Stripe trades 117% above Ant Group despite Ant clearing 18× the payment volume.
Three new pairs stress-test the founding thesis — enterprise AI, open-weight models, and payments. One of them produces the cleanest regulatory-overhang pricing in the whole dossier.
A wider cohort, a sharper boundary
Stripe trades 117% above Ant Group on a price-to-sales basis, despite Ant clearing about eighteen times Stripe's payment volume and posting nearly twice the revenue. The discount is the cleanest piece of regulatory-overhang pricing we have found anywhere in this cohort — Ant's halted November 2020 IPO and the surrounding crackdown still set the multiple six years later. We add it to the scoreboard alongside two other pairs that test the edges of the Western-premium thesis: Cohere against 01.AI in enterprise AI, where Lee Kai-Fu's pivot from consumer models to industrial customers makes revenue comparison structurally asymmetric; and Mistral against DeepSeek in open-weight frontier models, where the Chinese side is unambiguously the price-maker. The Mistral pair returns the cohort's second-largest inversion (roughly −82% to −89%). The three additions bring the documented cohort to twelve pairs across foundation models, semiconductors, robotics, and payments.
What the original cohort didn't cover
The founding cohort had a defensible perimeter — frontier foundation models, humanoid robotics public and private, accelerator silicon, logic foundry. The signal across those nine pairs was clean enough to publish, but the perimeter omitted three categories of pair the thesis ought to be checked against.
First, enterprise-tier AI.Cohere has built a defensible position in enterprise retrieval and embedding workloads from Toronto and San Francisco, with USD 250 million of annual run-rate revenue by end-2025; 01.AI under Lee Kai-Fu pivoted hard from consumer-facing models to industrial applications across 2024 and 2025, which makes its revenue comparison structurally asymmetric to a Western enterprise vendor. The premium is not computable on the standard basis — but the pair is instructive on capability ranking and on what Chinese enterprise revenue actually looks like at the line-item level.
Second, open-weight models. The frontier foundation-model pairs in the founding cohort assumed both sides had application-programming-interface revenue. The open-weight tier inverts that assumption: Mistral monetizes lightly through API and enterprise distribution, DeepSeek monetizes almost not at all by design, but its open-weight model files are the global free-tier baseline. The valuation gap on this pair is the closest the cohort gets to pricing the technology rather than the business.
Third, payments.Payments is the only frontier-tech sector where the regulatory overhang is unambiguous, dated, and load-bearing. Ant Group's November 2020 IPO halt, the subsequent restructuring under People's Bank of China pressure, and the discount that has held in the six years since make it a natural anchor for a regulatory-overhang reading. If the premium framework cannot see this pair clearly, the framework needs revisiting.
The cohort's cleanest regulatory-overhang reading
Ant Group cleared roughly USD 17.2 trillion in payment volume across Alipay's 1.3 billion monthly active users in the 2023 audited period. Stripe processed roughly USD 1.4 trillion in payment volume across 47 countries in the same period. Ant's 2023 audited revenue was USD 26.2 billion (CNY 184 billion). Stripe's 2024 company-reported revenue was USD 14.1 billion. Stripe trades at USD 91.5 billion (February 2025 tender). Ant trades at roughly USD 78 billion (Alibaba 6-K mark, March 2024).
On a price-to-sales basis: Stripe at about 6.5 times, Ant at about 3.0 times. The Western premium is +117%— Stripe trades at more than twice Ant's price-to-sales multiple despite Ant clearing eighteen times the payment volume and posting nearly twice the revenue. The ratio is not subtle. It is the largest single pricing distortion in the cohort that does not flow from a listing-venue mechanism such as a Hong Kong or STAR Market debut, and does not flow from a capability gap.
Ant's 2020 IPO halt is the load-bearing repricing event. The discount has not closed in the six years since — it has institutionalized.
The mechanism is well documented in public filings and contemporaneous reporting. The November 2020 IPO halt — at the time set to be the largest in history at roughly USD 34 billion — was followed by a forced restructuring of Ant's lending and consumer-finance arms under People's Bank of China pressure, a CNY 7.1 billion fine in July 2023, and a sustained markdown by Alibaba in its 6-K disclosures. The discount visible in the +117% premium is exactly the regulatory overhang priced as a permanent valuation haircut rather than a transient compliance cost.
