How much is Stripe paying for OpenRouter?

Neither company has disclosed a price. The New York Times reported $7.5bn on 19 August 2026, citing a person with knowledge of the agreement. Bloomberg had reported a figure above $7bn three days earlier. Every multiple in this analysis uses the $7.5bn figure and inherits its uncertainty.

Why are several different prices circulating?

Because no confirmed figure exists. The Wall Street Journal reported approximately $10bn on 23 July while the parties were in talks. Bloomberg reported more than $7bn on 16 August. Business Insider reported approximately $8bn. The New York Times published $7.5bn with a proceeds split on 19 August. The parties declined to disclose a value.

Is the deal cash or stock?

Not disclosed. Reporting references both cash and stock without stating the split. That split matters, because it determines how much of the reported value is realised at close and how much depends on Stripe equity.

How is the reported price split between founders and investors?

The New York Times reported approximately $1.5bn to the three founders and approximately $6.0bn to investors. Both figures come from the same single unnamed source.

When is it expected to close?

Not stated. The announcement contains no closing date, no conditions and no named regulatory jurisdictions.

What is the strategic rationale?

Stripe optimises the revenue side of its customers' businesses through payment methods, authorisation rates and fraud tooling. OpenRouter optimises the cost side by routing each request to an appropriate model on price, speed and reliability. Patrick Collison framed the combination as addressing both sides of profitability in the AI era.

Who advised on the transaction?

Not disclosed on either side. Advisers are frequently unnamed in private-to-private transactions where no filing requires disclosure.

Does OpenRouter continue as a standalone product?

Not addressed in the announcement. Stripe's precedents point both ways: Bridge kept its team and products after the 2025 acquisition, while TaxJar became Stripe Tax and Okay was folded into internal engineering.

What multiple does the deal imply?

Approximately 53.6 times enterprise value to annualised net revenue, using the reported $7.5bn against an estimated $140m. On gross profit at a reported 70 per cent margin the multiple is approximately 76.5 times. On gross routed spend of approximately $2.5bn it is approximately 3.0 times. All three describe the same transaction.

Which multiple should be used for comparison?

Enterprise value to net revenue, at 53.6 times, is the comparable measure. Gross routed spend is the money customers spend on inference, most of which passes straight through to model providers, so a multiple on that number understates the price by roughly a factor of eighteen.

Is 53.6 times high for a software business?

It is far above conventional software multiples and inside the range recent AI infrastructure transactions have cleared at. Nvidia paid roughly 106 times for Run:ai. Databricks paid roughly 65 times for MosaicML and a reported multiple in the region of 2,000 times for Tabular. Google paid roughly 43 times for Wiz on reported figures.

What justifies a multiple at that level?

Growth. OpenRouter is estimated to have moved from roughly $5m to roughly $140m of annualised revenue in twelve months, an increase of about twenty-eight times. A multiple set against that trajectory behaves differently from one set against a stable revenue base.

How does this compare with the May 2026 valuation?

The Series B was reported at $1.3bn post-money on 28 May 2026. A reported $7.5bn eighty-three days later is an uplift of approximately 5.8 times. The two are different instruments: a Series B prices a minority preferred stake with protective terms, while an acquisition prices the whole company including control.

How does the price compare with capital raised?

Total capital raised across all rounds is approximately $164m. A reported $7.5bn is roughly forty-six times that figure. On the reported split, investors receive approximately $6.0bn, or roughly thirty-seven times aggregate primary capital.

What is the revenue estimate based on?

The $140m annualised figure comes from Sacra, a research firm, published in July 2026. Neither company has confirmed revenue. The estimate is consistent with OpenRouter's published fee schedule and published token volumes, but it remains an estimate.

What does OpenRouter actually do?

It provides a single OpenAI-compatible API endpoint that reaches more than 400 models from more than 80 providers, and routes each request based on price, latency, availability and task fit. A developer integrates once rather than separately with each provider.

How large is OpenRouter?

More than 10 million developers and companies, more than 400 models, more than 80 providers, and a 90-person startup at announcement. Weekly token volume went from 5 trillion in November 2025 to 25 trillion in May 2026. Named customers include NVIDIA, Zoom and Lovable.

Total payment volume reached $1.9 trillion in 2025, up 34 per cent, with the company valued at $159bn in a February 2026 employee tender. The Information reported approximately $6.8bn of net revenue and approximately $3.2bn of free cash flow for 2025. The company remains private.

Is this Stripe's largest acquisition?

Yes, by a wide margin. Bridge at a confirmed $1.1bn in 2025 was previously the largest. At a reported $7.5bn this transaction is roughly 6.8 times that, and larger than every prior Stripe acquisition combined on any reasonable reading of the reported figures.

What is the main economic question in the transaction?

Whether volume growth continues to outrun price decline. OpenRouter charges a flat percentage of what customers spend on inference, and inference prices are falling three to five times a year on published projections. Monetisation per token declined roughly 60 per cent between March and August 2026 on third-party estimates, while volume grew fast enough to more than offset it.

Does neutrality survive corporate ownership?

Both chief executives addressed this directly in the announcement. OpenRouter's commercial position rests on being unaligned with any model provider and on not marking up tokens. Whether that holds is now something the combined company demonstrates through published pricing and routing behaviour rather than something guaranteed by independence.

Are there regulatory considerations?

At the reported size the transaction sits well above the 2026 US premerger notification threshold of $133.9m, so a Hart-Scott-Rodino filing is required. Stripe operates from San Francisco and Dublin and OpenRouter serves a global developer base, so more than one filing is likely. Neither company has named the jurisdictions or a timetable.

Does this set a benchmark for other AI infrastructure companies?

It adds a visible reference point. A three-year-old company with a 90-person team at announcement and an estimated $140m of annualised revenue reportedly clearing at $7.5bn is now a comparable that boards and investors in adjacent businesses will be asked about. How closely it applies depends on how similar the asset is, and the reported nature of the price should be stated whenever it is cited.

Stripe Acquires OpenRouter and Moves Into the Layer That Decides Which Model Gets the Request

01 · What happened

Stripe buys the layer that decides which model gets the request

Announced 19 August 2026. Price not disclosed by either company. Reported at $7.5 billion.

Stripe has agreed to acquire OpenRouter, the model gateway that routes requests across more than 400 models from more than 80 providers. The announcement went out from San Francisco and Dublin on 19 August 2026. It contains no price. The New York Times reported $7.5 billion the same day, citing a person with knowledge of the agreement, three days after Bloomberg reported a figure above $7 billion.

Three years ago OpenRouter did not exist. It was founded in 2023 by Alex Atallah, previously cofounder and chief technology officer of OpenSea, alongside Louis Vichy and Chris Clark. It employs roughly fifty people in New York. In May 2026 it raised $113 million led by CapitalG at a reported $1.3 billion valuation. Eighty-three days later it agreed to sell.

The strategic framing in the release is precise and worth reading literally. Stripe says it already helps businesses maximise revenue by optimising across payment methods, authorisation rates and fraud. OpenRouter optimises the other side, deciding which model handles which task, at what speed and at what price. Patrick Collison framed the combination as helping companies manage both sides of profitability in the AI era.

