Two AI Markets, One Headline Number: What Early-Stage Funding Patterns Actually Reveal

The Bifurcation: One Market, Two Economies

There is no longer a single AI funding market. There are two, and they operate on almost entirely different logic. At the top sits a hyper-concentrated, winner-take-all game dominated by a handful of frontier-model companies. Below it sits a far more competitive scramble for early-stage capital, where investors have become measurably sharper about what they will and will not underwrite.

The distinction matters because the two economies are routinely reported as one. "It's extraordinary how much money is being raised and how much money is being spent on just a few companies," said Kyle Stanford, Director of VC Research at PitchBook — a line that doubles as a warning about how to read any AI funding chart.

Everything that follows rests on that premise: the headline totals describe a handful of companies, not the experience of the typical founder. A seed-stage team reading that AI absorbed hundreds of billions of dollars is reading a number that has essentially nothing to do with the check it is trying to raise.

The Headline Numbers — and Why They Disagree

Start with the top line, and the first thing you find is that the sources do not agree. One puts global venture investment in Q1 2026 at $300 billion across 6,000 startups, up more than 150% both quarter over quarter and year over year. Another puts the figure at $255.5 billion deployed in the same quarter.

The framing of that quarter diverges too. One source says Q1 2026 absorbed close to 70% of all venture capital deployed in the whole of 2025. Another says the quarter surpassed the full-year 2025 total outright.

Even individual rounds are reported inconsistently. Anthropic's raise appears as $30 billion in Q1 2026 in one account, and as $65 billion at a $965 billion post-money valuation in May 2026 in another.

The disagreement is not a footnote — it is the signal. Differences in methodology, coverage windows and disclosure gaps make top-line venture totals genuinely unstable. Anyone quoting a single number for "how much went into AI" is quoting one dataset's answer, not a fact.

The Top of the Market: Concentration at Historic Extremes

What is not in dispute is the shape of the distribution. In Q1 2026, the top five AI firms captured roughly 73% of total value raised in AI. OpenAI alone closed a $122 billion round that quarter, described as one of the four largest venture rounds ever recorded.

The picture widened rather than narrowed as the year went on. In the first half of 2026, 78% of global VC funding — $399 billion — went to AI companies, with $237 billion of that going to OpenAI, Anthropic and xAI alone. Three companies, in other words, absorbed well over half of the AI total.

The trend was already established in 2025, when AI captured 65.6% of all US VC deal value — $222 billion of a $339 billion total — and AI-native startups took 53% of VC dollars globally.

• Top five AI firms, Q1 2026: ~73% of total AI value raised
• OpenAI, Q1 2026: $122 billion in a single round
• H1 2026: 78% of global VC ($399B) to AI; $237B to three companies
• Horizontal AI platforms, Q1 2026: $197 billion across 396 transactions

That last figure is worth dwelling on. Horizontal AI platforms — foundation models and infrastructure — took $197 billion across 396 transactions in Q1 2026, beating the segment's own full-year 2025 total in a single quarter. The biggest checks are going to the layer that requires the most capital to build.

Why the Average Round Size Lies to You

Concentration this extreme breaks the statistics founders most often use as benchmarks. In generative AI, the YTD 2026 average round was about $9.6 billion. The median was $65 million. The mean is not describing a market; it is mostly measuring OpenAI, Anthropic and xAI.

The concentration math is stark. In YTD 2026 generative AI, the largest single deal captured 50.8% of disclosed capital. The top three captured 90.4%. The top five captured 99.0% — meaning every other deal in the category shared the remaining one percent.

Category-level data tells the same story. Foundation Model APIs accounted for 24.0% of generative AI deals but 98.9% of disclosed capital. By dollars, generative AI is primarily a frontier-model recapitalization story rather than a broad application-layer boom.

And the growth is entirely in dollars, not activity. January through July 2026 saw 25 disclosed generative AI deals across 24 companies for roughly $240.0 billion — about 5.1x the comparable 2025 capital on essentially flat deal count, 25 versus 26.

The practical takeaway for founders and analysts is simple: benchmark against medians, never means. A $65 million median is a number you can reason about. A $9.6 billion average is a number about three companies.

The Early-Stage Reality: More Options, Fewer Dollars

Beneath the mega-rounds, the early stage is busy — just not rich. Generative AI first financings made up 48.0% of YTD 2026 deals but only 0.8% of capital. Investors are still creating new options at a healthy clip while reserving virtually all their dollars for perceived winners.

The stage mix captures the split precisely. Series A is the most common generative AI stage by deal count at 52.0%, while Series D+ captures 99.1% of capital. The market is simultaneously early in product formation and late-stage in infrastructure financing — two different clocks running in the same sector.

There is one genuinely encouraging pattern for founders: AI density rises with every stage. Across the broader market, 44.3% of pre-seed deals, 46.4% of seed, 53.5% of Series A and 59.2% of Series B deals are AI companies. AI-labelled startups are not just numerous at the entry point; they are increasingly the default as companies progress.

The Graduation Squeeze

Being numerous at seed is not the same as surviving to Series B. Only about 30–35% of companies that close a seed round reach a Series A. Series A to Series B graduation has fallen harder still, to 15–20% within 24 months for post-2021 cohorts, down from a historical 28–30%.

