Atlas of the 151 Grand Finale startups

Materials and methods

How each figure in this atlas was obtained, checked and scored. Data collected and verified in November 2026.

1. Population studied

The 151 startups qualified for the Startup World Cup 2026 Grand Finale in San Francisco, based on the three official lists displayed on site: the Top 10 and the 141 other finalists. They come from 74 countries.

2. Sources

Public sources only: official startup websites, press releases, business and trade press, public pages of funding databases (Crunchbase, Tracxn, PitchBook, Dealroom), public founder profiles, pages of the Startup World Cup regional finals and of Pegasus Tech Ventures. In total, 611 sources are cited, and each profile gives its own with a link.

The research was carried out by AI agents, each working on a batch of startups. Each result had to cite its source, then was reviewed, consolidated and corrected by hand before publication.

3. Identification

Each name was searched on its own, then with “Startup World Cup 2026” and the name of the regional final, to avoid namesakes. The confidence level of the identification is shown on each profile: high for 93 startups (official website and qualification confirmed), medium for 58 (a single source or partial information). 3D Armour remains the most uncertain identification.

4. Cross-check

A second, independent pass reviewed 150 profiles: country, sector, qualification, funding and product. 112 profiles were corrected or completed. The verification note and its sources appear on each profile.

5. Founders and funding rounds

• Coverage: founders documented for 136 startups, funding round details for 128.
• What is counted: equity (shares, convertibles) and, separately, public grants.
• What is excluded: debt, prizes paid in kind or in credits, and fundraising targets announced without a confirmed close.
• Conversion to dollars: fixed reference rates: €1 = $1.08, £1 = $1.27, CHF 1 = $1.12, AUD 1 = $0.65, ¥100 = $0.67, ₩1,000 = $0.72, CNY 1 = $0.138, HKD 1 = $0.128, NOK 1 = $0.093.
• Disagreements between sources: the best-documented amount is used, and a low and high range is shown.
• Status of each startup: amount disclosed (76), no funding (55), amount not disclosed (18), target only (2). The $278.2M total counts only disclosed amounts.

Scale of the regionals: the number of applications and finalists comes from releases by the regional organizers and from the local press, with no estimate when nothing is published. The total estimate multiplies the median of comparable regionals by the announced number of regionals, then cross-checks it against the organizer's figure.

6. Economic data by country

GDP, population and GDP per capita for 2025 come from the World Bank (indicators NY.GDP.MKTP.CD, SP.POP.TOTL, NY.GDP.PCAP.CD). For Taiwan, which is absent from the World Bank, the figures come from the IMF. The ratios set the total raised by a country's startups against its GDP and population. These are orders of magnitude on a small sample: a single startup can shift a country's result.

7. Maturity

• Funding stage: based on the last known round. Self-funded if no funding round and declared revenue; Undetermined if no reliable information was found.
• Product stage: R&D, prototype, pilot, or product on the market (sold to paying customers).
• Year founded: year of legal incorporation when a registry or the official website gives it; otherwise the launch year stated by the company or the databases, flagged as such. When sources disagree, the profile says so and cites its source. As a last resort, the first archive of the website (archive.org) or the year of the © serves as an estimate, marked “estimate” on the profile. No year is guessed: “not found” otherwise. Age is calculated as of 2026.
• Traction: none known, pilots or letters of intent, first customers, or strong traction (significant documented revenue or customers).

8. Artificial intelligence and LLMs

Three levels of AI use:

• AI at the core: the product does not exist without AI.
• AI integrated: AI is one feature among others.
• No documented AI: no public trace of AI. This does not prove its absence.

The LLM provider is shown only when public evidence exists (documentation, press release, job posting, partnership). This is the case for 34 startups; for the others, the field is left empty rather than guessed.

9. Investment-bank-style scoring

Four criteria scored from 1 to 10, each justified in one sentence on the profile:

• Disruption:
  • 1–3: incremental improvement;
  • 4–6: real innovation in a known category;
  • 7–8: new approach that can change practices in the sector;
  • 9–10: platform or technology breakthrough.
• Business model strength:
  • 1–3: no proven revenue or customers;
  • 4–6: pilots or first paying customers;
  • 7–8: recurring revenue or solid contracts;
  • 9–10: proven profitability at scale.
• Market potential: see the TAM rule below.
• Risk (10 = very risky): technology, regulation (health, finance, energy), execution and team, cash reserves relative to needs, competition and data uncertainty.

Overall score = (disruption + model + market + (11 − risk)) ÷ 4. Categories: A ≥ 7 · B 6 to 6.9 · C 5 to 5.9 · D < 5.

10. Global TAM and scalability

The TAM is the size of the global market in 2025, in billions of dollars of revenue, for the segment actually addressed. It is taken from a market research report cited on each profile, with the annual growth rate and the 2030 estimate.

Scalability (1 to 10) measures how easily the model can be replicated in other countries without proportional costs: software or hardware, country-by-country regulation, need for local partners.

Market score:

• base according to TAM: 3 below $1bn, 4 from 1 to 5, 5 from 5 to 20, 6 from 20 to 100, 7 above $100bn;
• +1 if scalability is at least 8;
• −1 if it is 4 or less;
• +1 if the startup sells in at least two countries or has a committed expansion.

Fairness calibrations, flagged on the profiles concerned:

• When a startup covers several segments, they are added together.
• When the published figure measures a volume rather than revenue (invoices financed, sales, loan book), it is converted into revenue: Btrust, Kauche, Monet.
• Very broad segments are flagged as upper bounds: Nuhpro.

11. Investors, Pegasus and mentors

• Investors: names were normalized (variants, subsidiaries, misspellings) before being cross-referenced. 18 investors appear in at least two startups.
• Pegasus Tech Ventures and mentors: the links come from Pegasus's public pages and those of the 397 Startup World Cup mentors and judges. A link is called direct if it is documented between the person and the startup, and indirect if it goes through a fund, a program or an employer. No link is kept without a source.

12. Data reported by the companies

For a few startups, non-public information (signed customers, team size, ongoing expansion) was provided directly by the company. It cannot be verified publicly and is presented as reported on the profile concerned. All startups are scored on the same grid.

13. Limitations

• Public data is incomplete, especially for young startups and for countries with little press coverage; many amounts are self-reported.
• Market research reports use different definitions and methods: their TAMs are orders of magnitude, not exact values.
• The scores are analyst judgments based on an explicit grid, not an investment recommendation.
• All data is as of November 2026.