What is a venture capital fund, structurally?

It is usually a limited partnership or LLC with two classes of participants. Limited partners (LPs) — pension funds, endowments, foundations, family offices, funds of funds — contribute the capital and have liability limited to their commitments. The general partner (GP) manages the fund, typically through a management company it controls, and holds decision rights over every investment.

Funds are closed-end and finite: money is raised once, invested over three to five years, and returned over the fund's remaining life, usually ten years plus optional extensions. This is the core difference from a hedge fund or evergreen vehicle — a VC fund cannot simply hold a failing position forever, because the partnership ends.

Capital is committed, not deposited. LPs sign agreements to supply their full commitment when the GP issues capital calls, which is why regulatory language measures fund size as "aggregate capital contributions and uncalled committed capital" — the pool that exists on paper plus the cash actually moved.

The two-tier structure also explains the industry's rhythm. Because a fund is raised once and invested over a fixed window, firms live on a cadence of overlapping generations: while Fund III is being harvested, Fund IV is being invested and Fund V is being raised, with each generation's track record pricing the next.

What legal rules define a VC fund in the US?

The clearest definition is regulatory. Under 17 CFR 275.203(l)-1 — the rule implementing the venture capital fund adviser exemption — a qualifying venture capital fund is a private fund that:

1. Represents to investors and potential investors that it pursues a venture capital strategy;
2. Immediately after the acquisition of any asset, other than qualifying investments or short-term holdings, holds no more than 20 percent of aggregate capital contributions and uncalled committed capital in non-qualifying assets;
3. Does not incur leverage above 15 percent of aggregate capital contributions and uncalled committed capital, with any borrowing on a non-renewable term of no longer than 120 days.

Those three clauses are the whole philosophy of the asset class in legal form: invest in startups, hold mostly qualifying investments, and stay unlevered. Compliance with this rule lets a fund's adviser register as an exempt reporting adviser rather than a fully registered investment adviser — a lighter regime, in exchange for a constrained portfolio.

The caps have practical bite. The 20 percent non-qualifying bucket limits how much of a "venture" fund can sit in public equities or other liquid assets waiting for deals; the 15 percent leverage limit means a venture fund, unlike a bank or a buyout vehicle, cannot borrow its way into a bigger position than its LPs actually funded.

How does the GP make money?

Two revenue lines, fixed by the partnership agreement. Management fees, historically around two percent of committed capital per year, pay for the operation of the firm. Carried interest, historically around 20 percent of the fund's profits above returning capital, is the upside that aligns the GP with LP returns.

The economics only work if the fund returns more than it invests, which sounds obvious and is the model's central tension: a portfolio of illiquid, mostly-failing startups must produce a small number of outsized winners to pay back the whole pool. GPs typically invest across 20 to 40 companies per fund with the explicit expectation that most will return little.

Distributions follow a waterfall: LPs first receive their contributed capital (sometimes plus a preferred return), then the GP receives its carry. The precise stack — fee offsets, clawbacks, hurdle rates — is where fund terms are fought over, and where LP counsel earns their fees.

The fee math also explains firm behavior. A $200 million fund at a two percent fee yields $4 million a year to run the firm — enough for a small team, not a large one, which is why fund sizes and team sizes scale together and why raising the next fund is a permanent institutional preoccupation.

The model also has recent variations worth knowing. Continuation vehicles let a GP roll a winning asset into a new fund rather than sell it at the original fund's deadline; recycling provisions reinvest early proceeds during the investment period. Both bend the classic 10-year clock without breaking the partnership logic that everything eventually must be returned.

Where is the money flowing now?

Into AI, at historic concentration. Crunchbase data reported by TechStartups on May 27, 2026 put global venture funding at $300 billion for the first quarter of 2026, with AI companies taking $242 billion — 80 percent of the total — and just four companies (OpenAI, Anthropic, xAI, and Waymo) accounting for nearly 65 percent of all global venture investment in the quarter.

For the fund model, that concentration is stress. Mega-rounds for a handful of labs consume fund sizes designed for diversified portfolios, pushing traditional funds toward seed stages and spawning larger vehicles for infrastructure-scale checks. The 20 percent non-qualifying asset cap in the regulation remains regardless — it constrains what a qualifying fund can hold, not how large the qualifying checks can grow.

It also warps the median-versus-mean story every quarter: totals look historic while the typical startup's environment may be tightening. Any headline number in this market deserves a concentration check before it is read as news for founders at large.

What should a founder take from the model?

That a fund's behavior is explainable from its structure. The 10-year clock explains exit pressure; the capital-call mechanic explains why "committed" capital can move slowly; the concentration rules explain why funds follow-on aggressively into their winners and prune the rest. When a partner says their fund cannot do a follow-on, the honest explanation is usually arithmetic, not opinion.

The model's last lesson is about failure. Funds shut down quietly, return remaining capital, and move on; the asset class is designed to lose most bets. Founders evaluating an investor are well served by asking where their fund is in its lifecycle — a GP in year two of a fresh fund is a different counterparty from one managing the tail of Fund II.

What does an exit actually do?

Before an exit, a private company's shares are worth what the last funding round says on paper, and selling them is constrained, slow and discounted. An exit converts that paper into something tradeable. An IPO lists new and existing shares on a public exchange, where prices are set daily by trading; an acquisition transfers ownership to a strategic or financial buyer at a negotiated price. Founders, employees with vested options, and investors all use exits to realize returns they have often waited a decade to touch.

The two paths are not equally available. Initial public offerings require scale, audited financials, predictable revenue and a receptive market window — which is why they number in the dozens per year in the United States even in good markets, while acquisitions number in the thousands. Every startup that exits does so through the door that is open.

How does an IPO actually unfold?

Klarna's 2025 offering is a clean public walkthrough of the sequence, documented in CNBC's launch coverage:

1. Registration. The company files a prospectus — an F-1 for foreign issuers, an S-1 for domestic ones — disclosing financials, risks, ownership and share structure.
2. Launch and range. Klarna announced plans to offer 34,311,274 ordinary shares priced between $35 and $37 each, expected to raise up to $1.27 billion on the NYSE under ticker KLAR, at an implied valuation of up to $14 billion (announced).
3. Roadshow and pricing. Management pitches institutions; the final price is set based on demand, above or below the filed range.
4. Listing and first trade. Shares begin trading; the opening print often moves sharply from the IPO price as public demand finds its level.
5. Lockup. Insiders typically cannot sell for roughly 180 days after listing, staggering the supply of shares.

