Research · April 2026

Series A Got Pickier About Product Categories

We classified 1,220 venture-backed startups into ten categories and tracked a decade of Series A rounds. The biggest shift was not AI rising. It was the generalist software bucket collapsing from 24.9% of Series A rounds to 9.6%.

The common reading of the 2022 to 2024 funding market is that AI ate everything. Our data does not support that, at least not in round count. AI/ML grew from 24.0% to 31.2% of Series A rounds between 2015-18 and 2022-24. That is a 7.2 point gain and the biggest category shift we measured, but it is not a reshaping of the market. Cybersecurity, Fintech, and Healthtech each grew alongside it.

The real rotation came from two places that get written about least. The generalist “Other” bucket fell from 13.5% of Series A rounds to 9.6%. Data Infra and DevTools/Infra, long-favored categories of the 2015-18 era, together gave up 12.8 points of share. Capital did not concentrate into AI. It concentrated away from categories that used to be the default software bet and away from companies that could not credibly claim a sector identity. This is the same dataset we used to map how long a Series A takes now; here we slice it by category instead of by time.

01The Dataset

The dataset is 1,220 US-based venture-backed companies. Every company had at least one funding event on record, and we pulled the full funding history for each: every seed round, Series A, Series B, and beyond, with dates and dollar amounts where disclosed. That produced 3,403 funding events. For this post, we focused on the 566 Series A rounds that cleared our classifier.

1,220
US-based, venture-backed
566
Series A rounds, 2015 onward
10
Categories, plus one Other
973
Classified companies (79.8%)

02How We Grouped the Companies

Industry labels in the source data were not useful. 86% of companies in the raw export tagged themselves as “information technology & services,” which is too broad to tell a story. We built a keyword-based classifier instead: each company has a list of self-reported tags like “machine learning platform” or “endpoint security,” and we mapped those tags against ten category dictionaries.

The dictionaries are deliberately narrow. A company only lands in “AI/ML” if its tags include phrases like “artificial intelligence,” “machine learning,” “deep learning,” or “generative ai.” Tagging a product as “predictive analytics” or “conversational ai” counts. Tagging it as “enterprise software” does not. Companies that did not match any category with at least two keyword hits landed in “Other.” 247 companies (20.2%) ended up in “Other,” which is itself a finding we return to in section 04.

03Share of Series A Rounds by Category

The first thing we wanted to see is whether the narrative of category rotation holds up. It mostly does not. Here are the ten categories ranked by 2022-24 share of Series A rounds.

Share of Series A rounds by category and window
2015-182019-212022-24
AI/ML
24.0%
28.9%
31.2%
+7.2pp
Cybersecurity
11.5%
14.4%
16.1%
+4.6pp
DevTools/Infra
17.7%
10.3%
12.4%
-5.3pp
Fintech
4.2%
9.8%
10.1%
+5.9pp
Other
13.5%
14.4%
9.6%
-3.9pp
Data Infra
16.7%
12.4%
9.2%
-7.5pp
Healthtech
2.1%
3.6%
4.1%
+2.0pp
Industrial/IoT
7.3%
4.1%
4.1%
-3.2pp
Climate/Energy
1.0%
0.0%
1.4%
+0.4pp
Consumer
2.1%
1.5%
0.9%
-1.2pp

Share of Series A rounds by category, grouped by calendar year of the round. Columns sum to 100%.

AI/ML and Fintech are the only categories with meaningful share gains (+7.2 and +5.9 points). Cybersecurity and Healthtech each added a few points. The biggest losers are Data Infra (-7.5), DevTools/Infra (-5.3), Other (-3.9), and Industrial/IoT (-3.2). The magnitudes do not support the framing of a market-wide rotation into AI. They support a slower story: the 2015-18 Series A default was a data-or-tooling company sold to other software teams, and that default lost its slot.

The 7.2 point shift in AI/ML between 2015-18 and 2022-24 matters, but so does what it displaced. A meaningful chunk came out of Data Infra, which lost 7.5 points of share partly because Data Infra companies rebranded themselves as AI/ML between 2019 and 2024. The category labels blurred as the underlying product did too.