Tokens are the central currency for companies building with AI, and it is clear that the real-world economic potential will depend on making good use of scarce compute resources.
Patrick Collison, cofounder and chief executive, Stripe

What is in the announcement, and what is not

The release is generous on strategy and silent on terms. It names three customers, NVIDIA, Zoom and Lovable. It states the model and provider counts. It quotes both chief executives at length on the multi-model thesis. It does not state a price, a consideration mix, a closing date, a condition, or a regulatory jurisdiction. A Stripe spokesperson told TechCrunch the company does not comment on rumours or speculation.

That places every number in this article in one of three categories. Published company figures, such as the platform fee and the model counts. Named press reporting, such as the price. Third-party estimates, such as revenue. The table below marks which is which, and the distinction carries through every chart that follows.

02 · The deal at a glance

Terms as reported

Transaction terms Reported Public record Acquiry calculation Not disclosed

Acquirer	Stripe, Inc., private, San Francisco and Dublin
Target	OpenRouter, Inc., New York, founded 2023
Announced	19 August 2026, Stripe newsroom
Reported consideration	$7.5 billion Reported New York Times, citing a person with knowledge of the agreement. Neither company has confirmed a figure.
Earlier reporting	More than $7bn (Bloomberg, 16 Aug); approximately $8bn (Business Insider); approximately $10bn while in talks (Wall Street Journal, 23 Jul)
Reported split	$1.5bn to founders, $6.0bn to investors Reported
Consideration mix	Not disclosed. Cash and stock referenced in reporting, no split given
Capital raised by target	Approximately $164 million across all rounds
Last priced round	$113m Series B, 28 May 2026, led by CapitalG at a reported $1.3bn
Uplift over the May mark	5.8× in approximately three months Acquiry calculation $7,500m ÷ $1,300m
Estimated annualised revenue	Approximately $140 million Indicative Sacra estimate, July 2026. Not a company figure.
Implied EV / revenue	53.6× Acquiry calculation $7,500m ÷ $140m
Platform fee	5.5 per cent on fiat credit purchases, 5.0 per cent on crypto and BYOK overage Public record
Models and providers	More than 400 models from more than 80 providers
Developers and companies	More than 10 million
Named customers	NVIDIA, Zoom, Lovable
Target headcount	90-person startup, per OpenRouter announcement, August 2026
Advisers	Not disclosed on either side
Break fee and deal protections	Not disclosed
Regulatory jurisdictions	Not named
Expected close	Coming weeks, subject to customary closing conditions¹
Status	Agreement announced. Price undisclosed by both parties

Where the reported consideration lands

One outlet has published a split of the price between founders and investors. Neither company has confirmed a figure of any kind.

Reported

The split comes from the New York Times, attributed to a single person with knowledge of the agreement. Against $164m of capital raised, $6.0bn to investors implies a strong return for a company founded in 2023. The mix between cash and stock is not disclosed, and that mix determines how much of the value is realised at close and how much rides on Stripe equity.

How the reported price moved between July and August

Four outlets published four figures over four weeks. The two companies published none.

ReportedNot disclosed

Reporting opened at roughly $10bn while the parties were in talks, settled above $7bn when Bloomberg reported the agreement, and arrived at a specific $7.5bn on the day of announcement. Stripe told TechCrunch it does not comment on rumours or speculation, and both companies declined to disclose the value. Every multiple on this page is built on the reported figure and inherits its uncertainty.

Acquiry calculationMultiple derivation and sensitivity

Only one of the three inputs to the headline multiple is a company figure. The price is press reporting and the revenue is a research estimate, so the multiple carries a band rather than a point value. The grid below runs both inputs across their reported range.

Price assumption	At $110m revenue	At $140m revenue	At $170m revenue	Basis
$7.0bn	63.6×	50.0×	41.2×	Bloomberg floor, 16 August
$7.5bn	68.2×	53.6×	44.1×	New York Times figure
$8.0bn	72.7×	57.1×	47.1×	Business Insider figure
$10.0bn	90.9×	71.4×	58.8×	Wall Street Journal, while in talks

Across the full grid the multiple runs from 41× to 91×. The range is wide because the inputs are wide, and no amount of arithmetic narrows it. What can be said is that on every combination in the table this is a growth multiple rather than an earnings multiple, and it is priced against a revenue line that has moved by an estimated factor of 28 in twelve months.

03 · The asset

What OpenRouter actually is, stripped of the gateway marketing

OpenRouter is one API endpoint that reaches more than four hundred models from more than eighty providers, and a routing engine that decides which one handles a given request. That sentence does more work than any product page. The single endpoint is the distribution mechanism: a developer integrates once and gains access to every model on the platform, including ones released after the integration was written. The routing engine is the retention mechanism: once an application is tuned to route on cost, latency and reliability rather than on a hardcoded model name, moving off the platform means rebuilding that logic.

The commercial signature follows from the structure. Adoption starts with a single developer and a test key. Expansion happens as applications move into production and volume grows. Removal is possible, because the API is deliberately OpenAI-compatible, but it means giving up provider failover, price arbitrage across four hundred models, and the accumulated knowledge of which model performs on which task. The asset is not the proxy. It is three years of demand data and eight million developers who already have a key.

Editorial illustration: a single luminous blue channel branching into many machined routes — One integration, four hundred destinations. The value is not in any single route but in the switching logic that sits above them, and in the demand data that accumulates every time a request is priced against an alternative.

The three-layer read

Access layer. One OpenAI-compatible endpoint, four hundred models, eighty providers, unified billing through a credit balance. Low integration cost. This is what gets it into the codebase.
Routing layer. Provider failover, price and latency optimisation, and quality-aware routing. Published model variants let a developer request the fastest provider, the cheapest provider, or the one tuned for tool-calling reliability. This is what makes it hard to replace.
Governance layer. Workspaces, spend management, guardrails and zero-data-retention policies. This is where enterprise contracts sit, and where the platform is youngest.

The development opportunity sits on the governance axis. OpenRouter scores highest in its category on breadth, adoption and neutrality. Enterprise controls, audit, observability and procurement-grade apparatus are where Portkey and Braintrust have invested most heavily, and where a 90-person company serving 10 million developers and companies has the clearest room to build. That headroom is a meaningful part of why a platform routing hundreds of trillions of tokens a month is estimated to earn roughly $140 million of net revenue.

Acquiry assessmentThe gap is the thesis

An estimated $2.5 billion of annualised inference spend crosses the platform, of which roughly $140 million is retained. Its take rate is a flat percentage applied at the point a customer loads credits, not a margin on the tokens themselves. The distribution is already won. The monetisation model is the open question.

Widening the take, or moving revenue from a transaction fee to a contracted enterprise relationship, calls for the governance and billing apparatus that Stripe has already built. That is the most direct reading of why these two businesses fit together.