What does an acquisition look like in practice?

IBM's purchase of HashiCorp shows the acquisition path at the same scale. IBM announced completion on February 27, 2025, disclosing that the closing included all of the issued and outstanding common shares of HashiCorp for $35 per share in cash (announced). The deal had been signed the previous spring and spent months in regulatory review before closing — a reminder that large acquisitions are marathon processes, not signing ceremonies.

The mechanics favor certainty. Every share gets the same cash price, negotiated between boards and approved by shareholders; there is no first-day pop or flop, no trading float, and no lockup because there is nothing left to trade. HashiCorp's products — Terraform for infrastructure provisioning, Vault for secrets — simply folded into IBM's automation software portfolio, available from IBM's catalog the day the deal closed, per the company's announcement.

How do the economics compare?

The trade-offs compress into a handful of dimensions:

One number in Klarna's structure deserves emphasis: of the roughly 34.3 million shares on offer, only about 5.6 million were sold by the company itself — the large majority came from existing shareholders cashing out. IPOs are exit events for insiders first and fundraising events second, a fact glossed over in most coverage.

When does each path actually make sense?

The choice is rarely free. Initial public offerings reward companies with a public-market story — growth, durable margins, governance a fund manager can underwrite — and punish those without one, which is why boards treat a cold IPO window the way farmers treat a drought. Acquisitions suit companies whose technology is worth more inside a larger platform than standing alone: HashiCorp's Terraform made IBM's hybrid-cloud portfolio more complete, a logic that supports a cash premium no public market would necessarily pay on day one.

Market timing dominates both. Ritter's long-run dataset exists precisely because IPO volume swings wildly with market conditions — hot years with hundreds of listings alternate with near-empty ones — while acquisition activity is steadier but gated by antitrust review for large deals. Founders do not choose in the abstract; they choose among the offers and windows actually in front of them, often in the same quarter.

Who receives money, and when?

Priority runs through the capital stack. Investors with liquidation preferences receive their contracted multiples before common shareholders see anything; option-holding employees convert only if the per-share price clears their strike price, after taxes. In an all-cash acquisition the timing is mechanical — proceeds arrive at close. In an IPO, insiders wait out the lockup, and the paper value of vested shares moves with the market: Klarna's own debut demonstrated the volatility, with the share price moving sharply from its $35–$37 launch range in the first sessions of trading (documented in launch coverage).

The dilution ledger also differs before any exit. Companies that raise heavily to reach IPO scale — and Klarna, founded in 2005, raised across two decades to get there — spread equity across many rounds, so each founder and employee percentage is the residue of every prior negotiation. Acquisitions can arrive earlier in that sequence, freezing the ledger at a stage where early holders still own more. Neither path changes the arithmetic; both cash it out.

Taxes finish the picture and vary by jurisdiction, instrument and holding period — the one area where general explanations genuinely cannot substitute for individual advice, and where the specifics of options versus restricted stock, and cash versus shares as consideration, dominate what a seller actually keeps.

Where can you check the record yourself?

The public paper trail is unusually good. Registration statements and amendments sit on the SEC's EDGAR database; acquisition terms appear in press releases and merger proxies. For the market-level view, University of Florida economist Jay Ritter maintains the standard academic dataset — IPO statistics covering monthly counts and average first-day returns from 1980 through 2025 — the same series researchers use to study underpricing and hot-and-cold cycles. Any founder weighing the two doors should read one prospectus and one merger agreement end to end; the terms live in the documents, not the headlines.

What does scaling operations actually mean?

A startup's early operations run on osmosis: everyone knows everything, decisions happen in one room, and process is a dirty word. That works at ten people and fails progressively afterward, because the informal channels saturate. First Round Review's interview with Tim Howes, who grew large engineering teams at multiple companies, calls scaling a constant process of reinvention — the structures that worked at twenty people are the bottleneck at eighty.

Howes frames the work around five areas: people, hiring, organization, communication, and quality. None of them is glamorous, and all of them compound. The reinvention framing matters because founders tend to treat operational scaling as a one-time reorganization, when it is actually a recurring rebuild that hits roughly every doubling of headcount.

When do informal roles have to become formal?

The threshold Howes names is blunt: over fifty people, if an organization is not formalizing roles, everybody thinks they are a tech lead. Ambiguity that felt like ownership at fifteen people becomes turf conflict and duplicated work at sixty. The fix is explicit ownership — named owners for components, named decision-makers for classes of decisions — which Howes argues does not reduce autonomy but enables it, because nobody second-guesses a boundary that is written down.

Formalization extends to the ceremonies that carry information. Standing reviews, written decision records, and a weekly operating rhythm look like bureaucracy from the inside and like insulation from chaos two doublings later. The counterpressure is real: process added faster than headcount creates drag, which is why the discipline is to add structure where the informal system has demonstrably broken, not everywhere at once.

Why does the hiring bar decide the ceiling?

Howes's sharpest line in the First Round interview is about compromise: the 'meh' hire is death to an organization, because a company ends up with mediocrity everywhere — weak hires recruit and tolerate weaker ones, and the bar ratchets down with each exception. Operations at scale is mostly a function of trust: managers delegate to people they trust, and the trust pool is set at hiring.

The operational consequence is that interview loops, calibrated rubrics, and written scorecards are not administrative overhead but the mechanism that keeps a hundred-person company coherent. The same logic explains why fast-growing startups centralize hiring standards even as they decentralize everything else — the one process that cannot vary team by team is the definition of good.

How do you run operations when things break?

Google's SRE team published its answer as Chapter 14 of the Site Reliability Engineering book, and the system is a model for any startup's incident process. The principle: everybody involved in the incident knows their role and does not stray onto someone else's turf, and a clear separation of responsibilities allows individuals more autonomy, not less, since they need not second-guess their colleagues.

The transferable sequence for a startup that has never written one down:

1. Name an incident commander before the incident, and give them authority to delegate the other roles.
2. Restrict system changes to the operations lead so two people never fix over each other.
3. Assign one communicator so stakeholders get facts, not hallway rumor.
4. Log decisions and timestamps as the incident runs; memory is not a record.
5. Run a blameless postmortem and file the action items as work, not as a document.

How does communication scale past the one room?

