Angel and VC investors drop an AI startup when its margin after AI model costs falls from 70 to 45 cents on the dollar: 12 of 30 would likely invest at 70 cents, 2 at 45, and none at 20

In November 2026, 12 of the 30 US angel, venture and fund investors on the Brox panel rated themselves likely to put $50,000 into a fast-growing AI software startup that keeps 70 cents of each revenue dollar after paying for its AI models, against 2 of 30 at 45 cents and none at 20 cents, and 20 of the 30 named a product that is a thin layer on another company’s AI model as what would most make them pass.

Lilith, Brox research agent· Reviewed: Lilith, AI reviewed· Published ·Updated
Who was asked
US angel and VC investors, 61 clones, US
When
November 2026 (fielded 1 November 2026)
How
Predicted responses from Brox clones, each built from a long-form interview with a real person.
What changed from the obvious question
The obvious question asks what VCs look for, which investment firms answer with checklists and no numbers. We turned it into a threshold: how likely these investors are to back one otherwise attractive AI startup as its gross margin after AI model costs falls from 70 to 45 to 20 cents on the dollar, with one forced pick each for the top reason to pass and the top reason to invest, asked before the margin question.

Section 01: What the panel said

  1. #

    12

    Offered $50,000 of an AI startup selling software to businesses, valued at $20 million with $1 million a year in revenue growing three times a year, 12 of the 30 private-market investors rated themselves likely to invest (4 or 5 on a five-point scale) when it keeps 70 cents of every revenue dollar after paying for its AI models, 2 of 30 at 45 cents and 0 of 30 at 20 cents; no clone rated a lower margin higher than a higher one

    base 30 · Invest in private companies (angels, professionals, through funds or syndicates)

  2. #

    20

    Asked for the ONE thing most likely to make them pass on an early-stage AI startup, 20 of the 30 private-market investors chose a product that is mostly a thin layer on top of another company’s AI model, 6 founders without deep industry experience, 2 a valuation too high for its revenue, and 1 each AI model costs eating the margin and no data of its own; across all 61 clones the thin layer led with 31 (50.8%)

    base 30 · Invest in private companies

  3. #

    0

    Asked what would most make them want to invest, 0 of the 30 private-market investors chose profit margins as high as a traditional software company’s: 12 chose data of its own that competitors cannot get, 9 paying customers who renew with revenue growing fast and 9 founders with deep industry experience

    base 30 · Invest in private companies · contrast

  4. #

    11

    No clone rated any version 5 (certainly would), and 11 of the 30 private-market investors rated all three versions 1 (certainly would not); 10 of those 11 explained it by their own sectors, strategy or appetite for risk rather than the margin

    base 30 · Invest in private companies · caveat

Offered $50,000 of an AI startup selling software to businesses, valued at $20 million with $1 million a year in revenue growing three times a year, 12 of the 30 private-market investors rated themselves likely to invest (4 or 5 on a five-point scale) when it keeps 70 cents of every revenue dollar after paying for its AI models, 2 of 30 at 45 cents and 0 of 30 at 20 cents; no clone rated a lower margin higher than a higher one

70 cents of each dollar (base 30)
12
45 cents (base 30)
2
20 cents (base 30)
0

Base 30 clones.

02Where do angel and venture investors walk away from an AI startup?

Somewhere between 70 and 45 cents of gross margin on each revenue dollar. We described one AI startup to 61 wealthy US technology and venture investors on the Brox panel: it sells software to businesses, it is raising at a $20 million valuation, and it has $1 million a year in revenue from paying customers, growing three times a year. We then changed one thing, how much of each revenue dollar it keeps after paying for the AI models its product runs on, and asked how likely they would be to invest $50,000.

Thirty of the 61 invest in private companies today. Among the 13 angels, 6 were likely to invest at 70 cents, 1 at 45 cents and none at 20 cents. Among the 9 who invest professionally, the same fall ran 5, 1, 0. Of the 12 private-market investors likely to invest at 70 cents, 10 were no longer likely at 45 cents. The 31 who do not invest in private companies barely moved off "certainly would not" at any margin (1 of 31 was likely at 70 cents), so this page describes the 30.

That reading is ours. The strictest reading is narrower: the number likely to invest fell at every step down in margin, no clone rated a lower margin higher than a higher one, and we cannot say exactly where between 70 and 45 cents the drop happens.

03Why is a thin layer on someone else's AI model the commonest reason to pass?

Because to these investors it is the margin problem in another form. Of the 20 private-market investors who chose the thin layer as the one thing most likely to make them pass, 19 talked about cost or margin in the same answer and all 20 talked about dependence on the model provider or a lack of defensibility. A product that resells someone else's model pays that company for every answer it gives, and can be copied or overtaken when the model changes.

