Theses›Software & applications›AI value capture moves downwards

AI value capture moves downwards

The AI application/agent software layer is where AI revenue SHOWS UP but not where it is KEPT — value behaves as an ingredient that flows to whoever already owns the data, distribution and workflow (incumbent SaaS re-rating on AI-attach) plus…

  • Bull
  • Closing
  • $538B by 2035

The callThe old view, that the application layer shows AI revenue but does not keep it, is losing: startups' share and margins are both rising.

D1D2R1R2R3C1foragainst D1D2R1R2R3C1foragainst
Fig. 1The balance. Drivers and catalysts against restraints, each block sized by its weight; catalysts hang dashed because they have not happened yet, sized by their odds. It tips toward the drivers.
  1. D1Agents turn software budgets into labour budgets: enterprises pay for work done rather than seats, and $234B of incumbent application spend is at risk by 2030.
  2. D2Independents are keeping more, not less: the startup share rose from 36% to 63% and gross margins widened 41%, 45%, 52% across 2024-26.
  3. R1Incumbents bundle: Microsoft 365 E7 ships Copilot and agents inside workflows it already owns, and Salesforce does the same with Agentforce.
  4. R2Inference eats about 23% of revenue, so even winners carry margins near 52% against the 80% software norm, and the labs are moving into first-party apps.
  5. R3At this age durable revenue cannot be told from experimental AI budgets that churn; Unify archived its main code repository in October.
  6. C1The margin checkpoint: whether independent AI-application leaders outside the top few show widening or shrinking gross margins, which settles whether the application layer keeps value.31 Mar 2027
Rests onthe beliefs underneath
Supported by · confidence
What would prove itdated tests, open and resolved
20272028today12predictionswatch 20272028today12predictionswatch
Fig. 2Each ring is a dated test: dashed while open, struck through when overdue, ticked or crossed once resolved.
  1. 1By end-2027 the AI application layer's value stays CONCENTRATED in incumbent AI-attach + the top 3-4 flywheel owners (code/law/support).Dec 2027 · open
  2. 2Value concentrates in incumbents plus the top 3-4 flywheel owners, or a broad basket of independents keeps growing above 40% with widening margins.Dec 2027 · open
Expressionslisted, private, technologies
  • CRMSalesforce
    • cheap
  • MSFTMicrosoft
    • full
  • NOWServiceNow
    • cheap
  • GOOGLAlphabet (Google)
    • cheap
     Sells Gemini assistants to Workspace business customers and Google AI consumer subscriptions, plus the Gemini Enterprise agent platform.
  • WDAYWorkday
    • cheap
     Sells AI agents for HR, finance and IT on its HR and finance software to enterprises, priced through usage-based Flex Credits.
  • INTUIntuit
    • not screened
     Sells the Intuit Assist generative AI assistant inside QuickBooks and Mailchimp to small businesses.
Private · 13 private companies, 11 of them in the US.

Otera (lntelligent NLU GmbH) · Fermat (Batou XYZ SL) · Unify (Deep Ivy Inc) · Wnstn Inc. · OpenAI · Anthropic · Cursor (Anysphere) · Harvey · Sierra · Glean · Cognition · Perplexity · ElevenLabs

Technologies
Who has already bet on itprivate rounds, from public sources

$197B raised by 9 companies, $197B of it in the last three years. Wave under way: Large rounds in the last three years; the category is funded and getting crowded.

Backed by a top-tier venture firm: OpenAI, Anthropic, Cognition, Cursor (Anysphere), Sierra, Harvey, Glean. Backed by a strategic investor: OpenAI, Cursor (Anysphere), Perplexity, ElevenLabs.

1 more funded company in the market.

What changedthe record, newest first
  • 26 Jul 2026My own primary source measured the opposite of what I assumed, in both level and direction.