Time between rounds has stretched alongside the failure rate: the median gap has moved from 20 months to 28 months. That is a straightforward operating implication — every stage now requires meaningfully longer runway to reach the metrics the next round demands.

AI startups do fare better than their peers on both counts. They convert between stages at higher rates than non-AI companies, and they command valuation premiums: roughly 25% higher at Series A for AI SaaS ($54M versus $43M) and a 41% premium at seed. The premium is real. It does not exempt anyone from the squeeze.

Where the Smart Early-Stage Money Is Going

After an earlier wave of generalist model bets, capital at the applied layer is consolidating into vertical AI — products where the software replaces or compresses a specific, expensive workflow.

The numbers describe a deliberate narrowing. Vertical AI application deal counts have fallen 51% from their Q1 2022 peak, while average deal size has more than doubled year over year to $24 million. Fewer companies, bigger checks, and a higher bar for what counts as a defensible moat.

Some of the largest applied bets are physical. Autonomous machines posted a record $29 billion across 118 deals, driven largely by Waymo's $16 billion Series D — a reminder that "applied AI" increasingly includes hardware-heavy categories with capital needs closer to infrastructure than software.

The mega-rounds at the top also reshape what happens below. They influence LP allocation toward AI-focused funds and accelerate deployment speed, with a predictable downstream effect: more capital chasing fewer breakout seed and Series A deals, at higher valuations, for companies that fit a tight thesis.

Agentic AI: The Counterexample of Broad-Based Formation

If generative AI shows what extreme concentration looks like, agentic AI shows what a healthier formation curve looks like. Funding rose from about $1.55B in 2024 to about $2.59B in 2025, and YTD 2026 had already reached about $2.51B by early July — roughly 3.0x the comparable 2025 period and about 97% of the full-year 2025 total.

Crucially, the growth is in deal formation, not just dollars. Deal count rose from 13 in the comparable 2025 period to 65 in YTD 2026. That is the opposite of generative AI's flat count and exploding capital.

Concentration exists here too, but at a manageable scale: the top 10 YTD 2026 rounds took 51.3% of capital, and rounds above $50M accounted for about $1.73B of the $2.51B total. The median round was $18M against a $38.5M average — a gap you can actually reason across, unlike generative AI's median-to-mean chasm.

• Vertical AI agents are the largest category at 50.8% of deals and 54.8% of capital, spanning customer support, procurement, accounting, security, recruiting, legal and compliance.
• Agent Execution Infrastructure is the fastest mover, from about $21M in the comparable 2025 period to about $504M in YTD 2026 — a shift from agent demos toward production deployment, security and governance.
• First financings were 43.1% of YTD 2026 agentic deals, up from 24.4% in full-year 2025, though still only 12.4% of capital.

The pattern is consistent with a category still forming its winners rather than recapitalizing them.

The Geography Gap

Frontier-model financing is an almost entirely North American phenomenon. North America accounted for 88.0% of YTD 2026 generative AI deals and 99.8% of disclosed capital, with Europe and Asia-Pacific appearing only through narrower examples.

Agentic AI looks distinctly more transatlantic. North America took 68.4% of YTD 2026 agentic capital while Europe rose to 28.8% — up sharply from 7.7% in full-year 2025.

The contrast points to a broader dynamic: emerging categories, where the capital requirements are measured in tens of millions rather than tens of billions, offer far more geographic openness than the frontier-model tier does.

What Investors Now Ask in Diligence

The sharpening at the early stage shows up most clearly in what investors ask. Diligence has moved away from product roadmap and toward harder structural questions: where the data moat sits, what inference costs look like at scale, and whether the business survives platform risk.

The central question is blunt — does the product still exist if a foundation model ships the same feature natively? For an application-layer company, that single question now determines whether a thesis is fundable.

Concentration at the top is what raises that bar. When a handful of model companies command the capital to build almost anything, "defensible" has to mean something more specific than a good interface on someone else's model.

The Longer Arc: Volume Without a Corresponding Boom

Step back a decade and the AI funding story looks different again. AI deal volume grew nearly 4x, from 240 AI-related rounds in 2010 to 7,646 in 2025, with 2,887 rounds recorded in Q1 2026.

But the path is not a straight climb. Round counts peaked earlier, at 6,842 in 2021, then dipped through 2022–2024 before recovering.

The most instructive detail: despite ChatGPT triggering an enormous wave of AI investment, the number of AI funding rounds stayed near 6,000 per year afterwards. That suggests the overall fundraising climate — rates, exits, LP appetite — may matter more to deal formation than the technology theme alone.

Dollars, not deal counts, are what exploded. The boom is real, and it is almost entirely a story about check size.

Methodology and Caveats

The figures in this article come from several analyses with materially different scopes, which is precisely why they disagree.

• StartupHub.ai draws on 500+ AI funding rounds across 2025 and early 2026, using PitchBook, Crunchbase, Carta and proprietary data.
• Aventis Advisors reviewed roughly 65,000 Crunchbase-tagged AI funding rounds from 2010 to Q1 2026.
• New Market Pitch analyses cover disclosed equity rounds of at least $300,000 for pure-play companies.

Differing scopes, inclusion criteria and disclosure thresholds explain much of the numerical disagreement across sources — including the conflicting Q1 2026 totals and the divergent Anthropic raise figures cited earlier. Treat any single top-line venture number as one dataset's answer, and read medians before means.