04 · Unit economics

The single most important thing to understand about how OpenRouter earns money

Tokens carry no mark-up. The per-token price a customer pays on the platform matches the price the provider charges directly. This is published in the company's own pricing documentation and it is the foundation of the neutrality position: there is no financial reason for the router to prefer one model over another, because it earns the same regardless of which one wins the request.

Revenue comes from a platform fee charged when a customer loads credits. 5.5 per cent on fiat, with an eighty cent minimum. 5.0 per cent on cryptocurrency. 5.0 per cent on the list-price equivalent for customers bringing their own provider keys, after the first million requests a month. Those are published figures, not estimates.

The consequence matters for every multiple in this article. A reported $140 million of annualised revenue at a blended 5.5 per cent fee implies roughly $2.5 billion of annualised inference spend flowing across the platform. Gross routed spend and net revenue differ by a factor of about eighteen, and using the wrong one produces a multiple that is wrong by the same factor.

Gross flow against net revenue Public record Indicative Acquiry calculation

Measure	Approximate value	Basis
Annualised gross inference spend routed	$2.5bn	Acquiry calculation: $140m divided by a 5.5 per cent blended fee
Annualised net revenue	$140m	Sacra estimate, July 2026. Not a company figure
Reported gross margin	~70%	Secondary reporting. Inference cost is paid by the customer from their credit balance, not out of the fee
Implied gross profit	~$98m	Acquiry calculation
EV on net revenue	53.6×	Acquiry calculation at the reported $7.5bn
EV on gross profit	76.5×	Acquiry calculation
EV on gross routed spend	3.0×	Acquiry calculation. Included to show how far the framing moves the answer

The bottom three rows are the same transaction described three ways. Anyone quoting a multiple on this deal should say which line they are using. Three times gross flow sounds inexpensive; seventy-six times gross profit does not. Both are arithmetically correct.

Why routing has commercial value

Published list price per million input tokens across four widely used models.

Public record

The spread between the cheapest and the most expensive input token in this set is thirty times. That spread is the entire commercial case for a routing layer. Sending a simple classification task to a model priced at three dollars per million tokens when one priced at ten cents would do the job is the cost problem OpenRouter exists to solve, and it is the problem Stripe says it wants to solve alongside the revenue side.

05 · Buyer rationale

Why a payments company writes its largest cheque for an inference router

The buyer has an identifiable acquisition pattern: buy a capability that plugs into the existing money flow, keep the team, fold the product into the platform. Paystack brought African payments. TaxJar became Stripe Tax. Bridge brought stablecoin rails at a confirmed $1.1 billion. Metronome brought usage-based billing. Every one of those was bought to extend what Stripe already did for its customers.

This transaction fits the pattern on strategy and breaks it on scale. At a reported $7.5 billion it is roughly 6.8 times Bridge, which was the largest deal Stripe had done before it, and larger than every prior acquisition combined on any reasonable reading of the reported figures.

The four rationales, ranked by how well the evidence supports them

Strategic rationale, assessed Acquiry assessment

Rationale	What supports it	What to consider	Support
Owning both sides of the AI margin	Stripe sees what an AI product earns. OpenRouter sees what it costs to serve. Token Billing and Metronome already sit in the stack and need exactly the usage signal OpenRouter produces.	Nothing material. This is the strongest leg of the case and it is the one both chief executives articulated.	Strong
Distribution overlap	Stripe reports serving 78 per cent of the Forbes AI 50. Those are the same companies buying inference at scale. OpenRouter brings 10 million developers and companies.	Developer count is not revenue. The conversion from free and low-volume users to enterprise contracts is the work.	Strong
Take rate arbitrage	Stripe's blended payments take is roughly 36 basis points. OpenRouter charges 550. Acquiring a fee stream fifteen times richer than the core business is commercially rational.	The richer rate applies to a flow roughly seven hundred times smaller, and it is applied to a unit price that is falling.	Moderate
Agentic settlement	Shared Payment Tokens, the Agentic Commerce Protocol and streaming payments all point at machine-to-machine transactions. Combining the request layer with the payment layer is the long-dated case.	The market for autonomous agent settlement is still forming. This is optionality rather than a near-term revenue line.	Directional

The transaction is underwritten principally on the first two rationales, and they reinforce each other. Stripe is buying the cost side of a customer base it already serves on the revenue side. The remaining items are optionality layered on top rather than the basis for the price.

07 · Metrics

What the buyer is acquiring, quantified

What the buyer is actually acquiring

Platform reach and provider coverage, on published figures.

Public recordAcquiry calculation

Models available through one endpoint

400+ models

More than four hundred models from more than eighty providers, reachable through a single integration. Every one of those provider relationships had to be negotiated, maintained and kept current as models are released and repriced.

Developers on the platform

10m+ developers and companies at announcement

Reported as more than 10 million developers and companies at announcement. This is the asset that cannot be rebuilt quickly. The routing logic is reproducible; eight million developers who already have an API key are not.

Annualised revenue per employee

~$2.8m per employee per year

Acquiry calculation: approximately $140m of estimated revenue across roughly 50 staff. The bar is drawn against a $5m reference ceiling for visual scale. Very few software businesses at any stage operate at this ratio.

Stripe coverage of the Forbes AI 50

78% of the AI 50

Stripe reports serving 78 per cent of the Forbes AI 50 for payments. Those are the same companies that buy inference at scale, which is where the commercial overlap between the two businesses sits.

Two of these four bars describe the same underlying asset from different angles: distribution that took three years to build and would take a competitor about as long to replicate. The revenue per employee figure is the one that explains the price. A 90-person team supporting 10 million developers and companies is an unusual operating profile.

What is published, and what a fuller picture would add

Model counts, provider counts, developer counts and the fee schedule are published. Revenue, margin, net revenue retention and customer concentration are not. For a platform earning a percentage of customer spend, retention is the metric that shows whether growth comes from existing applications scaling up or from new developers arriving. A business at $140 million growing on expansion within existing accounts has a different profile to one growing on new sign-ups, and the two value differently.

The published token volume figures give a partial answer. Weekly volume went from 5 trillion tokens in November 2025 to 25 trillion in May 2026, a fivefold increase in six months. That is expansion within production applications rather than sign-up growth, because eight million developers do not arrive in a quarter and trial usage does not reach trillions of tokens.

The eight numbers that would settle the operating case

Net revenue retention. Enterprise share of revenue. Concentration in the top ten accounts. Blended effective take rate against the 5.5 per cent headline. BYOK volume as a share of total. Gross margin from the company rather than from research estimates. Contracted against pay-as-you-go mix. Revenue per token routed, over time.

None are in the public record. The last one is the number that matters most, because it is the single measure that shows whether volume growth is outrunning price decline.

08 · Precedents

How the implied multiple sits against AI transactions of the last three years

Private AI acquisitions have produced a reference set that behaves nothing like software M&A. Nvidia paid roughly $700 million for Run:ai on approximately $6.6 million of revenue, a multiple above one hundred times. Databricks paid roughly $1.3 billion for MosaicML on approximately $20 million, about sixty-five times. Databricks also paid roughly $2 billion for Tabular on approximately $1 million of annual recurring revenue, which is a multiple in the region of two thousand times and is excluded from the chart below because it renders every other bar invisible.