04The Real Rotation: Out of Generalist Software

The Other bucket is not a huge row in the table. It dropped from 13.5% to 9.6% of Series A rounds, a 3.9 point loss. Taken alone, that number looks modest. Taken together with the 12.8 point loss across Data Infra and DevTools/Infra, a pattern shows up: the categories that lost share were the ones where “we make infrastructure for software teams” used to be a sufficient thesis.

These companies did not disappear. They either repositioned into AI/ML, repositioned into Cybersecurity, or failed to raise. The companies raising Series A now have sharper category identities than the ones raising a decade ago. Founders learned that “enterprise SaaS” is not a narrative, and investors learned that “enterprise SaaS” is not a thesis.

13.5%
Other 2015-18 (13 rounds)
16.7%
Data Infra 2015-18 (16 rounds)
17.7%
DevTools 2015-18 (17 rounds)
16.7
Combined point drop by 2022-24

This is the signal operators should look at if they are building a company today. The AI tailwind is real but saturated. The Fintech tailwind is real but gated by regulation. The Cybersecurity tailwind is real but dominated by incumbents. What changed most is that raising a Series A as a horizontal software or developer-tools company got meaningfully harder. Not impossible. Forty-eight companies still cleared it in 2022-24 across Other, Data Infra, and DevTools. Meaningfully harder.

05Where Each Category Lands on Size and Share

Share tells you how many rounds happened. Round size tells you how much capital each company absorbed. The chart below plots both at once. The x-axis is the median Series A round-size multiple from 2015-18 to 2022-24. The y-axis is the change in each category's share of all Series A rounds over the same window. Bubble size is the number of Series A rounds each category logged in 2022-24.

Round-size growth vs. share change, by category
1x2x5x10x+8+40-4-8Round-size multiple (2015-18 → 2022-24)Share change (percentage points)Gained share, bigger roundsLost share, bigger roundsAI/MLCybersecurityDevTools/InfraFintechOtherData InfraHealthtechIndustrial/IoTClimate/EnergyConsumer
Bubble area = count of Series A rounds in 2022-24. X-axis uses log scale.

Round-size multiple is the ratio of median Series A size in 2022-24 over median in 2015-18. Share change is percentage-point delta in each category's slice of all Series A rounds over the same window.

The chart has four quadrants and the story sits in the contrast between the top and bottom halves. AI/ML and Cybersecurity sit in the upper left: modest round-size growth (1.5x to 1.6x) but meaningful share gains. Data Infra and DevTools/Infra sit in the lower middle: bigger round-size growth (2.5x to 2.7x) but share loss. Each dollar of AI funding went to a new company; each dollar of Data Infra funding went to an existing category with fewer winners per year.

Fintech is the outlier on the right: a 12x jump in median round size, driven by a thin base in 2015-18 when Fintech medians sat at $1.5M. Once Fintech crossed a credibility line with growth-stage investors, round sizes reset hard. Healthtech and Industrial/IoT show the same pattern at lower magnitudes (5.2x and 4.1x): narrow-buyer categories that historically raised small rounds now raise normal-sized ones when they raise.

The broader point is that round-size inflation was not an AI phenomenon. Seven of ten categories more than doubled median Series A size. AI/ML grew in round count, not per-round size. If you hear “AI round sizes are out of control,” the truth is that Series A sizes are out of control, and AI happens to have the most rounds.

The number that does stand out is Vertical SaaS at $24.0M median for 2022-24, four times its 2015-18 value. That is a small sample, only two rounds in 2022-24, and should be read with caution. The direction is consistent with what we hear anecdotally: once a vertical SaaS company clears the “is this a real market” bar, investors fund it aggressively.

06Where the Old Playbook Still Works

In the previous post in this series, we identified 86 companies in the dataset that raised a seed round, did not raise again for at least three years, cleared $3M in annual revenue, and kept revenue per employee above $150K. We called them the efficient compounders: companies that took the path of building a real business instead of raising more capital. The share of each successive seed-year cohort taking this path has shrunk over time, from 11.5% of 2014-16 seeds to 3.1% of 2020-22 seeds.