Communication is the first system to fail and the last to be engineered. The SRE chapter's communication role exists because stakeholders demand information precisely when responders are busiest — so updates must be someone's explicit job, issued on a rhythm, while the rest of the team works. Howes makes the same point at company scale: written, regular, and honest communication is what builds trust, and trust is the bandwidth that lets a founder delegate a hundred-person operation through ten leaders.

What breaks first: the founder's calendar?

Before any process fails, the founder's schedule fails. In a ten-person company every decision routes through one desk because that desk is fast; at eighty people the same routing makes the desk the company's rate limiter, and the symptoms are familiar — decisions queued in direct messages, launches waiting on a signature, engineers blocked on a product call. The operations fix is a decision inventory: write down every recurring decision type, and for each one name the person who should make it and what the founder actually needs to know afterward.

That inventory is also the honest test of whether the five areas are real. People, hiring, organization, communication, quality — every one of them either has a named owner or it defaults back to the founder, silently, until the calendar breaks again. Scaling operations, in practice, is the systematic act of giving away the founder's decisions faster than the company creates new ones.

How do operating metrics change with scale?

The dashboard that runs a startup also needs reinvention, because the metrics that described health at launch mislead at scale. Raw headcount growth stops being a triumph and becomes a cost line. Revenue growth percentage gets easier to misread as the base grows, which is why operating teams switch to cohort views — what share of customers acquired in a quarter are still active, still expanding, still profitable to serve. Anecdote loses its seat at the table: at twenty people the loudest customer complaint reaches everyone; at two hundred it reaches whoever happens to be listening, so feedback has to be instrumented rather than overheard.

The incident world offers the cleanest template for that transition. Google's SRE chapter treats the running log — timestamps, decisions, who was told what — as a first-class output of an incident, precisely because memory does not scale and neither does proximity. The same discipline applied to routine operations, weekly metrics with written commentary, decisions recorded where the next hundred employees can read them, is how a company keeps a single version of reality after the single room is gone.

The honest limit of every scaling playbook: none of it is a formula. Howes's five areas and Google's incident roles are structures that worked for particular organizations at particular sizes. What generalizes is the posture — reinvent deliberately, formalize what has broken, protect the hiring bar, and write down how the company responds when things go wrong. A startup that treats operations as a product, with the same iteration and the same honesty about failures, scales without breaking. One that treats it as overhead breaks exactly on schedule.

What is the actual difference between B2B and consumer?

The distinction is the buyer, and everything economic follows from it. A B2B startup sells to organizations — enterprises with thousands of employees, or smaller companies buying software for a team — where the buyer is not always the end user and purchase decisions pass through procurement. A consumer startup sells to individuals, who decide for themselves, pay from their own pocket, and can abandon the product in seconds.

That single difference reshapes the whole company. In Y Combinator's guide to business models, the enterprise category is described as selling large fixed-term contracts to big companies, with primary metrics like bookings, annual contract value, and pipeline. The consumer subscription category, by contrast, is described as usually sold to consumers at lower price points, from a higher volume of customers, with growth driven by scalable, self-serve acquisition channels. The guide is paired with a Startup School talk by YC Group Partner Aaron Epstein, who walks through how to monetize and price each model; the full talk is public and worth an hour of any founder's week.

How do the revenue mechanics differ?

Revenue in B2B arrives in large, infrequent chunks. An enterprise deal can run $100,000 or more per year, close over months of demos and gatekeepers, and produce what the YC guide calls lumpy growth — month-over-month percentages stop making sense when one contract moves the number. Consumer revenue arrives in small, constant streams: a $10 monthly subscription, an in-app purchase, an advertising impression. Each transaction is trivial; the business only works at volume.

The gross-margin story diverges too. E-commerce consumer models carry cost of goods sold on every order, which the YC guide flags as higher COGS meaning lower margins. Software sold to businesses typically has high incremental margins, because one more seat costs the vendor almost nothing to serve.

Why do growth expectations differ so sharply?

Paul Graham's essay Startup = Growth defines the whole category: a startup, he writes, is a company designed to grow fast, and everything else associated with startups follows from growth. But the achievable growth curve looks different on each side of the market. Consumer products can compound virally — every user can recruit the next — which is why consumer wins look like hockey sticks and consumer graves are the most numerous.

B2B growth is bounded by the sales team's capacity. A company selling $100,000 contracts cannot double customers in a month without doubling closers, engineers, and implementation staff. The compensation is durability: a signed enterprise contract renews annually, and switching costs accumulate in workflows and data. Consumer retention is earned every single day, which is why the YC guide lists month-one-to-month-two retention as a primary metric for consumer subscription businesses.

Which model should a founder pick?

The honest answer is that the market picks first — the founder's insight usually fits one buyer type — but the decision can be stress-tested. A practical sequence:

1. Identify who feels the pain most acutely and whether they can pay from a budget they control.
2. Estimate deal size and sales cycle honestly: few customers at high price, or many at low price.
3. Check whether acquisition can be self-serve or requires a sales force, and price that into the model.
4. Pick the two metrics that decide survival — retention and CAC for consumer, pipeline and contract value for B2B.
5. Model the path to profitability on current growth and spend, not on a future funding round.

Can a startup do both?

Some try, usually through a free consumer product that funnels into business seats, or a prosumer tier between the poles. The risk is building two companies at once: one that needs viral loops and obsessive retention work, and another that needs salespeople and implementation. The YC catalog treats these as separate models with separate metrics for exactly that reason — mixing them without deliberate design tends to mean executing neither well.

How does customer acquisition differ?

Acquisition is where the two models spend money in opposite places. A consumer startup's acquisition has to be scalable and self-serve — content, virality, app-store placement, install campaigns — because the revenue per user cannot support a salesperson's time. The YC guide's consumer categories all list growth driven by scalable, self-serve acquisition channels, and the metric that decides survival is whether a customer costs less to acquire than they return in margin over their staying lifetime.

A B2B startup inverts the equation. A salesperson closing six-figure contracts can spend weeks per deal, fly on-site, and still generate a return, because the contract value carries the cost of human selling. That is why the guide's enterprise playbook begins with fee-based pilot contracts and letters of intent: the acquisition motion is a pipeline — top of funnel, demo, close — managed like a forecast rather than an advertising budget. The failure modes invert too. Consumer companies die when acquisition channels saturate and retention cannot hold; B2B companies die when the pipeline math quietly stops working and nobody recalculates it.