Investors and non-investors saw this differently. Only 1 of the 30 private-market investors chose AI model costs eating the margin as their top reason to pass, against 12 of the 31 who do not invest in private companies. People who back startups name the cause, the thin layer; people who do not name the symptom, the cost.

04Are margins a reason to invest in an AI startup?

Not for these investors. None of the 30 private-market investors chose software-like margins as the thing that would most make them want to invest; their picks split between data of its own, renewing customers with fast growth, and founders who know the industry. Yet margin was the only thing we changed on the startup, and it took the number likely to invest from 12 of 30 to 2 of 30.

Our reading is that margin works as a gate, not a draw. Nobody backs an AI startup for its margins, but a margin that has gone to the model provider is enough to stop a deal that would otherwise get a meeting.

05What would a founder or an investor-relations lead do with this?

  1. Our reading, not the panel's: put gross margin after AI model costs on the first page of the pitch, beside growth, because investors appear to check it before they get interested.
  2. Our reading: answer the thin-layer question before it is asked, with the data, workflow or model work the startup owns, since for most of these investors that is the same question as the margin.
  3. Our reading: a startup with margins near 45 cents is better served by a plan to raise them than by a stronger growth story, since growth tripling every year did not hold investors at that margin here.

06What else do people ask about this?

  • At what gross margin do investors stop backing an AI startup? Between 70 and 45 cents on the dollar: the private-market investors likely to invest fell from 12 of 30 to 2 of 30, and to none at 20 cents (November 2026).
  • How many investors pass on an AI startup that is a thin wrapper? 20 of 30 US angel, venture and fund investors named a thin layer on another company's AI model as the one thing most likely to make them pass.
  • What share of investors back an AI startup for its margins? None of the 30 private-market investors; 12 named data of its own, 9 renewing customers with fast growth and 9 founders with deep industry experience.

Section 07: What they said, and the interview line it rests on

“I'd put myself at a 2. On paper, a few things fit my profile: $50,000 is right in the sweet spot of what I typically write for angel checks, and I actively look to stay connected with what founders are building in AI. Seeing $1 million in ARR growing 3x year-over-year at a $20 million valuation is undeniably solid top-line momentum. However, having spent decades in enterprise software and cloud infrastructure as an engineering leader and CTO, a 20% gross margin is a massive red flag. True software economics rely on 70% to 80%-plus gross margins. If you are keeping only 20 cents on every dollar because 80 cents goes straight to underlying model providers, you are effectively reselling compute and foundational model APIs with no pricing power and minimal defensibility. Unless there is a credible engineering roadmap to bring model inference in-house, fine-tune smaller open-source models, or drastically reduce token costs, you can't build a sustainable, venture-scale enterprise business on those unit economics. Because I do angel investing partly for intellectual curiosity and mentorship rather than purely optimizing returns, I wouldn't say 'certainly not' (a 1) without hearing the founder's story, but the unit economics make it an uphill battle. So I lean strongly against it at a 2.”

Predicted from this line in the interview

“Then in 2006 I joined what became my defining company - a cloud infrastructure startup when "cloud" wasn't even a common term yet. I came in as VP of Engineering, became CTO within two years.”

Angel investor, US, a former chief technology officer in enterprise software; rated the 20-cent version 2 supports finding 1

“Coming from private equity, I'm fundamentally focused on defensibility, enterprise value, and whether a business actually has staying power. If an early-stage startup's product is essentially just a thin layer or wrapper sitting on top of an underlying foundation model from OpenAI or Anthropic, there is simply no moat. The second the underlying model provider releases an update or a feature that replicates that functionality, the startup's entire value proposition vanishes overnight. While unit economics and valuations certainly matter, you can fix pricing or negotiate terms; you can't fix a business model that has no proprietary foundation to begin with. Without real defensibility, it's just speculation, and I've seen far too many hype-driven things blow up to put money behind that.”

Predicted from this line in the interview

“And I'm suspicious of anything where the primary use case seems to be speculation.”

Private equity professional, US; would pass on a thin layer over another company’s AI model supports finding 2

“Coming from an enterprise data infrastructure background, that is immediately where my attention goes: proprietary data. In today's AI landscape, anyone can call an API or build a wrapper around an existing foundation model, and early growth or renewals can easily evaporate once commoditization hits. Having built an enterprise software company and seen how rapidly the landscape can shift, the only true long-term moat is having data of its own that competitors simply cannot get. Without unique data, you don't really have an enduring business in enterprise AI.”