At a reported $7.5 billion on an estimated $140 million, this transaction clears at 53.6 times, which sits inside that range rather than outside it.

EV / revenue multiples, recent AI infrastructure transactions

Reported enterprise values against reported revenue at announcement. Only the ServiceNow price is confirmed by primary source.

IndicativeReported

Tabular, acquired by Databricks in 2024 at a reported $2bn on approximately $1m of annual recurring revenue, is excluded from the chart because a 2,000x multiple compresses every other bar to nothing. It belongs in the record and is discussed in the text. Two things follow from the chart. OpenRouter sits inside the range set by Nvidia and Databricks rather than outside it, and the revenue bases are not consistent: some are annual recurring revenue, some are estimated run rate, and only one price in the set is confirmed by a primary source. The set gives a range, not a valuation.

Reality filterThree reasons the comparison needs context

One. Only one price in the set is confirmed. ServiceNow disclosed $2.85 billion for Moveworks. Every other figure in the chart is press reporting, including this one. Comparing reported numbers to reported numbers is internally consistent, and it is not the same as comparing audited ones.

Two. The revenue bases are inconsistent. Some precedents are measured on annual recurring revenue, some on estimated run rate, some on a trailing period. OpenRouter is measured on a research firm's estimate of annualised net fee revenue. These are not the same denominator and the resulting multiples are not strictly comparable.

Three. Growth rates are not held constant. Run:ai at one hundred times and MosaicML at sixty-five times were both acquired for technology and team ahead of commercial scale. OpenRouter is estimated to have grown revenue by a factor of twenty-eight in twelve months. A multiple set against that trajectory is a different instrument from one set against a research asset.

The 2026 window

Capital moved into inference infrastructure at scale through 2026, which gives a second reference set. Fireworks raised $1.5 billion in July at a $17.5 billion valuation. Baseten raised $1.5 billion in June at $13 billion. Together AI raised $800 million at $8.3 billion. Groq raised $350 million at $3.5 billion.

Both Fireworks and Baseten carry marks above the reported OpenRouter price. The distinction is that those are private marks on primary capital and this is a price for the whole company, which is a harder number to set and a different kind of evidence.

Where the price sits among 2026 AI infrastructure marks

Reported valuations from funding rounds completed in the same year, against the reported consideration for OpenRouter.

Public recordReported

Capital has moved into inference infrastructure at scale through 2026. Fireworks and Baseten both carry marks above the reported OpenRouter price, and both raised more than a billion dollars in a single round. The distinction is that those are private marks on primary capital and this is a price paid for the whole company, which is a harder number to set.

09 · Comparables

Where the implied multiple sits against the gateway peer set

There is no listed comparable for this business. Every direct competitor is private and none publishes revenue, so the peer set has to be assembled from reported valuations and judgement about scale. The chart below plots implied multiple against estimated growth and places OpenRouter at the extreme of both axes.

Fee rate against growth, gateway and inference comparables

OpenRouter is plotted at its implied 53.6 times on estimated revenue. Every other position is Acquiry judgement.

IndicativeAcquiry calculation

Gateway and inference comparablesOpenRouter implied by this transactionIndicative peer band

Multiples and growth rates for every private comparable other than OpenRouter are indicative. Private inference and gateway businesses do not publish revenue, so these positions are Acquiry judgement against reported valuations and are shown to give shape to the peer set rather than to price it. The growth axis is what carries the multiple. OpenRouter is estimated to have moved from roughly $5m to roughly $140m of annualised revenue inside twelve months. No other position on the chart is growing at that rate, and a multiple set against that trajectory looks different from one set against a mature business.

Every position other than OpenRouter carries an indicative label. They are Acquiry judgement against reported valuations, not measured multiples, and they are shown to give the peer set shape rather than to price it. The relationship is the useful output.

What the chart shows is that the multiple and the growth rate move together. OpenRouter is estimated to have moved from roughly $5 million to roughly $140 million of annualised revenue in twelve months. No other position on the chart is growing at that rate, and a price set against that trajectory looks different from one set against a mature business, whatever the headline multiple says.

Acquiry assessmentWhat the multiple is paying for

Accumulated demand data across four hundred models, gathered over three years. Eight million developers with a working integration. More than eighty negotiated provider relationships. A position at the point where the model choice is actually made, which is upstream of both the provider and the application.

None of those appear on a revenue line, and all of them take time rather than capital to build. Acquiry's read is that the price is set against the position rather than against the current revenue run rate, which is the normal way strategic buyers value infrastructure early in a category.

11 · Reference points

Four numbers attached to the same company, on four different bases

OpenRouter valuation path, 2025 to 2026

Two priced rounds and two reported transaction figures inside sixteen months.

ReportedIndicative

Priced roundsReported figures

Three months separate a $1.3bn Series B from a reported $7.5bn agreement, an uplift of roughly 5.8 times. A venture round and an acquisition price are not the same instrument. A Series B prices a minority stake in preferred stock with protective terms attached; an acquisition prices the whole company including control. Some of the gap is the instrument, and the balance is what changed in the business over the quarter.

Plotted over time the pattern is compressed rather than gradual. Two priced rounds inside twelve months, a reported figure of roughly $10 billion while the parties were in talks, and a reported agreement at $7.5 billion four weeks later. There is no long history to draw on, because the company is three years old.

Four reference points on the same company

Each column states its own basis. They measure different things and are not a time series.

ReportedIndicative

Capital raised of $164m is shown as a dashed reference because it is an input, not a valuation. The reported consideration is roughly 46 times the capital that went into the business and roughly 54 times estimated annualised revenue. Both statements are true, and neither is a multiple that would be recognised in a mature software transaction. This is a growth-rate price, not an earnings price.

The dashed column is the informative one. Total primary capital raised of approximately $164 million sits against a reported consideration roughly forty-six times that figure. On the reported proceeds split, investors receive $6.0 billion against $164 million deployed across all rounds.

Read plainly: this is a company that raised comparatively little, grew very quickly, and sold early. It is the opposite shape to the 2021 growth-equity vintage, where large amounts of capital were raised against valuations that took years to grow into.

12 · Capital history

Four rounds, $164m, and one that dominates the total

OpenRouter primary capital raised, by round

Approximately $164m in total. The Series B alone accounts for more than two thirds of it.

ReportedIndicative

Total raised is reported at $164m across all rounds. The split across the pre-seed and seed is not published, so the two earliest bars are indicative and sum to the reported balance. The company raised $113m in May 2026 and agreed to sell three months later. A round of that size normally funds eighteen to twenty-four months of building. The compression between the round and the agreement is the most informative feature of the funding history.