The cleanest cross-jurisdictional open-weight comparison
Mistral closed its Series B in June 2024 at USD 6.4 billion and is talked at USD 11 billion in Q3 2025 funding discussions. Revenue scaled from roughly USD 30 million in 2024 to USD 100 million in annual run-rate by late 2025, per Sacra. DeepSeek is talked at USD 45 billion in the Big-Fund-led round of May 2026 against estimated API revenue of USD 50 to 100 million.
On absolute valuation, DeepSeek prints at 4.1 times Mistral. On capability, DeepSeek-V3 and R1 post MMLU 87 to 90 against Mistral Large 2 and Magistral at 84. On a price-to-sales basis, taking the midpoint of the DeepSeek revenue range, the signed premium runs −82% to −89%: the Chinese open-weight firm carries a substantially higher price-to-sales multiple than its Western open-weight counterpart, in the direction opposite the default Western-premium reading.
The mechanism here is different from the Hong Kong listing-window dynamic that drives the Anthropic and Zhipu inversion. DeepSeek has no public listing; the talked valuation is private. The premium is being set by capital that views DeepSeek's open-weight release cadence and reasoning-eval performance as a global free-tier baseline, while Mistral is being priced as a regional incumbent. The valuation has run ahead of the revenue on both sides, but it has run further on the Chinese side.
Revenue-asymmetric, but instructive
Cohere is the cleaner of the two on standard frontier-AI metrics: roughly USD 5.5 billion valuation at its July 2024 Series D, USD 250 million in annual run-rate revenue by end-2025, Command R+ in production at enterprise scale, and a defensible niche in retrieval and embedding workloads. 01.AI sits north of USD 1 billion from its March 2024 round, with the 2025 pivot to industrial applications obscuring any clean run-rate figure. The official Yi-Lightning model carries an LMArena T6 ranking and an MMLU of 76 — a tier below the frontier but credible for industrial deployment.
The premium is not computable on the standard basis because the revenue base on the Chinese side is the wrong shape: industrial-application revenue does not aggregate into an annual run-rate comparable to enterprise software. The instructive read is qualitative: 01.AI's post-2024 pivot is itself a data point about what is and isn't monetizable in the Chinese enterprise context. The Cohere and 01.AI cell sits in the dossier as an example of why the framework does not, and should not, force a numeric premium on every pair.
The cohort grew, the median moved slightly
With the three additions, the cohort is now twelve documented pairs across seven sector buckets: frontier foundation model (three), multimodal model (one), humanoid robotics public and private (three), AI accelerator silicon (one), logic foundry (one), enterprise frontier model (one), open-weight frontier model (one), and payments infrastructure (one). Eight pairs resolve to a numeric signed premium; the remaining four are reported as not computable on a like-for-like basis because one side is pre-revenue, structurally non-comparable, or qualitatively unbounded.
The aggregate median sits at −19.5% across the eight numeric pairs, with an interquartile range of 169.25 percentage pointsrunning from a first quartile of −75.75% (the Hong Kong listing-window foundation-model inversion zone) to a third quartile of +93.5% (the private-stage Western premium zone). The Stripe and Ant +117% reading is the cleanest single regulatory-overhang outlier and sits alone in the payments cell; the next-largest Western premium remains Boston Dynamics and Unitree at +233% in humanoid robotics.
Technical detail
- The documented cohort now covers twelve pairs across foundation models, semiconductors, robotics, enterprise AI, open-weight models, and payments.
- Cohere and 01.AI— enterprise frontier model. Premium not computable on a like-for-like basis. 01.AI's 2025 pivot to industrial applications is the load-bearing reason.
- Mistral and DeepSeek— open-weight frontier. Premium −82% to −89% (DeepSeek trades at a price-to-sales premium to Mistral); DeepSeek talked at 4.1 times Mistral on absolute valuation.
- Stripe and Ant Group— payments at scale. Premium +117%. Ant's November 2020 IPO halt is the load-bearing repricing event.
- Primary-source valuation evidence pinned for sixteen firms across IPO prospectuses, 10-K, 10-Q, 6-K filings, S-1s, and disclosed funding rounds.
- Trajectory data captured for twenty-four of thirty-six pair-period cells — the foundation of the temporal series in the next piece.
What this opens. The three additions did what cohort expansion is supposed to do: they moved the headline median slightly but reframed the boundary cases entirely. The Stripe and Ant pair is now the cohort's canonical regulatory-overhang reference; the Mistral and DeepSeek pair is its canonical open-weight inversion reference. The next piece disaggregates the cohort into sector subgroups — the point at which the −19.5% headline median resolves into four very different regimes it was always hiding.