If you remember one number from YC's 2026 batches, make it this: 61% of the companies are B2B. Out of 478 companies in the current snapshot, 292 sell to businesses. Consumer? 20 companies — about 4%.
A decade ago, YC was shorthand for consumer internet. In 2026 it reads more like a map of enterprise workflows. Here's what the data says, and why I think it marks a structural shift rather than a hot sector.
61% is not a rebound — it's the new baseline
Look at the five-year trend for B2B share of each YC year:
| Year | B2B share | Consumer share |
|---|---|---|
| 2021 | 47% | 11% |
| 2022 | 47% | 9% |
| 2023 | 68% | 6% |
| 2024 | 63% | 10% |
| 2025 | 66% | 7% |
| 2026 | 61% | 4% |
B2B jumped from 47% to 68% in 2023 and has held above 60% for four straight years. That's not a cycle — it's a regime change. And consumer's curve runs the opposite direction, ending at its lowest point in the series.
The economics are blunt: consumers will try a new app but won't necessarily pay; enterprises move slowly, but once your product is embedded in a workflow, the money, data, permissions, and renewals all live inside it. Post-generative-AI, the default startup idea is no longer "a better interface for users" — it's "eat a specific job, a specific process, a specific system integration."
B2B no longer means "selling SaaS"
The 2026 sub-industry breakdown shows how wide this wave runs:
- B2B (general): 97 companies
- Infrastructure: 62
- Engineering, Product & Design: 39
- Productivity: 17, Operations: 15
- Security: 10, Marketing: 9, Finance & Accounting: 8, Legal: 8, Supply Chain & Logistics: 8
This isn't a fresh crop of CRMs. It's a map of the inside of a company — code, procurement, compliance, finance, support, logistics, data observability — with AI-native tools growing at every node. A few one-liners from the batch make the pattern concrete:
- Canary — "the first AI QA engineer that understands your code"
- Rubric AI — reasoning and verification infrastructure for AI
- Carrot Labs — track and attribute AI spend across every provider
- Pollinate — AI agents for the supply chain
None of these are selling a smarter chat box. They're selling capacity a specific department can put to work.
Software is turning from tool into employee
The keyword data is even more telling. Across the 477 real companies (one obvious test record excluded), 226 — 47.4% — match terms like agent, copilot, operator, or assistant in how they describe themselves. Nearly half the year is packaging its product as a job, not a tool.
Old SaaS was a tool: you log in, click buttons, fill forms, export reports. The new wave behaves like a role: it makes the calls, drafts the documents, reviews the contracts, runs QA, processes approvals, flags anomalies. What the buyer purchases isn't seats — it's the automation of a category of repetitive labor.
This is also the deeper reason B2B share climbed. The first place AI creates provable value isn't open-ended creativity; it's closed loops: clear inputs, verifiable outputs, quantifiable labor costs, bounded error. Enterprises are full of exactly those loops.
Two people can now take on a vertical workflow
One more number matters: among companies with known team size, the median team is 2 people (average 3.1). There are 205 two-person teams and 34 solo founders. The B2B surge did not bring back sales-heavy, implementation-heavy org charts.
The opposite happened. AI lets tiny teams wedge into an industry crevice: find one workflow that's frequent, painful, and clearly budgeted, then go deep with models, data connections, and orchestration. Two people used to struggle covering product, sales, support, and delivery at once; with those internal capacities amplified, small teams now iterate faster than big ones.
Which explains why horizontal platforms feel less exciting than vertical wedges this year. Insurance inventory audits, PE deal sourcing, CAD for mechanical engineers — each looks narrow, but each has budget, pain, and headcount it can replace.
A takeover doesn't mean everyone wins
Crowded B2B also means brutal filtering. Enterprise buyers don't pay because you used AI. They ask: can you plug into our existing systems, actually reduce headcount-hours, pass compliance, and deliver reliably? When half the batch says "agent," the real differentiation isn't in the demo video — it's in data access, workflow detail, and cost of deployment.
So the honest reading of YC 2026's B2B takeover isn't "enterprise software is a safe lane." It's that the default battleground moved: from capturing attention to capturing workflows, from building apps to building roles, from selling interfaces to selling outcomes.
In 2021 the founder question was "will users open it every day?" In 2026 it's: "will a company hand you a piece of its work?"
Data notes
- Source: public data from ExploreYC and the YC Startup Directory, current snapshot.
- "YC 2026" covers the Winter, Spring, Summer, and Fall 2026 batches; Summer and Fall data are likely still incomplete (Fall has only 4 records so far).
- Raw 2026 count is 478; one obvious test record is excluded where noted, giving 477.
- Keyword screens are coarse heuristic matches and overlap is allowed — treat percentages as directional.
- This is research and analysis, not investment advice.
Slice the data yourself
Every cut in this post — by industry, by batch, by keyword — came from queries you can run yourself with the ExploreYC Startup Research Agent. It works against the same public dataset, so you can test your own hypotheses instead of taking mine. Background on the integration is on the ecosystem page, and the agent runs on ClawMama.
Top comments (0)