There’s no doubt that AI is reshaping entire industries and, as a result, the way young entrepreneurs build companies. But as AI evolves, what does that mean for the kinds of businesses investors are actually funding? To find out, I analysed more than 2,000 startups that joined the cohorts of the world’s most famous accelerator, Y Combinator, over the past four years. The analysis reveals which industries make up the largest share of funded startups, how the picture differs across countries and over time, and how startups’ positioning around AI has changed since 2023.
Industrials are growing, but B2B still dominates in 2026
One clear shift in the startups that YC funds is the rise of Industrials. Under Industrials, YC includes sectors such as agriculture, automotive, aviation and space, climate, defense, drones, energy, manufacturing, and robotics.
Its share increased from just 2.43% in 2023 to 18.23% in 2026, making the sector roughly 7.5 times as large by share as it was in 2023. In absolute numbers, this change is significant, too: the number of Industrials startups in the YC directory grew from 12 in 2023 to 132 in 2026, more than a tenfold increase.
B2B remains by far the largest sector in YC cohorts, but its share has declined. It accounted for 68.02% of the 2023 cohorts, falling to 62.88% in 2024, 64.96% in 2025, and 56.63% in 2026. However, the absolute number of B2B startups increased from 336 in 2023 to 410 in 2026. The broader picture is that the YC startup landscape is becoming more diversified. B2B still dominates, but Industrials is taking an increasingly large share of each new cohort, while AI, agents, and robotics are becoming central themes across the ecosystem.
AI positioning is changing
As AI capabilities evolve, so does the way startups describe their products and position AI within their businesses. I used natural language processing to identify patterns in startup names and one-line descriptions on the YC directory.
One notable shift I found is that founders are moving away from describing their product as “AI-powered” toward “AI-native.” This shift makes sense. “AI-powered” generally means that businesses use AI to enhance their existing product, while “AI-native” suggests that AI is fundamental to development or operation.
At the same time, the fact that founders define their businesses less frequently as “platform” suggests a broader change in positioning. Rather than identifying primarily as infrastructure or a platform, more startups are emphasizing what exactly the system itself can do.
The share of descriptions containing “autonomous” increased from just 0.24% in the 2023 cohort to 4.01% in 2026, which is more than a sixteen-fold increase.
Other terms have followed a similar path. The change is notable for “robots,” which appears more than thirteen times as often as in 2023. “Agent” appears nearly five times as often, while “agentic” has gone from a marginal term to one appearing in 2.63% of captured descriptions.
One possible interpretation is that startup positioning is moving from tools that help users perform tasks toward systems that perform more of those tasks themselves. The rise of “agent,” “agentic,” and “autonomous” language is consistent with this shift.
Another interesting change is happening in how founders name their startups. The share of YC companies with an explicit “AI” marker in their name in 2026 is about 42% lower than in 2023, despite AI becoming more prominent in startup descriptions.
The term “AI” appears in 21.03% of 2023 descriptions and in 43.02% of 2026 descriptions. In other words, AI has more than doubled as a description-level signal while becoming less common as a naming signal.
There are several possible explanations. One is that AI has become so widespread that founders no longer need to put “AI” directly into the company name to signal what they do. AI may now be treated as an underlying technology rather than a differentiator.
Startup name structure itself has remained stable. Median name length stays around seven to eight characters, while average word count ranges from approximately 1.21 to 1.33 words across the period.
Industry mix and AI positioning by country
As YC is an American accelerator, it is no surprise that the overwhelming majority of funded startups are based in the United States. Across the full dataset, 82.24% of records are associated with the US, with San Francisco alone accounting for 1,456 startups.
American startups account for 81.03% of B2B records and 86.89% of Industrials records. This means that the growth of Industrials is not being driven primarily by a particular international market; it is largely a US-based phenomenon within the YC ecosystem.
However, B2B remains the largest category everywhere, ranging from 60% of Canadian startups to 77.5% of German startups. Industrials has a smaller but noticeable presence in France and Canada, where it accounts for 12.5% and 13.33% of records, respectively.
India stands out in a different way: 23.08% of its startups are classified as Fintech, compared with 7.11% in the US and 11.49% in the UK. This suggests that although the YC ecosystem is heavily concentrated in the US, the industry mix can differ considerably in smaller national cohorts.
AI terminology is also not distributed evenly across countries. The UK has the highest share of descriptions mentioning “AI”, while Canada has the lowest. The vocabulary also differs beyond the simple presence of “AI.” In the US, “agents” appears in 11.65% of descriptions, while the figure reaches 12.94% in the UK. France has 10%, Germany 7.69%, Canada 6.67%, and India 8%.
The UK also has a relatively high share of descriptions containing the phrase “AI agents”: 9.41%, compared with 6.11% in the US.
Restricting the comparison to B2B startups provides a more focused view of how companies are positioning AI in one of YC’s largest categories. In the US, 187 of 1,179 B2B descriptions mention “agents,” or 15.86%. The comparable figure is 8 of 55 B2B descriptions in the UK, or 14.55%.
Overall, the country-level data suggests that AI and agent terminology is widespread across YC’s major markets, but not uniformly so. The US remains the center of the ecosystem by scale, while some smaller markets show a higher concentration of AI-related language despite having far fewer startups.
What these numbers mean for aspiring founders
The data does not provide a formula for getting into YC, but it does show how the startup landscape and the language founders use to describe their companies are changing:
- AI is becoming more of a baseline. Founders may therefore get more value from explaining what the AI does than from simply highlighting that they use AI.
- AI-native positioning is becoming more common. This suggests that startups are increasingly presenting AI as part of the product’s core architecture rather than as an added capability.
- The physical world is becoming a larger part of the YC landscape. This suggests an expanding overlap between AI and robotics, manufacturing, and other physical-world applications.
- Geography still matters, but the trends are not exclusively American. The US accounts for 82.24% of the dataset and 86.89% of Industrials records, but smaller markets show their own patterns — for example, India has a 23.08% Fintech share, while the UK has a particularly high prevalence of AI-related language.
Taken together, the data suggests that AI is becoming more widespread while the language around it is becoming more specific: from AI-powered products toward AI-native systems, agents, autonomy, and increasingly physical applications.
About the Data
For this research, I analyzed the YC cohorts from winter 2023 to fall 2026, totaling 2,393 startups. It’s worth noting that the data is disproportionate, and the dataset includes 494 startups in 2023, 590 startups in 2024, 585 startups in 2025, and 724 startups in 2026, as many as were accepted into YC each year. This analysis therefore compares cohort groups as they appear in one directory snapshot and does not track the same companies’ language over time. I collected available information from the YC directory, categorized it, and ran the analysis with Python.