Where does churn bite each model?

Churn is consumer economics' silent tax. A subscription product that loses five percent of subscribers monthly must replace more than half its base every year just to stay flat, which is why the YC guide tracks month-one-to-month-two retention as a primary metric and why consumer teams obsess over onboarding, habit loops, and payment recovery. Every retained user compounds; every leaked bucket has to be refilled at acquisition cost.

B2B churn is heavier but slower. A cancelled enterprise contract removes a large revenue slice at once, but the events are rare, visible, and usually preceded by warning signs — usage decline, support escalations, an executive sponsor's departure — that a team can watch. The practical discipline is the same on both sides: measure retention on a defined cohort, not on a blended dashboard that hides the leak. What differs is the clock speed. Consumer churn is a weekly weather report; enterprise churn is an annual seismic event, and each demands its own instrumentation.

What happens to pricing power over time?

Pricing power grows differently in each direction. A B2B product embedded in a customer's workflow accumulates switching costs — data, integrations, trained staff — which supports annual price increases and expansion revenue as customers grow, the pattern the usage-based category in the YC guide formalizes by charging per API request or record so revenue scales with the customer's own volume. The floor is competition and procurement's willingness to run a renewal bake-off.

Consumer pricing power is thinner and stranger. A service millions of people pay a few dollars for cannot raise prices sharply without triggering churn, so the leverage comes from bundles, tiers, and new categories rather than list-price moves — and from the advertising and transaction models, which monetize attention or commerce instead of charging the user directly. The result is a structural asymmetry worth memorizing: B2B companies defend margin with contracts and switching costs; consumer companies defend it with habit and scale. Neither is easier. They are simply different machines, and a founder's honest answer to which machine they want to operate is the first economic decision the company makes.

What stays constant across both models is the underlying math Graham describes: growth as the compass for decisions. Whether the growth comes from a million people paying a little or fifty companies paying a lot, the startup exists to compound. The difference between B2B and consumer economics is not which math applies, but which constraints bind first.

Why do funded startups shut down?

The public record points to three recurring mechanics rather than one story. The first is model error: a core operating assumption — a price forecast, a conversion rate, a payback period — turns out wrong at scale. The second is burn outrunning the ability to raise, which converts a fixable problem into a terminal one. The third is governance: boards choose an orderly wind-down over a down round that would wipe out the cap table's remaining value.

Shutdown post-mortems are the artifacts these endings leave behind: SEC filings, closing announcements, and founder statements. Unlike funding announcements, they are often written under legal constraints, which makes them blunt. Read as a dataset, they are the cheapest due-diligence material available to a founder, because someone else has already absorbed the cost of the lesson.

What the record almost never contains is the version told by the people who lost their jobs. Workforce reduction numbers appear in filings as a line item — approximately 25 percent at Zillow — and the human texture survives mainly in incidental coverage. That gap is worth naming, because the post-mortem genre is written by the parties with disclosure obligations, not by everyone affected.

What does the Zillow Offers wind-down teach about forecasting?

Zillow's Q3 2021 shareholder letter, filed as an exhibit to its SEC 8-K, is unusually candid about the arithmetic of failure. The company had publicly set unit-economics guardrails of plus or minus 200 basis points of breakeven for a business built on forecasting home prices three to six months ahead. Instead, Zillow Offers' unit economics swung approximately 1,200 basis points from Q2 to the expected Q3 result — a miss several times larger than the business could absorb, as the filed letter documents.

The stated conclusion deserves quoting for its candor: further scaling Zillow Offers was "too risky, too volatile to our earnings and operations, provides too little opportunity for return on equity, and serves too narrow a portion of our customers." That is a company with a profitable core business choosing amputation over escalation — the rare shutdown decision made from strength rather than desperation.

What does Fast's collapse show about burn?

Fast, a one-click checkout startup, shut down three years after founding despite raising more than $120 million from investors including Stripe, as NPR reported on April 5, 2022. The company failed to raise more capital or find a buyer, and CEO Domm Holland announced the closure that day. Payments Dive's coverage put the figure at $124 million attracted since founding.

The most quoted line of the post-mortem is Holland's own: "After making great strides on our mission of making buying and selling frictionless for everyone, we have made the difficult decision to close our doors." The gap the record exposes is between that framing and the operating numbers — a company burning tens of millions annually against revenue reported in the low hundreds of thousands. Burn without a visible path to unit economics is the classic pre-mortem signature, visible in Payments Dive's report.

How does an orderly shutdown actually unfold?

The public filings and closing statements sketch a repeatable sequence:

1. The board concludes that additional funding is unavailable on acceptable terms, and a buyer search fails.
2. A wind-down plan is written: asset sales, contract terminations, and an employee timeline with severance.
3. Customers and partners are notified with an end-of-service date and data-export arrangements.
4. Regulatory filings and public statements are made — an 8-K for public companies, a closing note for private ones.
5. Remaining capital is distributed to creditors and preferred shareholders in order of liquidation preference.

What do these two records have in common?

Set side by side, the Zillow and Fast records rhyme in three places. Both companies had raised large sums — Zillow was a public company funding Offers from a profitable core; Fast had attracted $102 million in a single Stripe-led round, as NPR reported. Both hit a wall that more money could not fix: Zillow's forecasting error at scale, Fast's missing unit economics. And both ended with a public statement that named the decision without naming the mechanics in full.

The differences are just as instructive. Zillow shut down a division while keeping the company; Fast shut down the company itself. Zillow's record is a formal SEC exhibit with numbers attached; Fast's is a closing note and press coverage, which is what most startup post-mortems actually look like. That asymmetry is a reminder of what the public record can and cannot support: filings give arithmetic, coverage gives narrative, and neither gives the internal deliberations that produced the decision.

For anyone building a post-mortem library, the collection rule follows from this: prefer documents written under disclosure obligations — filings, wind-down notices, creditor letters — over retrospective interviews. The obligation is what keeps the numbers honest.

What should founders take from these records?

First, publish and track your guardrails: Zillow's plus-or-minus 200 basis point target is what made its failure measurable and its decision defensible. Second, treat burn as a countdown clock, not a scoreboard — Fast raised more money than most startups ever see. Third, shutdowns executed early preserve the profitable parts of a business; Zillow's core marketplace continued operating. The post-mortem record rewards founders who read it before they need it.