Predicted from this line in the interview

“Enterprise data infrastructure, very unsexy but real problem.”

Invests in startups through syndicates, US, built an enterprise software company; would back data of its own supports finding 3

Section 08: How much should you trust this?

What we would defend

  • The direction of the margin ladder: no clone rated a lower margin higher than a higher one (0 of 61), and each version was reasoned on its own figure (70 cents named in 25, 45 cents in 36 and 20 cents in 49 of the 61 answers to that version).
  • The order of reasons to pass among the 30 private-market investors: the thin layer on another company’s AI model (20) far ahead of founders (6) and every other reason (2 or fewer), and first among angels (9 of 13), professionals (6 of 9) and fund or syndicate investors (5 of 8).
  • Margins as a reason to walk rather than a reason to invest: 0 of 30 picked software-like margins as the top reason to invest, while the number likely to invest fell from 12 to 2 of 30 as the margin went from 70 to 45 cents.

What we would not

  • The exact margin at which investors walk away: we tested 70, 45 and 20 cents only, so the drop lies somewhere between 70 and 45 cents and we cannot place it more precisely.
  • Any level as a market share: 12 of 30 is a stated intention on a small base of clones, and no clone was certain it would invest at any margin.
  • The whole-panel ranking of reasons to pass: the 31 clones who do not invest in private companies answered hypothetically and named AI model costs most often (12 of 31), against 1 of 30 among those who do.
  • That the thin layer and low margins are separate objections: 19 of the 20 private-market investors who chose the thin layer also cited cost or margin in the same answer.
  • 6 of the 61 answers to the reason-to-pass question reasoned from the valuation or growth figures of the margin question asked later in the study, a small leak between questions.

The main limitation

Every answer to the main questions was inferred from what each person said in interview about their investing, work and attitude to risk, not from anything they said about AI startups or margins: 0 of 61 answers were evidenced on each main question. Each person’s own lines about startup investing ground whether they would back a startup at all (21 to 24 of the 30 private-market investors per version), while the reaction to margin is the clones applying general software economics to those profiles (10 of 30 named a benchmark such as 70% to 80% at 45 cents, 17 of 30 at 20 cents). Read the levels as direction with size, on a base of 30.

Section 09: How was this run?

Panel composition
SegmentClones
Invest directly in startups as angels13
Invest professionally at a VC or private equity firm, or for a family office9
Invest in startups only through funds or syndicates8
Do not invest in private companies31

Question wording as fielded

  1. Which of these best describes how you invest in private companies today?
  2. When you look at an early-stage AI startup today, which ONE of these would most likely make you pass on investing in it?
  3. When you look at an early-stage AI startup today, which ONE of these would most make you want to invest in it?
  4. An early-stage AI startup that sells software to businesses is raising money at a $20 million valuation. It has $1 million a year in revenue from paying customers, growing three times a year. After paying for the AI models its product runs on, it keeps 70 cents of every dollar of revenue as gross margin. On a scale of 1 to 5, where 1 is certainly would not and 5 is certainly would, how likely would you be to invest $50,000 in this startup if you were offered the chance?
  5. An early-stage AI startup that sells software to businesses is raising money at a $20 million valuation. It has $1 million a year in revenue from paying customers, growing three times a year. After paying for the AI models its product runs on, it keeps 45 cents of every dollar of revenue as gross margin. On a scale of 1 to 5, where 1 is certainly would not and 5 is certainly would, how likely would you be to invest $50,000 in this startup if you were offered the chance?
  6. An early-stage AI startup that sells software to businesses is raising money at a $20 million valuation. It has $1 million a year in revenue from paying customers, growing three times a year. After paying for the AI models its product runs on, it keeps 20 cents of every dollar of revenue as gross margin. On a scale of 1 to 5, where 1 is certainly would not and 5 is certainly would, how likely would you be to invest $50,000 in this startup if you were offered the chance?

What was not asked, and why

We held everything but the margin constant (a $20 million valuation, $1 million a year in revenue growing three times a year, a $50,000 check) and did not test margins between 45 and 70 cents. We named no firm, fund, startup or AI model maker and listed no brands. Ask Brox fielded the three margin versions as one question with three rows and added no questions of its own.

What changed from the obvious question

The obvious question asks what VCs look for, which investment firms answer with checklists and no numbers. We turned it into a threshold: how likely these investors are to back one otherwise attractive AI startup as its gross margin after AI model costs falls from 70 to 45 to 20 cents on the dollar, with one forced pick each for the top reason to pass and the top reason to invest, asked before the margin question.

Fielded through Ask Brox on 1 November 2026.

Ask this of your own customers: humans@brox.ai

Related insights

All in this topic