Back out the percentages and the more useful numbers appear. A 716.89% one-year gain on a $1.38 trillion estimate means Anthropic was being carried at roughly $169 billion a year ago [1][1], and OpenAI's 140.08% implies about $375 billion [2][2]; between them the pair went from $544 billion to $2.28 trillion, which is $1.74 trillion of paper value created [3]. That gain is five times the $344.7 billion that Stanford's 2026 AI Index counted as every dollar of global private AI investment in 2025 [13][4]. Marks compound faster than cash arrives, because a mark is an extrapolation from the last thin slice sold.
Forge's account of the mechanism is the honest one: sovereign wealth funds, private equity firms, corporate investors and crossover funds are now competing with traditional venture capital for access, and that competition lifts valuations earlier [10]. A buyer bidding for allocation rather than for ownership is not solving a discounted cash flow, and the price it accepts on a small primary tranche becomes the reference for the entire cap table. Forge's own framing is that AI-native businesses reach meaningful revenue and strategic relevance faster than earlier generations [9], which is plausible and also exactly what a mark-setter would say.
Gartner puts spending on AI models and platforms at $64.25 billion this year, up 63.4% from 2025 [14]; Anthropic's mark alone is 21.5 times that entire line [5], and that line is 6.4% of the roughly $1 trillion Goldman Sachs Research expects to be invested in AI during 2026, $581 billion of it in the United States [11][6]. Compounding at 63.4% closes a 21.5x gap in about six years if nothing decelerates [10]. That is a real answer, but not a near one.
The evidence permits three resolutions. Model spending keeps compounding and the marks look conservative when the S-1s land. Or the first listing prices inside the private range, the crossover funds that set the reference take the difference on their own books, and the primary window narrows for the rest of the 182 names DeFiLlama tracks [3]. Or nobody lists, and the estimates stay unfalsified. Nasdaq Private Market's own July note argued the market is differentiating winners with durable advantages from valuations driven by enthusiasm [17], while reporting its private tracker up another 10% from around June 1 [16], which is a claim about discrimination supported by a number that moved in one direction.
This desk reads the $1.38 trillion as a price for access to a slice rather than a value for the whole, and that is probably wrong in one specific way worth naming: if adoption really does pull revenue forward the way Forge describes, the 4.5 years Anthropic took to reach $100 billion, against 19.6 for SpaceX [8], is evidence about the business rather than about the bidding. Falsification is clean either way. An AI listing that prices at or above its last private mark and holds it through lockup expiry, or Gartner's model-and-platform line doubling twice, would settle it. Meanwhile the capital going to frontier labs and the infrastructure beneath them [18] is capital not underwriting the rest of that board, which now raises against a reference two companies set.