OpenRouter priced rounds Reported Indicative

Round	Date	Raised	Post-money	Lead investors
Pre-seed and seed	2023 to 2024	~$15m	—	Angel syndicate, Andreessen Horowitz
Series A	May 2025	~$36m	—	Menlo Ventures
Series B	28 May 2026	$113m	$1.30bn	CapitalG
Total		~$164m	—	PitchBook, cited by the New York Times

The Series B is the round that matters, and its participant list is more informative than its size. Alongside CapitalG it drew NVentures, the venture arm of NVIDIA, plus ServiceNow Ventures, MongoDB Ventures, Snowflake Ventures and Databricks Ventures, with Andreessen Horowitz and Menlo Ventures following on. Four enterprise data platforms and a chip manufacturer put money into a model router three months before it agreed to sell.

The shape of the table also explains the founder proceeds. Two thirds of all capital raised came in a single round three months before the agreement, which means the preference stack is shallow and recent. Reported founder proceeds of $1.5 billion, or twenty per cent of the total, are consistent with a company that raised late relative to its growth.

Acquiry assessmentWhat the timing indicates about the process

Eighty-three days between a $113 million round and an agreement to sell is unusual. A round of that size normally funds eighteen to twenty-four months of building, and a board that has just priced the company at $1.3 billion does not typically run a sale process the following quarter.

The Wall Street Journal reported talks on 23 July, fifty-six days after the round closed. The sequence is more consistent with an approach than with a process, and an approach at roughly six times the round price is one a board is obliged to consider seriously.

14 · Where the value goes

Where the reported $7.5bn lands, to the limit of the public record

Where the tokens actually go

Share of weekly token volume by model provider, June 2026, from OpenRouter’s published rankings.

Public record

Chinese-origin providers account for 46.4 per cent of weekly volume against 35.7 per cent for United States providers. The single largest provider holds under eighteen per cent, which is the strongest evidence that the routing layer is genuinely multi-model rather than a front end for one lab. That distribution is also what makes the neutrality question commercially material rather than philosophical.

What the reported price buys, and what is not knowable

Proportions are stated against reported or published figures only.

ReportedAcquiry calculationNot disclosed

Reported investor proceeds against capital raised

$6.0bn of $164m

On the reported split, investors receive approximately 37 times the primary capital that went into the business across all rounds. The distribution across rounds depends on preference terms that are not public.

Reported founder proceeds

$1.5bn of $7.5bn reported total

Twenty per cent of the reported consideration to three founders and, on the reporting, no other party. Founder ownership at that level after two priced rounds is high, and consistent with a business that raised late relative to its growth.

Estimated annualised revenue against reported price

$140m of $7.5bn

Revenue equals roughly 1.9 per cent of the reported price, which is the same statement as a 53.6 times multiple expressed the other way round. Both figures rest on a third-party revenue estimate.

Consideration mix, retention terms and close date

Not disclosed

Cash versus stock, escrow, earn-out, retention pools and the closing timetable are all outside the public record. Without them the reported headline cannot be converted into value actually received at close.

Acquiry does not publish a cap table for this transaction. Ownership percentages, preference terms and option pool treatment are not public, and the proceeds split itself rests on one unnamed source. This figure states only what the reporting supports.

The reported split is $1.5 billion to the three founders and $6.0 billion to investors. Against approximately $164 million of primary capital across all rounds, that implies roughly thirty-seven times on invested capital at the aggregate level. How it distributes across individual rounds depends on preference terms that are not public, and the Series B at $113 million represents most of the money at risk.

Acquiry publishes an equity waterfall when the preference terms are documented. For this transaction they are not, and the proceeds split itself rests on a single unnamed source in one outlet. The figure above states what the reporting supports and nothing beyond it.

15 · Market context

Why this transaction is happening now

Three conditions converged in 2026 to make this transaction possible, and none of them are about OpenRouter specifically.

The first is that inference became the cost line that matters. Gartner expects inference to account for sixty-five per cent of AI compute by 2029. Goldman Sachs projects token consumption growing twenty-four times between 2026 and 2030. For any company building on models, the question moved from whether the technology works to what it costs to run at volume, and that is a procurement question rather than a research one.

The second is that token prices fell faster than anyone budgeted for. Inference cost declined roughly tenfold annually between 2021 and 2025 on published estimates, and is projected to fall three to five times a year through 2027. A thirty-fold spread now separates the cheapest and most expensive input token among widely used models. When the price difference between two adequate options is that large, choosing between them becomes a commercial function rather than an engineering preference.

The third is that payments companies started competing on AI infrastructure. Stripe shipped Token Billing in 2025, co-developed the Agentic Commerce Protocol with OpenAI in September 2025, and launched streaming stablecoin micropayments at Sessions in 2026. Two hundred and eighty-eight product launches in a single April. The company had already decided that AI transaction infrastructure was its next market before this agreement.

Editorial illustration: dark columns of varying height with three edge-lit in blue and a routing path threading between them — Three conditions arrived together: inference became the dominant compute cost, token prices fell far enough to make model choice a commercial decision, and payments infrastructure moved into the AI stack. This transaction sits where they meet.

Token volume against revenue earned per token

Monthly volume rising while the fee earned on each token falls. Index rebased to 100 in November 2025.

Public recordIndicative

Monthly token volumeRevenue per token, indexed

Monthly token volume against an index of revenue earned per token, rebased to 100 in November 2025. Volume points for November 2025 and May 2026 derive from OpenRouter's published weekly figures. The intermediate points and the monetisation index are third-party estimates. This is the central economic question in the transaction. A flat percentage fee on a falling unit price produces less revenue per unit, and only volume growth closes the gap. Volume has grown faster than price has fallen for the last four quarters. Whether that continues determines whether the reported multiple looks conservative or full in three years.

The buyer's market context

Total payment volume reached $1.9 trillion in 2025, up 34 per cent, against a $159 billion valuation in a February 2026 tender. The Information reported approximately $6.8 billion of net revenue and approximately $3.2 billion of free cash flow for the year, which implies a margin near forty-seven per cent. The company describes itself as robustly profitable in its annual letter.

A reported $7.5 billion is roughly 4.7 per cent of the current valuation and slightly more than two years of free cash flow. John Collison told CNBC in February 2026 that an IPO is not among the company's top twenty priorities, so the funding for a transaction this size comes from the balance sheet and the stock rather than from public markets.

16 · Competitive landscape

Where OpenRouter leads, and where the headroom sits

The field divides into three groups. Dedicated gateways such as Portkey, Martian, Requesty and the open-source LiteLLM. Observability and evaluation platforms that added routing, principally Braintrust and Helicone. And platform gateways shipped by companies with existing developer distribution: Cloudflare AI Gateway, Vercel AI Gateway, and Google Cloud's API Gateway, which now accepts OpenAI-compatible requests and routes them to Gemini or Claude.

Capability matrix, AI gateway and routing platforms

Acquiry assessment on a four-point scale. This is judgement, not measurement, and is presented as such.

