AI Startup Rankings: Methodology
The below methodology is determined by Claude based on the following prompt: “Rank the top 100 AI startups right now, as judged by you. Search the web for current data first. Choose your own criteria for judging them.”
What the rankings measure
Each ranking scores private AI startups on how strong a business they are today and how well placed they are to keep growing. It is a judgment-based ranking built from public reporting, not a valuation model or investment advice. Scores are comparable across rankings because every list uses the same criteria, weights and scale.
Who qualifies
A company is included only if, as of the data date, it is:
- Private. Not publicly listed, and not in an IPO that has already priced.
- Independent. Not acquired, not under a signed acquisition agreement, and not majority-owned by a large technology company.
- AI-native. AI is the core product, not a feature added to an older business.
Notable companies that fail one of these tests are listed on each sheet's Excluded tab with the reason and a source. Examples: xAI (merged into SpaceX, now public), Cursor (acquired by SpaceX), Hugging Face (Nvidia acquisition agreed), Waymo and Isomorphic Labs (controlled by Alphabet), and Ramp in the fintech list (corporate cards and spend software with AI features, rather than an AI-first product).
For a category ranking, the category is defined at the top of the sheet's source note. The fintech list covers financial research, accounting, fraud and identity, compliance, lending and mortgages, insurance, and tax.
How the data was gathered
- Candidate list. Built from funding trackers and rankings (Value Add VC, data.prompt20.com, aifunding.me, Forbes AI 50, Tech Funding News) and recent coverage of the largest funding rounds.
- Company research. For each candidate: latest valuation, latest funding round (amount, date, lead investor), a traction figure (revenue, recurring revenue, customers, users or contracts), product, headquarters, founding year, website and founders.
- Status checks. Each company was checked for acquisitions, mergers, IPO filings and blocked deals before scoring.
- Sources. Up to three source links per company are in columns U to W of the Ranking tab. Sources are company announcements, news coverage and financial data sites.
Valuation rules. The Valuation ($B) column holds the latest confirmed or widely reported valuation. Funding talks that have not closed, secondary share sales and IPO targets are described in the Valuation detail column instead, so they do not inflate the number. A blank valuation means none has been disclosed.
Website rule. Websites are listed only when a source confirms them. Where none could be confirmed, the sheet says "No public site found" rather than guessing a domain.
Scoring
Each company is scored from 1 to 10 on five criteria:
| Criterion | Weight | What it measures | 10 means | 5 means | 1 means |
|---|---|---|---|---|---|
| Traction | 30% | Revenue, recurring revenue, paying customers, users and deployments today | Tens of billions in run-rate revenue | Roughly $100M in recurring revenue, or broad paid use | No product |
| Growth | 20% | Revenue growth and valuation increases over the last 12 months | Several-fold revenue growth | Healthy, ordinary growth | Stalled or shrinking |
| Technology | 20% | Frontier capability, proprietary models, hardware or data, research talent | Sets the state of the art | Strong product built mostly on others' models | Commodity |
| Market position | 15% | Size of the market and how defensible the company's place in it is | Leader in a market worth hundreds of billions | Contender in a sizable niche | Small or crowded market |
| Capital and durability | 15% | Money raised, investor quality, runway, and risks such as customer concentration or regulation | Effectively unlimited capital and low key risks | Funded but exposed | Fragile |
Traction carries the most weight so that proven revenue counts for more than funding or hype. As a result, well-funded labs without a product score lower than smaller companies with real customers.
Composite score. Score = 10 x (0.30 x Traction + 0.20 Growth + 0.20 Technology + 0.15 Market position + 0.15 Capital and durability), giving a score out of 100, rounded to one decimal place.
Ranking and ties. Companies are ranked by composite score. Ties are broken by higher valuation. Companies with the same score and the same (or no) valuation share a rank.
Category rankings. In a single-category list, market position is judged within that category's market, while the rest of the scale stays the same as the overall list.
Limitations
- Private companies disclose little. Many figures come from press reports, some describe funding talks rather than closed rounds, and a few data points are a year or more old (each is dated in the sheet).
- The criterion scores are editorial judgments made from the evidence listed, not outputs of a statistical model. Reasonable analysts could score the same company a point or two differently.
- The data is a snapshot as of the date above and will go out of date as companies raise, get acquired or go public.
- Small differences in score (under about one point) should be read as ties.
Disclosure
The rankings were researched and scored by Claude, an AI model made by Anthropic. Anthropic appears in the AI Startup 100 and ranks first by half a point over OpenAI. That gap is within the margin treated as a tie, and readers may wish to weigh that placement with this relationship in mind.