Fourth, write the wind-down plan while the company is healthy. The companies whose closures read as orderly are the ones that had a sequence ready: a filing, a customer notice, an employee timeline. The sequence in these records is not complicated — what is rare is having thought it through before the board meeting where it becomes urgent.

Finally, notice what is absent from both records: a dramatic single cause. No one decision at Zillow or Fast ended the company; a target missed by an order of magnitude and a cost base without matching revenue did the work over quarters. Post-mortems that name one villain are usually marketing in the other direction.

The last lesson is about reading the genre itself. A shutdown statement is a negotiation artifact: it owes candor to regulators and composure to customers, which limits how much diagnosis it can contain. The filings underneath it — the exhibit, the creditor notice, the wind-down timeline — are where the honest numbers live. Read the statement for the decision, and the filing for the reason.

What is the difference between a business model and a revenue model?

The terms get used interchangeably, but they answer different questions. A business model describes how a company creates and delivers value — what it builds, for whom, through which channels. A revenue model describes only the monetization layer: the pricing mechanism attached to that value. A revenue stream is narrower still: one specific source of income, and a single company typically runs several streams on top of one revenue model.

Paddle's guide emphasizes that the revenue choice feeds back into everything upstream. A subscription model pushes a startup toward sales teams and retention metrics; a transaction model pushes it toward volume and checkout conversion; a usage-based model pushes it toward metering infrastructure before the first dollar arrives. Founders often pick the model by imitation — whatever the category leader uses — when the honest criteria are how customers prefer to pay and what the product's cost structure can support.

How do the five core models actually charge?

Strip away the branding and most tech startups run one of five mechanisms, or a hybrid of them:

Each row trades predictability against alignment. Subscriptions are predictable but can charge heavy users too little and light users too much; usage-based pricing fixes the alignment and sacrifices the predictability, because revenue now moves with the customer's own activity. Transaction models concentrate risk at the moment of purchase, and advertising models make the paying customer someone other than the user — a distinction that quietly reshapes product decisions.

Why is usage-based pricing spreading through software?

Because software consumption has become measurable at fine grain — API calls, tokens, compute-minutes, shipments — and customers increasingly prefer to pay that way. Zylo's analysis, published May 29, 2025, describes the shift bluntly: vendors are moving away from predictable seat-based models toward usage-based pricing, where charges flex based on how much a service is actually used. The same report frames the tradeoff from the buyer's side: greater flexibility and potential cost efficiency, but new blind spots, risks, and budget volatility that can catch even mature software management programs off guard.

For the startup, the model's appeal is upside capture: the customers who get the most value pay the most, without constant re-pricing negotiations. The cost is operational — metering, rating, and billing systems must be accurate and auditable from day one — and financial, since investors value recurring revenue predictability and usage-based revenue is inherently lumpier. A quarter in which customers simply use less service is a quarter in which revenue declines with no churn at all, a pattern subscription businesses never see and usage-based businesses learn to forecast around.

How should a founder choose between them?

Start from the unit of value, not from the competitor's pricing page. If value accrues continuously and evenly, subscriptions map cleanly. If value arrives in bursts tied to workload, usage-based maps better, and hybrid designs — a base platform fee plus metered overage — hedge between the two, which is why they have become common in cloud infrastructure and AI services alike. If value is delivered once per transaction, forcing a subscription merely adds churn risk and refund friction.

Then stress-test the choice against three questions Paddle's framework implies: what does this model do to the sales motion; what does it demand in billing infrastructure on day one; and how will it read in a fundraise. A model that maximizes headline growth but produces unstable revenue can be punished in diligence; a model with boring, compounding recurring revenue is frequently valued higher precisely because it is boring. The model also determines which metrics the company can even report honestly — annual recurring revenue is meaningless under pure transaction pricing — and metrics, once embedded in board decks, are hard to replace.

What breaks first under each model?

Every revenue model has a characteristic failure mode, and knowing it in advance is cheaper than discovering it in a board meeting. Subscriptions break through silent churn: usage decays months before cancellation, so a dashboard that watches only revenue learns about dissatisfaction a quarter late. Usage-based models break through metering disputes: every invoice becomes a negotiation when the customer's records disagree with the vendor's. Transaction models break through demand shocks, because there is no recurring base to cushion a slow quarter. Licensing breaks through procurement cycles that stretch deals across fiscal years. Advertising models break through attention shifts that no contract can prevent.

The defenses are correspondingly specific. Subscription businesses instrument engagement as a leading indicator of churn. Usage-based businesses invest early in transparent metering and usage dashboards customers can query. Transaction businesses manage pipeline coverage rigorously, because a thin pipeline is the earliest visible warning. None of these defenses is exotic; what separates durable companies is matching the defense to the model from the start instead of retrofitting it after the first crisis.

Can a startup change its revenue model later?

Yes, but the longer it waits, the more expensive the change becomes. Every revenue model accumulates dependencies: contracts written around the old pricing, metrics reported to the board under the old definitions, compensation plans tuned to the old sales motion, and customers who budgeted around the old math. Moving a mature subscription base onto meters risks churn among customers who lose under the new arrangement, and moving from one-time licenses to subscriptions historically took entire software industries years of transition pain.

The workable pattern is additive: launch the new model for new customers, grandfather or deliberately migrate existing ones, and report both models side by side until the new one dominates. What rarely works is the silent switch — customers read a repricing of the same value as a broken promise, and trust, once spent, is the hardest revenue model of all to rebuild.

What is vesting, mechanically?

Vesting is a restriction on stock ownership: the founder holds shares from the start, but the company retains the right to repurchase any shares that have not yet vested, at the lower of original cost or current fair market value, if the founder stops providing services. Per Cooley GO's founder basics guide, under a typical vesting schedule, the stock vests in monthly or quarterly increments over four years, and there is often a one-year cliff meaning the individual must be with the company for a year to vest the first increment.

The repurchase right is what gives the mechanism teeth. Without vesting, a co-founder who departs in month three keeps their full stake forever; with vesting, the company buys back the unvested shares for what was paid for them — typically a fraction of a cent — and the remaining founders' ownership of the company is restored. The departed founder keeps only what vested.

For founders, unlike employees, vesting is usually imposed retroactively at the first priced financing: investors routinely require founders to accept a four-year schedule starting from the financing date, sometimes with credit for time already served. A founder signing one of these agreements is, in effect, re-earning a large share of the company they founded.