Indicative

	Model breadth	Provider breadth	Intelligent routing	Enterprise controls	Observability	Developer adoption	Billing integration	Neutrality position
OpenRouter
Cloudflare
Vercel
Portkey
Braintrust
LiteLLM
Martian
Helicone
Kong
Google Cloud

NoneLimitedCompetitiveLeading

OpenRouter scores highest on breadth, adoption and neutrality, and has the most headroom on enterprise controls and observability, which is where Portkey and Braintrust are strongest. The column that changes with this transaction is billing integration, where the combination with Stripe moves it beyond anything else in the set. Neutrality is the column to watch, since it is the one attribute ownership can affect.

Read the OpenRouter row and the pattern is clear. It leads on model breadth, provider breadth, developer adoption and neutrality position, and it is competitive on routing intelligence. Enterprise controls and observability are the axes with the most room to develop, and they are the axes that support the move from developer credit card to enterprise contract.

The column that changes with this transaction is billing integration, where the combination moves beyond anything else in the set. Stripe already runs the invoicing, tax, revenue recognition and usage-based billing apparatus that a gateway needs in order to sell to a procurement department. Governance investment shows in the revenue line more slowly than product integration does, so the sequencing after close is the thing to watch.

Acquiry assessmentOn the reproducibility argument

The most common technical objection is that a model router is a reverse proxy and could be rebuilt quickly by a competent team. The architecture argument is largely correct and the commercial conclusion does not follow from it.

What is reproducible is the proxy. What is not reproducible on the same timescale is eighty-plus negotiated provider relationships, three years of demand data showing which model actually performs on which task, published rankings that developers use as a reference, and eight million existing integrations. The engineering is the cheap part of the asset and the distribution is the expensive part, which is the same pattern that has held in payments infrastructure for two decades.

17 · Value levers

Five ways the buyer creates value, rated on confidence of delivery

The following assessment separates what the buyer has demonstrated it can execute from what it would need to build. Confidence ratings are Acquiry judgement, not disclosed guidance, and the horizons are estimates.

Value levers and Acquiry confidence

Five levers available to the buyer, rated on confidence of delivery and horizon.

Indicative

Internalised payment processing

Very high confidence

0-6 months

OpenRouter processed its credit purchases through Stripe. That fee now stays inside the group. Mechanically certain, modest in absolute quantum.

Billing and metering attach

High confidence

6-18 months

Token Billing and Metronome already sit in the Stripe stack. OpenRouter supplies the usage signal those products need, and Stripe supplies the invoice. The two halves were built separately and fit.

Enterprise governance upsell

Moderate confidence

12-24 months

Workspaces, guardrails and zero-data-retention are built. Selling them into Stripe's enterprise base is a distribution question rather than a product one, and enterprise cycles are long.

Agentic settlement

Moderate confidence

18-36 months

Shared Payment Tokens, the Agentic Commerce Protocol and streaming payments point at machine-to-machine settlement. Combining the request layer with the payment layer is the strategic case, and it depends on a market that is still forming.

Cross-sell into the AI 50

High confidence

6-18 months

Stripe reports serving 78 per cent of the Forbes AI 50. Those companies buy inference. The overlap between the two customer bases is the most direct commercial synergy in the transaction.

The highest-confidence lever is also the smallest: OpenRouter paid Stripe to process its credit purchases, and that fee now stays in the group. The largest lever is the one furthest out. Combining the request layer with the payment layer to settle machine-to-machine transactions is the strategic case for the price, and it depends on a market that is still being built.

Acquiry assessmentWhat to measure, and when

The highest-confidence lever is the smallest. OpenRouter processed its own credit purchases through Stripe, so that fee now stays inside the group, and the saving is mechanical from day one. The largest lever is the furthest out, because agentic settlement depends on a market that is still forming.

The practical implication is about measurement rather than outcome. The informative metric is enterprise contracted revenue in the eighth quarter after close, not the payment-processing saving in the second. The first tells you whether the governance and billing thesis worked. The second was always going to happen.

18 · Risk register

Eight items to monitor, scored on materiality and likelihood

Materiality and likelihood scores are Acquiry judgement on a five-point scale. They are not disclosed figures and should be read as a structured opinion rather than a measurement.

Risk register, materiality against likelihood

Eight items scored on a five-point scale. Position is Acquiry judgement. Hover any point for the full assessment and what would resolve it.

Indicative

Materiality 5Materiality 4Materiality 3 or below

Two items score five on materiality. The first is that no price is confirmed by either company, which is a disclosure position rather than a commercial one and resolves the moment either party states a figure. The second is fee compression, and that one resolves only with several quarters of data on revenue earned per token routed. Everything else on the map is ordinary transaction risk.

Risk register, detail Acquiry assessment

#	Item	Assessment	Mat	Like
1	Price is not confirmed by either company DISCLOSURE	Every price in circulation is press reporting. The $7.5bn figure rests on one person with knowledge of the agreement, quoted by one outlet, and the companies declined to disclose a value. Earlier reporting ranged from $7bn to $10bn. Every multiple in this article inherits that range.	5	5
2	Fee compression as token prices fall REVENUE MODEL	Revenue is a flat percentage of what customers spend on inference, and inference prices are falling three to five times a year on published projections. Monetisation per token declined roughly 60 per cent between March and August 2026 on third-party estimates, while volume grew fast enough to more than offset it. How long that offset holds is the central economic question.	5	4
3	Revenue and margin are third-party estimates REVENUE QUALITY	The $140m annualised figure and the roughly 70 per cent gross margin come from research firms. Net revenue retention, enterprise share and customer concentration are not in the public record. The published token volumes are the only company-sourced scale data available.	4	5
4	Routing is becoming a platform feature COMPETITIVE	Cloudflare, Vercel and Google Cloud ship routing. Cursor, Databricks, Ramp and Kong have built it into their own products. The technical function is reproducible. The provider relationships, demand data and developer habit are the parts that take time rather than capital.	4	4
5	Neutrality perception under corporate ownership COMMERCIAL	OpenRouter built its position on being unaligned with any model provider, and does not mark up tokens. Both chief executives addressed neutrality directly in the announcement. Developer confidence in that commitment is now something the combined company demonstrates rather than asserts.	4	3
6	Concentration in open-weight model volume REVENUE QUALITY	Chinese-origin open-weight models moved from roughly 2 per cent of platform usage in mid-2025 to 46.4 per cent by June 2026. Cheaper tokens carry the same percentage fee on a smaller base, which changes the revenue mix even as volume rises. This is visible in OpenRouter's own published rankings.	3	4
7	Multi-jurisdiction clearance timing EXECUTION	At the reported size the transaction sits well above the 2026 US premerger notification threshold of $133.9m. Stripe operates from San Francisco and Dublin and OpenRouter serves a global developer base, so more than one filing is likely. Neither company has named the jurisdictions or the expected timetable.	3	3
8	Retention of a 90-person team HUMAN CAPITAL	OpenRouter describes itself as a 90-person startup operating a platform processing hundreds of trillions of tokens a month. Retention arrangements are not disclosed. Reported founder proceeds of $1.5bn are large enough that continuity is worth confirming rather than assuming. Stripe kept the Bridge team and products running after that acquisition, which is the relevant precedent.	3	3

The two highest-scoring items sit at five on materiality, and they resolve on very different timescales. The disclosure question resolves the moment either company states a figure, or on the first regulatory filing that makes one public. The fee compression question resolves only across four to six quarters of data on revenue earned per token routed, and that data is not currently published by anyone.

