16z Growth funds, with correspondingly compressed multiples. The undiluted first-mover premium survives most clearly at the managers whose early cheques are not followed by large later capital in the same vehicle family — 8VC, Spark and Khosla.

Selection quality within the shared fairway is competitive but not first among peers. On identical-round entries Cosmic sits alongside the leads — Anrok, Ventrilo, Convex, Metronome, the Anduril Series B at the same $920m post as a16z, Cognition ahead of both Khosla and 8VC’s follow-ons, Harvey nine months ahead of Kleiner Perkins at one-eighth the price. The distinctive Cosmic result is repeated first- or same-day access to the current AI cohort (Harvey, Perplexity, Cognition, Braintrust, Decagon, Character, Mistral, Pika, Recraft, Socket), which is precisely the access argument underpinning Cosmic 5. For portfolio construction, however, the 34-name a16z edge means an investor holding both managers is doubling exposure to a common AI and infrastructure factor rather than diversifying; Cosmic’s 121 non-overlapping companies, roughly a third of invested capital and mostly small Aleph positions, are the genuinely diversifying part of the platform.

Data sources

Cosmic position data is taken from the manager-supplied Q3 2026 workbook of platform positions (241 rows across Yeda, Aleph 1–4, Bet 1–4 and the Bet 3 and Bet 4 Durable vehicles), cross-checked against the Q1 2026 schedules of investments. Peer manager data is taken from the internal Portfolio positions database, covering 8VC, a16z, DST Global, Greenoaks, Kleiner Perkins, Khosla Ventures and Spark Capital across approximately 2,900 positions and 2,528 unique companies. a16z Growth fund MOICs are supplemented from a separately supplied a16z late-stage venture position dataset covering AH Growth Funds I–IV, marked as of Q2 2025, whose IRRs reconcile with the positions database. All figures are gross and denominated in US dollars as recorded by each manager, and are not all of a common mark date: a16z Growth positions carry Q2 2025 marks against Cosmic’s Q3 2026.

Overlap matching

Company names were normalised by extracting doing-business-as and formerly-known-as forms, stripping parentheticals and iteratively removing corporate suffixes, then matched server-side against the database’s own normalised company master using batched pattern queries. A second pass caught cases where the database stores a shorter form than the manager file (Anduril for Anduril Industries; Recraft for Recraft AI). Every ambiguous hit was then verified manually against legal name, sector, round date and round size.

Confirmed true matches with dissimilar names include Anchor Labs as Anchorage, Degree as Lattice, People Center as Rippling, Alpha Exploration as Clubhouse, Mellis as Pika, FirstLayerAI as Medallion, Plus Five Five as Resend, and Sprig Technologies as Sprig. Rejected false positives — excluded from all figures — are AnchorZero and AnchorUSD against Anchor Labs; Avenue Technologies against Aven; IfOnly and IFTTT against IF Labs; Framework Computer against Frame; LFG Future against LFG Labs; Magic Leap against Magic AI; Rain Financial against Rain Neuromorphics; and Charm Therapeutics against Charm Industrial.

Stage normalisation

Round stages are taken as recorded by each manager and mapped to a common axis: pre-seed and seed to Seed; lettered rounds to their letter through Series D; Series E and beyond, together with rounds labelled growth or late, to Series E+/Growth; listed-security purchases to Public; token purchases and unclassifiable instruments to Other. Greenoaks’ secondary purchases are retained as a distinct Secondaries row rather than folded into the corresponding primary stage, since the entry economics differ materially.

Valuation conventions

Entry valuations are entry equity values as recorded in the source. The convention records the post-money at the manager’s first investment in a company and carries it forward across follow-ons, so a follow-on row’s recorded valuation is not necessarily the valuation of that round. Cosmic’s entry valuation field is populated on 63 of 241 positions; database rows frequently carry zero or blank valuations, which are treated as missing rather than as zero. Only the earliest dated position with a populated valuation is used for each manager-company pair. Two values are known artefacts and are flagged in place rather than corrected: the Lattice entry in KV VI and the Cosmic dbt Labs entry.

Date conventions

The a16z Growth-fund source records dates in day-month-year order, and both that source and the corresponding positions-database rows had been ingested month-first, transposing day and month wherever the day was twelve or below. Affected Growth-fund dates were corrected by swapping day and month, validated against externally known rounds (Character AI to March 2023, the round a16z led; Mistral to December 2023, the round a16z co-led). The a16z early-fund rows do not share the defect — the Mistral entry in AH Fund VIII, correctly recorded in December 2023, would be impossible if transposed — and were left untouched, as were all other managers. The corrections change displayed entry months on eight positions and no entry rankings.

First-mover ranking

Rank is assigned on each manager’s own earliest recorded commitment date within a given company. Same-day entries share a rank. Positions without a recorded date — predominantly Cosmic’s Bet 3 and Bet 4 rows, plus one Khosla holdco residual — cannot be ranked and are excluded from first-mover statistics while remaining in the capital and value tables. Because rank derives from recorded commitment dates rather than round closes, two managers in the same round days apart can appear in either order.

Value and return aggregation

Total value is computed as invested capital multiplied by reported MOIC. The positions database reports IRR without MOIC for the a16z Growth funds; those 38 positions (roughly $4.7bn of invested capital in this universe) were marked using the separately supplied a16z late-stage venture dataset, matched on normalised company name and fund. The supplied MOICs reconcile with the IRRs already recorded in the database on every matched row, confirming both sources reflect the same mark set, so MOIC coverage is complete for all eight managers and the value and MOIC tables carry no exclusions. Any position still reporting no MOIC — a handful of Cosmic Bet 4 rows recorded without cost or value — is excluded rather than imputed at cost. Manager-level IRRs shown are capital-weighted averages of position IRRs, which approximate but do not equal a true pooled cash-flow IRR; positions recording an IRR of exactly zero are excluded from the weighting.

Pooled company MOIC

For the first-mover analysis, each manager’s MOIC in a company is the sum of value over the sum of cost across all of that manager’s positions in that company, which means a strong initial cheque can be diluted by large later follow-ons. This is the economically correct measure of what the manager earned on the name, but it is not a measure of initial-cheque selection quality, and the two should not be conflated when reading the Cosmic differential.