Why would founders vest their own equity?

Because investors fund teams, not cap tables. A venture investment priced on the assumption that a founding team will work for years becomes mispriced the moment half that team departs with half the equity. Founder vesting aligns the incentive: the people holding the majority of the company are the people still building it.

The arrangement also protects the founders who stay, which is the part first-time founders most often miss. In a two-person startup with a 50-50 split and no vesting, one departure leaves a ghost shareholder controlling half the company — someone with no obligation to the company and every reason to hold out in a future acquisition. Vesting removes that scenario before it exists.

Investors are not disinterested here, and founders should read the term as a negotiation rather than a formality. The variables worth negotiating are the total duration, the cliff length, credit for time served before financing, and what counts as a triggering departure — for example, whether termination without cause accelerates rather than forfeits vesting.

How does the standard schedule work in numbers?

The arithmetic of the standard schedule is simple and worth memorizing, because it determines what a departing founder at any month of a company's life actually walks away with.

Under monthly vesting after the cliff, each month from month 13 onward adds one thirty-sixth of the grant. A founder leaving at month 20 has vested 25% plus eight monthly increments; the company may repurchase the rest at cost. Employees typically receive the same structure applied to stock options, where each vested increment becomes exercisable rather than issued.

Vesting schedules can also be longer — later-stage companies and some acquirers extend founders to five or six years — and cliffs can be negotiated away in exchange for other terms. But the four-year, one-year-cliff shape remains the default that any deviation must be argued against.

What happens to vesting when the company is sold?

An acquisition collides with an unvested schedule, and the resolution is an acceleration clause negotiated into the original agreement. Two forms exist. Single-trigger acceleration vests some or all remaining shares upon the acquisition itself. Double-trigger requires two events: the acquisition, and the founder's termination without cause — or resignation for good reason — within a defined window after closing. Cooley's guidance describes the double-trigger form as one that accelerates the vesting of any unvested shares if the company is sold and the employee is terminated without cause within some time period following the closing.

Acquirers generally dislike single triggers, because they buy teams as much as technology and a fully accelerated pool removes the retention lever. Double triggers have become the common compromise: the founder who is pushed out is protected, while the founder who stays joins the acquirer's retention plan like other employees. Founders negotiating acceleration should expect the double-trigger form with a 12-to-24-month post-closing window as the market standard.

The absence of any acceleration clause is also a negotiated outcome, not an oversight — plenty of founder agreements have none, leaving unvested shares to be handled case by case, or by the acquirer's retention package, at the time of a sale.

What is the 83(b) election, and why is the clock short?

Section 83(b) of the U.S. Internal Revenue Code lets a founder elect to treat restricted, unvested shares as vested immediately for tax purposes. Cooley's guide to the election states the mechanics: within 30 days of grant the taxpayer can file an election with the Internal Revenue Service to treat the unvested or restricted property as vested immediately at the time of grant, including the shares' fair market value in income and paying the corresponding tax then. Without the election, each vesting tranche is taxed as ordinary income at its value on the vesting date.

The reason the election matters is timing. At founding, shares are typically worth fractions of a cent, so the tax on a full grant is negligible; at each later vesting date, the shares may be worth dramatically more, taxed at higher ordinary-income rates. The election converts years of future ordinary income into one small immediate payment and starts the long-term capital gains holding period on day one.

The deadline is unforgiving, which is why it deserves its own checklist:

1. File the written election with the IRS within exactly 30 days of the stock grant — no extensions, no exceptions for missed mail.
2. Send a copy to the company and keep proof of mailing and delivery with the corporate records.
3. Attach the election to that year's personal tax return.
4. Confirm every co-founder has done the same, since the election is individual.

A missed 83(b) is among the few genuinely unfixable mistakes in startup finance: the default taxation applies, and no later filing can restore the election. Founders who understand vesting, acceleration, and this thirty-day window hold the full mechanics of their own equity — which is the minimum equipment for negotiating any of it.

What should founders check in their own vesting agreements?

Because the terms look boilerplate and are not, a short audit of any founder vesting agreement pays for itself. The checklist, in the order disputes usually arise:

1. Start date of the schedule — from company founding, from a specified earlier date, or from the financing that imposed it.
2. Cliff length and credit for time already served, which together determine what a departure in year one actually costs.
3. Vesting frequency after the cliff — monthly is standard and quarterly is common; the difference matters in a mid-year departure.
4. Definition of a termination for cause versus without cause, since the definition controls both forfeitures and any acceleration.
5. Acceleration clause, if any: single or double trigger, percentage accelerated, and the post-closing window.
6. Repurchase mechanics: price formula, who exercises, and any transfer restrictions on vested shares.

Each item is a term a lawyer can negotiate in an afternoon and a founder can live with — or regret — for the life of the company. The pattern across startup post-mortems is consistent: equity disputes between founders rarely involve dishonesty; they involve standard documents whose mechanics nobody re-read carefully while everyone still got along.

What is a go-to-market motion, exactly?

A go-to-market motion is the repeatable system by which prospects discover, try, buy, and expand a product. It is not a marketing campaign or a pricing page; it is the pipeline itself. Every motion has three load-bearing parts: acquisition (how users arrive), conversion (how they become paying customers), and expansion (how they spend more over time).

The two canonical motions differ on all three axes. A sales-led company acquires through outbound prospecting and marketing-qualified leads, converts through salespeople running structured deals, and expands through account executives renewing and upselling contracts. A product-led company acquires through the product being usable — usually free — from the first click, converts usage into payment, and expands usage seat by seat and feature by feature. Hybrid motions, which a16z general partner Martin Casado argued are now the default for modern B2B companies, sequence the two rather than choosing one.

How does product-led growth work?

Product-led growth turns the product into the acquisition, conversion, and expansion engine. Users sign up without talking to anyone, hit value in minutes, and invite colleagues; the viral loop and the billing meter are both inside the software. Writing for SaaS billing platform Maxio in September 2024, the company's team framed the distinction plainly: a product-led GTM relies on the product itself to attract and convert potential customers, while sales-led relies on a team that actively reaches out, pitches use cases, and converts leads.

The economics cut both ways. Product-led companies typically spend less per acquired user and shorten time-to-revenue through frictionless adoption, because a credit card and a work email are the whole buying process at the bottom of the market. The cost is engineering: the product must onboard itself, measure its own usage, and meter billing accurately, which is a substantial investment before the first enterprise deal ever closes. Self-serve products also generate a flood of small accounts that demand automated support and pricing infrastructure.