20 · Integration

Fifty people joining a company of nine thousand

Integration risk in this transaction runs opposite to the usual direction. OpenRouter has roughly fifty employees against Stripe's approximately nine thousand. There is no cost synergy to extract, no duplicated function to rationalise, and no overlapping product to sunset. The risk is not that integration is difficult. It is that integration is unnecessary and might happen anyway.

Editorial illustration: a small precisely machined component descending into a much larger machined housing — A 90-person platform joining a nine-thousand-person company. The engineering fit is straightforward. The question is how much of the acquired product's independence survives contact with a larger organisation's roadmap.

What the precedent suggests

Bridge is the closest comparison. Roughly sixty staff, $1.1 billion confirmed, closed February 2025. Stripe kept the team and continued running the products. That is the pattern this transaction most resembles in shape, if not in scale.
TaxJar was absorbed. Roughly two hundred staff, acquired April 2021, and the product became Stripe Tax. A capability folded into the platform rather than a brand kept alive.
Okay disappeared entirely. Acquired May 2023 and folded into internal engineering. The smallest acquisitions have consistently been absorbed rather than maintained.
Neither company has said which model applies here. Whether OpenRouter continues as a standalone brand is not addressed in the announcement, and it is one of the more commercially significant unanswered questions for existing customers.

The technical integration is genuinely simple. OpenRouter already used Stripe to process credit purchases, so the payment path is built. Stripe's Token Billing product needs exactly the usage signal OpenRouter generates. Metronome, acquired earlier in 2026, handles the usage-based invoicing. The three products were built separately and fit together without modification, which is unusual and is a reasonable part of the strategic case.

21 · Sequence

How the transaction got here, and what remains

Transaction sequence

Announced. Filing jurisdictions, conditions and close date are not stated.

ReportedPublic recordNot disclosed

Eighty-three days separate the Series B from the announcement. The Wall Street Journal reported talks on 23 July, fifty-six days after the round closed. A company that has just raised $113m at a $1.3bn valuation is not usually in sale discussions two months later, which suggests the approach came to them rather than the other way round.

The compression between 28 May and 19 August is the most informative element of the sequence. Eighty-three days from a $113 million Series B at a reported $1.3 billion to an announced agreement at a reported $7.5 billion. The Wall Street Journal reported talks on 23 July, fifty-six days after the round closed.

The inference is that Stripe approached OpenRouter rather than the company running a process. That is an inference, clearly labelled, not a reported fact. A board that has just priced its company at $1.3 billion with fresh capital in the bank does not usually seek a buyer eight weeks later, and an approach at roughly six times the round price is one the same board is obliged to take seriously.

What remains is entirely unstated. No filing jurisdictions or regulatory clearances are named. OpenRouter says the transaction remains subject to customary closing conditions and expects close in the coming weeks. At the reported size the transaction sits well above the 2026 US premerger notification threshold of $133.9 million, so a Hart-Scott-Rodino filing is required. Beyond that, the timetable is not in the public record.

22 · Consequences

Who benefits, and who now has a harder conversation

Beneficiaries

OpenRouter shareholders

On the reported split, approximately $6.0bn to investors against roughly $164m of primary capital. A 2023 company reaching this outcome in three years is an exceptional result on any measure.

The three founders

Reported at $1.5bn between them, twenty per cent of the total. Founder ownership at that level after two priced rounds reflects a company that raised late relative to its growth.

CapitalG and the May syndicate

A reported $1.3bn entry in May against a reported $7.5bn agreement in August. NVentures, ServiceNow, MongoDB, Snowflake and Databricks Ventures all participated in that round.

Stripe's AI customer base

Companies already using Stripe for payments gain a route to unified cost and revenue reporting on their AI products. Stripe reports serving 78 per cent of the Forbes AI 50.

Other AI infrastructure founders

A validating reference point for the category. A three-year-old company with a 90-person team reaching this outcome resets what boards consider achievable on this timeline.

Facing a harder conversation

Independent gateway competitors

Portkey, Martian, Requesty and Helicone now compete with a category leader carrying Stripe's balance sheet and enterprise distribution behind it.

Platform gateway teams

Cloudflare and Vercel ship routing as a feature. A dedicated competitor with billing integration and eight million developers changes what feature parity means in this category.

Developers valuing strict neutrality

Both chief executives addressed neutrality directly in the announcement. It is now a commitment the combined company demonstrates rather than a structural property of an independent company.

Model providers with premium pricing

A routing layer with better cost visibility, owned by a company that specialises in optimising unit economics, makes price differentials more visible to buyers than they were.

Late-stage investors pricing this category

A reported 53.6 times net revenue is now a visible reference point. Boards holding inference infrastructure positions have a comparable they will be asked about either way.

24 · Acquiry verdict

Strategically coherent, priced on growth, and reported rather than confirmed

Acquiry verdict · 21 August 2026

A clear strategic fit at a growth-stage multiple. The price is press reporting, and the economics turn on one variable.

On strategic logic, this is among the cleaner fits of the year. Stripe optimises the revenue side of its customers' businesses. OpenRouter optimises the cost side. Token Billing and Metronome already sit in the Stripe stack and need exactly the usage signal OpenRouter produces. The three products were built separately and fit together without modification, which is unusual and is the strongest part of the case.

On price, the multiple is set against trajectory rather than run rate. A reported $7.5 billion against an estimated $140 million is 53.6 times net revenue, or 76.5 times gross profit. Those are growth multiples. They sit inside the range set by Run:ai at roughly one hundred times and MosaicML at sixty-five, and they are underwritten by an estimated twenty-eight-fold revenue increase in twelve months rather than by current earnings.

On the economics, one variable carries the outcome. Revenue is a flat percentage of customer inference spend, and the unit price of inference is falling three to five times a year. Volume has outrun price so far, comfortably. Whether it continues to is the question that determines whether this multiple looks conservative or full in three years, and it is measurable only with data neither company currently publishes.

On disclosure, everything material about the terms is press reporting. No price, mix, close date, condition or jurisdiction is confirmed by either party. The $7.5 billion figure rests on a single unnamed source in one outlet. That is normal for private-to-private AI transactions, and it means every multiple here carries a range rather than a point.

Net. The strategic case is strong and internally consistent. The price is high on every current measure and defensible on trajectory. The most useful thing anyone can do with this transaction is watch revenue per token routed, because that single series answers the only question that matters.

Strategic logic

Strong

Price against run rate

Growth multiple

Integration risk

Low

Disclosure to date

Reported only

Editorial illustration: two aligned machined slabs separated by a luminous blue gap with an unclamped seal above — Announced and aligned, but not closed and not priced on the record. The seal is in place unclamped, which is roughly where this transaction stands until either company states a figure.