That rate varies sharply by category.

Efficient-compounder rate by category
Industrial/IoT
16.4%
Healthtech
12.8%
Fintech
9.2%
Other
8.9%
Cybersecurity
8.1%
Data Infra
5.8%
AI/ML
5.2%
Consumer
3.1%
DevTools/Infra
1.5%

Only categories with at least 30 companies shown. Climate/Energy and Vertical SaaS had zero efficient compounders in our dataset but sample sizes were too small to report.

The pattern inverts the progression-rate pattern. Industrial/IoT, which has the lowest raw progression rate, has the highest share of efficient compounders. Healthtech follows. Fintech and Cybersecurity sit in the middle. AI/ML and DevTools/Infra, the categories where capital compounds fastest, have the lowest efficient-compounder rates by a wide margin.

The reading is that enterprise sales cycles and specialized buyer categories still support the old playbook. Industrial companies selling to factories, health companies selling to hospitals, and fintech companies selling to regulated counterparties run slower sales motions that reward capital efficiency. AI/ML and DevTools sell into developer or business-user audiences where growth can compound monthly, which makes not raising economically irrational once the product-market fit signal is clear. The same category-level divergence shows up in revenue per employee: lean sales orgs win where buyers can self-evaluate, and the categories that dominate our efficient-compounder list are exactly the ones where that is hardest.

The non-raising path is not dead. It is category-specific. If you are building in a category with a twelve-month enterprise sales cycle, raising past seed is optional. If you are building in a category with a two-week product-led growth loop, raising past seed is structurally required.

07Implications for Founders and Operators

01The horizontal software bet lost its default slot.
Data Infra, DevTools/Infra, and the generalist Other bucket together gave up 16.7 points of Series A share over the decade. If your pitch deck leads with “we make infrastructure for software teams,” the data suggests you are raising against a shrinking slice of investor interest. Pick a sector, pick a buyer, pick a problem. Do it before you raise, not after.
02Round-size inflation is broad, not AI-specific.
Seven of ten categories saw median Series A sizes at least double between 2015-18 and 2022-24. The pattern is a market-wide shift in what a Series A is, not a category-specific phenomenon. Operators benchmarking against 2019 round sizes are using obsolete reference points regardless of sector.
03AI/ML grew in round count more than in round size.
AI/ML Series A round count grew from 23 in 2015-18 to 68 in 2022-24. Median round size grew 1.6x over the same window, less than DevTools (2.7x), Data Infra (2.5x), and Fintech (12x). The AI boom, at least at Series A, looks more like “more companies got funded at slightly bigger rounds” than “fewer companies got funded at much bigger rounds.” Plan for a crowded category, not a winner-take-most race.
04The non-raising path survives where sales cycles are slow.
Industrial/IoT (16.4%), Healthtech (12.8%), and Fintech (9.2%) have the highest efficient-compounder rates. These categories share a structural feature: long sales cycles, enterprise pricing, and narrow buyer populations. AI/ML (5.2%) and DevTools/Infra (1.5%) have the lowest. If you want the option to build without raising past seed, the category you pick in year zero largely determines whether that option stays available. Workforce contraction data tracks the same split: the departments under the most pressure are the ones AI can most readily substitute for.
05Category boundaries are blurring, and it shows in the data.
Data Infra lost Series A share between 2019-21 and 2022-24, some of which shifted to AI/ML as formerly-data-focused companies rebranded around model infrastructure. Category membership is a narrative choice as much as a product choice. Founders who own that choice early and can defend it against the “you're actually an X company” question from investors progress faster.

08Methodology

Dataset: 1,220 US venture-backed companies enriched with full funding histories via the Apollo API. Classifier: keyword-based mapping against ten category dictionaries, minimum two hits per company, 79.8% classified cleanly. Round sizes: medians of disclosed amounts in USD. Efficient-compounder filter: ≥36 months since last round, ≥$3M revenue, ≥$150K revenue per employee, <$5M total funding, ≥10 employees. Category shares are directional; revenue figures are Apollo estimates.