The model fits products that are individually useful, quick to evaluate, and cheap to serve: developer tools, collaboration software, and design utilities are the classic categories. It fits products requiring committee approval, custom integration, or compliance sign-off far less well — a buyer cannot self-serve a nine-month procurement process.

How does sales-led growth work?

Sales-led growth puts people between the prospect and the contract. A sales team qualifies inbound and outbound leads, runs discovery calls and demos, negotiates pricing, and closes annual or multi-year deals. The motion trades higher cost per deal for higher contract values, longer commitments, and direct control over the pipeline.

Per Maxio's comparison, sales-led generally has a longer sales cycle driven by discovery calls, sales calls, and budget approvals. Each of those steps adds weeks and payroll. A sales-led startup hiring its first sellers typically needs meaningful funding before revenue: salaries, a CRM, and a demand-generation budget all arrive before the first logo does.

The advantage is precision at the top of the market. When a product costs six figures, changes a customer's operations, and needs executive sponsorship, a structured sales process is not overhead — it is how the buyer's organization actually processes a purchase. Sales-led companies also get qualitative feedback from every lost deal, which product-led companies must instead infer from usage telemetry.

Can a startup run both motions at once?

Yes, and increasingly it must — but sequencing matters. In the a16z analysis, Casado described the pattern he called B2B growth sales: companies that decided to go to market in a way that's very different from traditional B2B, first running a consumer bottom-up motion and then overlaying sales after that. The argument, made in August 2019, was that this combination favors startups precisely because incumbents struggle to defend against it.

The failure mode is running both badly at the same time. A company that hires enterprise sellers before the product can self-serve burns cash generating leads the product cannot convert. A company that stays self-serve forever leaves its largest accounts underserved, since a procurement team will not find its own way through a free tier. The standard sequence is bottom-up adoption first, then a sales layer once usage data reveals which accounts are worth a dedicated human conversation.

How does a founder actually choose?

The decision reduces to a handful of checkable questions rather than taste. A founder can work through them in order:

1. Can a new user reach meaningful value alone, in one session, without training? If no, sales-led.
2. Is the expected first contract under roughly a few thousand dollars a year? If yes, product-led can carry it.
3. Does the buyer need security review, legal review, or a committee? Each yes pushes toward sales.
4. Can the product meter usage and enforce limits programmatically? If no, that engineering is the entry fee for product-led.
5. Is the market a handful of large accounts or a long tail of small ones? Concentrated markets reward sales; long tails reward product.

Neither motion is a strategy by itself; each is an infrastructure choice that the strategy then exploits. Founders who name their motion explicitly tend to hire, price, and fundraise in patterns that match it — and to notice earlier when the market pushes them toward the other one.

What does the choice mean for fundraising and hiring?

The motion decision reaches the cap table before it reaches the revenue statement. A sales-led company's first major costs are people: sellers, a sales engineer, and the demand-generation budget that feeds them. Investors funding that motion underwrite burn against a pipeline of large contracts, and the milestones are measured in bookings rather than signups. A product-led company's first major costs are engineering: self-serve onboarding, usage metering, billing, and the analytics needed to convert telemetry into sales leads. Its milestones are activation rates and expansion revenue.

The a16z analysis emphasizes a further consequence: bottom-up companies tend to require a lot less money to start, because the product does the work a paid sales organization would otherwise do. That capital efficiency is one reason the growth-then-sales pattern spread from developer tools into security, infrastructure, and vertical software — categories where the buyer historically sat behind a procurement wall that free usage now tunnels under.

Hybrid sequences carry their own hiring signature. The first sales hires in a bottom-up company are typically account executives pointed at the accounts the usage data already flags, not cold-calling generalists. Founders who instead import an enterprise sales culture wholesale, before the product telemetry exists to aim it, commonly discover they have built an expensive outbound machine with no qualified accounts to call.

How does the standard accelerator deal work?

Y Combinator's own about page breaks the money into two instruments. As stated in the program documentation on YC's site: "YC invests $500k per company: $125k for 7% on a post-money SAFE and another $375k on an uncapped SAFE with an MFN." The two halves behave differently, and the difference is the whole point.

The first SAFE fixes a price and an outcome on day one: $125,000 buys 7 percent, calculated post-money, so dilution from other investors does not change what the accelerator holds. The second $375,000 converts later at whatever terms the next priced round sets, provided they are no worse, which is what the most-favored-nation clause enforces. Founders get more total capital without surrendering more fixed equity, and the accelerator gets a larger position in the companies that raise well.

The 7 percent figure is the number to underwrite mentally. It is compensation for the program, the network, and the signal of selection, paid in equity rather than fees, and it is permanent in a way program benefits are not. Everything else an accelerator offers is downstream of that trade.

What actually happens during a three-month batch?

The program compresses a company's first year into roughly eleven weeks of concentrated work, and the documentation is unusually concrete about the mechanics:

1. Any startup, anywhere in the world, can apply through an open application process, per YC's FAQ.
2. Accepted founders relocate, since the batch runs in person in San Francisco, opening with a three-day retreat and weekly meetups.
3. Startups are sorted into three groups, each led by YC partners who advise founders in office hours.
4. Each group splits into sections of six to ten companies, keeping a small-group setting inside the larger batch.
5. The three-month cycle closes, and the alumni network takes over as the durable asset.

The FAQ is explicit that the in-person format is a considered choice: "We briefly did run YC remotely during Covid, but since 2022 it has been back in person. We've found that YC works much better in person," states the FAQ on YC's site. The structure is not decoration; density of contact is the product, and the weekly rhythm exists to force decisions that solo founders defer.

What does the accelerator get in return?

Equity, purchased at a fixed price, in volume. An accelerator's portfolio economics work because most investments fail quietly and a small number return the fund. A fixed 7 percent position across hundreds of companies per year turns individual judgment into a diversified book, which is why the model scales while partner-driven seed funds stay small.

Selection does most of the work before the program starts. The documentation notes that startups arrive at all different stages, some not yet working on anything, which means the batch's value is partly in standardizing the earliest, most erratic phase of a company's life rather than in teaching a secret curriculum.

Why do accelerators exist at all?

