Exits - death, acqui-hire, and the rare durable company#
Contents
Formation cheapens entry. Selection determines who can hold rent. Venture funds the attempt. Exits are where the thesis becomes a number on someone else's ledger - acquisition price, IPO multiple, or zero. Under cheap cognition the exit distribution shifts: more zeros, more small talent deals, fewer mid-tier product acquisitions, and a thin upper tail of companies that own real complements.
The four exit modes, reweighted#
| Mode | What buyer / market pays for | AI-era weight |
|---|---|---|
| IPO / durable independent | Proven margin, growth, governance | Rarer for pure software; still available for complement owners |
| Strategic acquisition (product) | Roadmap fill, customers, data | Selective; buyers rebuild features more often |
| Acqui-hire / talent | Team, not ARR | Rises as product rebuild cost falls |
| Quiet death / fire sale | Nothing / residual assets | Modal outcome; arrives faster |
The mechanism behind the reweighting is simple: if the acquirer can rebuild the product in a quarter with a small team and models, the product is not the asset. What remains acquirable at product-era prices is whatever still takes years to recreate - customer contracts with high switching cost, regulated licenses, proprietary datasets with legal exclusivity, physical footprint, brand trust in a domain where trust is scarce, or a team whose tacit knowledge is the integration layer rather than the CRUD.
Why mid-tier product M&A thins#
The 2010s produced a thick middle: $50–300M acquisitions of vertical SaaS that "fit the roadmap." That middle assumed the roadmap was expensive to fill internally. Software and formation attack that assumption directly. Expect:
- More "build" decisions inside strategics for anything without a unique data or distribution wedge.
- Lower revenue multiples for growth-without-moat companies when they do sell - the buyer is purchasing a temporary customer list and a brand, depreciating fast.
- Faster time-to-kill at the board - when the next pivot is cheap to attempt, the option value of keeping a zombie alive falls; shutdown becomes the clean choice sooner.
- Acqui-hire as the honest name for deals formerly dressed as product acquisitions - watch whether the product is deprecated within a year of close.
for US software companies founded 2024–2027 that exit by 2032, the share of exits that are primarily talent transactions (product sunsetting within 18 months, or explicit acqui-hire labeling) is higher than the corresponding share for 2016–2019 foundations (~65% confidence). The falsifier is a thick middle of product-preserving acquisitions at 2010s-like multiples for undifferentiated AI feature companies.
Who buys, and for what#
| Buyer type | Still pays up for | Stops paying up for |
|---|---|---|
| Big Tech / hyperscalers | Talent, distribution adjacency, rare data, infra defaults | Generic AI wrappers, undifferentiated agents |
| Vertical incumbents | Customers in their channel, compliance position, domain data | Features their own AI program can ship |
| PE / roll-up | Cash flow with switching cost; cost-cut narratives | Growth stories without margin |
| Model labs | Talent, eval data, distribution into enterprises | Application features (they partner or ship) |
| Foreign strategics | Geography-specific licenses and relationships | Pure code without local trust |
Game 1 labs and hyperscalers remain the high-price talent bidders when research or applied taste is scarce. That market is real and does not require the startup's product to work. Founders optimizing only for acqui-hire will still sometimes "succeed" on paper - LPs who thought they bought product equity will not.
IPO as a complement filter#
Public markets eventually demand either growth with a path to durable margin or margin with a story. Red Queen SaaS shows up as usage up, price down, sales efficiency flat - the pattern B5 and Game 3 already flag. Companies that reach IPO scale while pricing seats into a deflating cognitive basket will be punished unless they reprice to outcomes or reveal a non-software rent. The IPO window does not close; it discriminates harder. Regulated fintech with licenses, marketplaces with liquidity, industrial data platforms with hardware-installed bases - these still look like companies. "ChatGPT for X" without X's data rights does not.
Death rate and the welfare reading#
A higher death rate among startups is not automatically a social loss. If attempts are cheap, failed attempts are cheap too - less capital incinerated per zero, fewer years of misallocated senior labor, faster recycling of founders into the next attempt or into employment. The welfare risk is elsewhere:
- Founder and junior apprenticeship - repeated shallow attempts may not produce the same learning as one deep operating experience (formation, Game 4).
- Noise in the ecosystem - customers, open-source maintainers, and investors pay the review tax for a flood of low-quality products (software commons failure).
- Narrative misallocation - talent and LP capital chase theme exposure instead of complements, which is a real cost even when each check is small.
So "more startups die" can coexist with "the startup system is healthier" if the deaths are early and cheap and the survivors are more complement-native. It is unhealthy if death is late (zombie rounds), expensive (oversized raises into feature companies), or if survivors are mostly platform tenants with no equity value.
Second-order: what exits do to formation and venture#
Exit expectations feed back:
- If founders believe product M&A is dead, more optimize for revenue and profitability early (good for selection) or optimize purely for talent signaling (bad for customers, fine for Big Tech HR).
- If LPs believe pure SaaS multiples are structurally lower, funds reweight portfolios toward capital-heavy and licensed categories - which is venture equilibrium B.
- If employees believe equity is lottery tickets on feature farms, comp shifts toward cash, which raises burn for the same headcount and partially undoes formation-cost collapse for hiring (not for tooling).
The rare durable company#
The upper tail still exists. It looks like the selection map's durable rows executed with operational intensity: a firm that becomes the system of record for a workflow and owns the data rights; a licensed intermediary that AI makes more efficient without making entry free; a physical network whose software layer improves density; a verification or trust vendor the rest of the economy has to plug into (compressed version). These companies may use more AI than anyone and still not be "AI companies" in the narrative sense - they are complement companies with AI as cost structure. That is the exit this section treats as success.
Failure modes#
- If rebuild-inside stays hard (context debt, compliance, multi-year integrations), mid-tier product M&A and classical SaaS IPOs persist - this page was early.
- If one platform's marketplace becomes the default exit (acquire-by-ranking, revenue share, or soft acqui), independent exit statistics mislead because value left through a different door.
- If correlated AI liability events freeze M&A diligence (insurance, Uncertainty 6), exit timelines slip without changing the long-run reweighting toward complements.
Ask what still exists if the product is rewritten from scratch in ninety days. If the answer is "nothing," there is no exit - only a slower zero.
Related: Selection · Venture · Software · Game 3 · Assets · Insurance · B5