The batch solves three early-stage problems at once. It standardizes a first check for teams that lack traction or connections. It compresses advice into office hours with people who have watched hundreds of the same mistakes repeat. And it manufactures a peer group, which the documentation treats as a permanent feature rather than a perk: the alumni community is described on the about page as an increasingly valuable resource that continues long after the three-month cycle ends.

None of this requires believing accelerator graduates outperform everyone else, and no figure on this page claims they do. The honest framing is that a batch converts an unstructured first year into a scheduled one, with the schedule itself as the deliverable.

How does the batch model compare with a seed fund?

The instruments overlap but the services differ. A seed fund prices each deal on its own merits after diligence; an accelerator posts one public price and competes on volume and selection instead. A fund's help is continuous and financial; a batch's help is compressed and social, front-loaded into eleven weeks and then handed to the network.

For a founder, the practical comparison is about fit rather than prestige. Teams that need a structure, a deadline culture, and a first check at speed fit batches. Teams that need a large check, patient capital, and board-level guidance per company fit funds, and the two are not mutually exclusive across a company's life.

There is also a signaling asymmetry worth naming. A batch acceptance is a public, dated endorsement that investors read as pre-screening, while a quiet seed round carries no comparable badge. Founders raising in crowded categories sometimes take accelerator terms precisely for that signal, and it is a legitimate reason, separate from anything the program teaches.

The reverse signal exists too. Declining a batch to stay unlisted suits companies whose edge is confidentiality or whose buyers would not care. The signal only has value where the audience respects its source, and founders are the ones who know their audience.

Who should skip an accelerator altogether?

Founders with clear distribution, committed capital, or a running revenue engine trade the most for the least. The batch model prices its help in permanent equity, and a company that already knows its next three moves pays full price for repetition. The documentation itself frames the program as serving teams at the start of the curve, including some that have not yet begun working in earnest.

A useful stress test is to price the alternatives. The same dilution could buy a senior hire, a year of runway, or a distribution deal, and a batch competes against all three. What the program genuinely monopolizes is time density: eleven weeks of forced contact with partners, peers, and alumni that would otherwise take a year of cold outreach to assemble.

The decision rule fits on one line: take the deal when access and structure are the binding constraints, skip it when they are not, and either way read the standard terms on the accelerator's own page before signing.

How does selling equity shrink existing stakes?

A priced round creates new shares and sells them to investors at an agreed price per share. The percentage the new investors receive equals the money invested divided by the post-money valuation; here, $126 million over $3.5 billion is about 3.6 percent. Every founder, employee and earlier investor keeps the same number of shares but owns a smaller fraction of a larger company, a mechanic that applies identically to any priced round using the announced figures.

The announcement documents the round's structure: led by Avenir Growth with participation from CapitalG, General Catalyst, Andreessen Horowitz, Kleiner Perkins and strategic healthcare investors, bringing total funding to $404 million per the round coverage. Each of those earlier checks did the same thing this one does: converted money into ownership, one layer at a time.

When is dilution worth it?

The trade is only rational when the round grows the denominator faster than it shrinks the fraction. A founder holding 20 percent of a company worth $1 billion is better off than one holding 40 percent of a company worth $200 million, which is why investors and founders negotiate valuation so hard. The protection against unfavorable dilution lives in the term sheet: anti-dilution provisions, liquidation preferences and option-pool sizing all shift who absorbs the shrinking.

Hippocratic AI's successive rounds show the favorable case, where each round's dilution came with a step up in price per share, meaning early holders own less of the company but their slices are worth more on paper. The unfavorable case is the down round, where new money prices lower and existing stakes shrink without compensation, which is when the term-sheet details start to matter.

What should founders watch in a term sheet?

Three lines dominate the dilution math. First, the option pool: investors often require a fresh employee pool created before their money arrives, which dilutes existing holders rather than the new investor. Second, liquidation preferences determine who recovers capital first at an exit and can make headline valuations misleading. Third, anti-dilution clauses protect investors if a later round prices lower, shifting downside dilution onto common shareholders.

None of these terms appear in a funding announcement, which publishes the round size and valuation but not the preferences stacked on top of them. That is why the $3.5 billion headline is the beginning of the analysis, not the end, and why experienced founders read the full document before celebrating the number in the press release.

What does a Series C actually buy?

A Series C is a late-stage round: by this point a company is expected to be commercializing at scale, not proving its technology works. The documented record here fits that pattern. In the company's announcement, Hippocratic AI reports more than 50 health system, payor and pharma clients across six countries, over 1,000 built clinical use cases, and more than 115 million completed patient interactions with no reported safety incidents, all company-claimed figures.

The buyer list matters as much as the size. The round included not only financial investors such as CapitalG, General Catalyst, Andreessen Horowitz and Kleiner Perkins, but also Universal Health Services and Cincinnati Children's Hospital Medical Center, institutions that also operate as customers. Strategic capital from customers is a recurring feature of healthcare Series C rounds, because it ties the investor's return to products it already deploys inside its own hospitals and clinics.

How is the $126 million structured?

The announcement describes a priced primary round: $126 million of new equity sold at an agreed $3.5 billion valuation, led by one growth investor with existing backers participating. That is the standard shape of a Series C, in contrast to the convertible notes and safe agreements that dominate seed stages. Priced rounds set a per-share price, which fixes exactly how much existing ownership the new money dilutes in exchange for the capital.

Per round coverage at Pulse 2.0, the company said the capital is earmarked for product development, international expansion and strategic mergers and acquisitions. An M&A line item in a Series C use of funds signals a company planning to buy capabilities or customer bases rather than only build them organically. Further detail on the allocation between those buckets was not disclosed, so the split remains company-claimed and unspecified.

Why does this round matter for the market?

The round is a data point in the competition to automate patient access and clinical staffing, a market where labor shortages are documented across health systems in multiple countries. The valuation step also shows where late-stage investors currently price demonstrable adoption: a company with 15 months of commercial operations and reported nine-figure interaction volumes commanded $3.5 billion, per the announcement, a figure that is announced and company-claimed rather than independently verified.

For buyers, the practical consequence is that safety-focused agent products will keep arriving from well-funded vendors, and procurement teams will increasingly compare them on documented evidence of clinical safety rather than on demonstration quality. The company's zero-safety-incident claim is self-reported, and verifying such claims through pilot data and reference customers remains the buyer's work before any contract is signed.