25 · What to watch

Eight resolvable questions, and where each one gets answered

Investor and operator watch list Acquiry assessment

#	Question	Where it resolves	Timing
1	Does revenue per token routed stabilise?	Third-party estimates, or company disclosure if given. The single most informative series in the transaction	4-6 quarters
2	Does either company confirm a price?	A regulatory filing, a subsequent funding round disclosure, or a voluntary statement	Unknown. Possibly never
3	Does the 5.5 per cent platform fee change?	OpenRouter's published pricing page. Directly observable	Any time post-close
4	Does OpenRouter remain a standalone brand?	Product announcements after close. Bridge was kept, TaxJar became Stripe Tax, Okay disappeared	At or shortly after close
5	Do the three founders stay?	Announcement of post-close leadership, or absence of one. Reported proceeds make this worth confirming	At or before close
6	Does model provider mix keep shifting to open weights?	OpenRouter's own published weekly rankings. Directly observable and updated continuously	Continuous
7	Which jurisdictions review the transaction?	Filing announcements or clearance notices. HSR is required at this size	Weeks to months
8	Does Stripe acquire again at this scale?	Subsequent announcements. This transaction is the reference case for whether the pattern has changed permanently	12-24 months

Questions three and six resolve through observation rather than disclosure, because OpenRouter publishes both its pricing and its model rankings. Question one is the one that matters most and it has no reliable public source. Anyone underwriting this category should build their own token-level monetisation series, because the published data will not do it for them.

26 · Sources

Source ledger

Twelve sources underpin every figure on this page.

01Stripe newsroom announcement, 19 August 2026Primary source for the agreement, model and provider counts, named customers, and both chief executive statements. Contains no pricePrimary

02New York Times, 19 August 2026Sole source for the $7.5bn figure and the $1.5bn founder / $6.0bn investor proceeds split. Attributed to a person with knowledge of the agreement. States that the companies declined to disclose the valueReported

03Bloomberg, 16 August 2026, and Wall Street Journal, 23 July 2026Bloomberg reported a figure above $7bn. The Journal reported talks at approximately $10bn. Business Insider reported approximately $8bnReported

04OpenRouter published pricing documentationPrimary source for the 5.5 per cent fiat platform fee, 5.0 per cent on crypto and BYOK overage, the $0.80 minimum, and the statement that per-token cost matches provider list pricePrimary

05OpenRouter published rankings and platform statisticsWeekly token share by provider for June 2026, weekly volume of 5 trillion tokens in November 2025 and 25 trillion in May 2026, and the 8 million developer figurePrimary

06Sacra research, July and August 2026Source for annualised revenue of approximately $140m, the earlier $50m and $5m estimates, 29 per cent month-on-month growth, the roughly 60 per cent decline in monetisation per token, and the estimated monthly token volumeEstimate

07OpenRouter Series B announcement, 28 May 2026Round size of $113m, CapitalG as lead, and the participant list including NVentures, ServiceNow Ventures, MongoDB Ventures, Snowflake Ventures and Databricks Ventures. The $1.3bn valuation is reported rather than confirmedPrimary

08Stripe annual letter and company disclosuresTotal payment volume of $1.9tn for 2025, 34 per cent growth, the $159bn February 2026 tender valuation, and the 78 per cent Forbes AI 50 figurePrimary

09The Information, 2026Stripe net revenue of approximately $6.8bn and free cash flow of approximately $3.2bn for 2025. Stripe is private and does not publish audited statementsReported

10Precedent transaction set, 2023 to 2026Reported enterprise values and revenue for Run:ai, MosaicML, Tabular, Wiz, Windsurf and Moveworks. Only the ServiceNow price for Moveworks is confirmed by primary sourcePublic

11Stripe acquisition historyIndex 2018, Paystack 2020, TaxJar and Recko 2021, Okay 2023, Lemon Squeezy 2024, Bridge 2025 at a confirmed $1.1bn, Metronome 2026. Four of the nine have no disclosed pricePublic

12Acquiry Deal Intelligence, internal analysisAll multiple derivations, the implied routed spend calculation, revenue per employee, the sensitivity grid, risk materiality and likelihood scores, lever confidence ratings, the capability matrix, and every indicative peer positionAcquiry

Deal snapshot21 Aug 26

AcquirerStripe, Inc.

TargetOpenRouter, Inc.

Reported price$7.5bn

Confirmed priceNone

Est. revenue~$140m

Implied EV / revenue53.6×

Platform fee5.5%

Developers + companies10m+

Announced19 Aug 2026

Expected closeComing weeks*

StatusAnnounced

The acquirerAs at 19 Aug 26

STRIPEPrivate · San Francisco and Dublin

$159bnFeb 2026 tender

2023 $50BN · 2025 $91.5BN · 2026 $159BN

TPV 2025$1.9tn

TPV growth34%

Free cash flow~$3.2bn

OpenRouterPrivate

Valuation pathCalculated

May 2026 Series B$1.3bn

Reported price$7.5bn

Uplift5.8×

Elapsed83 days

Capital raised~$164m

Price vs capital~46×

EV / gross profit76.5×

SequenceClose not stated

Nov 2025

Weekly volume at 5 trillion tokens

28 May 2026

Series B, $113m led by CapitalG

23 Jul 2026

Wall Street Journal reports talks

16 Aug 2026

Bloomberg reports a deal above $7bn

19 Aug 2026

Stripe announces the agreement

Not stated

Regulatory clearances and close

Buyer track recordPublic record

Metronome2026 · ~$1bn

Bridge2025 · $1.1bn

Lemon Squeezy2024

Okay2023

TaxJar2021 · ~$850m

Recko2021 · ~$150m

Paystack2020 · ~$200m

Index2018

Considering an AI infrastructure acquisition or exit?

Acquiry executes buy-side and sell-side mandates in software and AI infrastructure from USD $1m to $500m. Valuation benchmarking, target origination, and transaction execution.

Start a conversation Valuation benchmarking

Open questions8 to resolve

01Does revenue per token routed stabilise?

02Does either company confirm a price?

03Does the 5.5 per cent fee change?

04Does OpenRouter stay a standalone brand?

05Do the three founders remain?

06Does the open-weight mix keep shifting?

07Which jurisdictions review the deal?

08Does Stripe acquire again at this scale?

Related deal intelligenceAcquiry

AI infrastructurePricing the inference layer in 2026Where gateway, routing and serving businesses are clearing

ValuationSector multiples database, Q3 2026EV/ARR and EV/EBITDA ranges by software vertical

MethodReading a deal announcement for what is missingA disclosure checklist for buy-side diligence

Section	Sources
What happened	01 Stripe announcement, 02 New York Times
Terms	01 Stripe announcement, 02 NYT, 03 Bloomberg and WSJ, 12 Acquiry calculation
The asset	01 Stripe announcement, 04 OpenRouter pricing, 05 OpenRouter statistics
Unit economics	04 OpenRouter pricing, 06 Sacra, 12 Acquiry calculation
Buyer rationale	01 Stripe announcement, 08 Stripe disclosures, 11 acquisition history
Distribution	01 Stripe announcement, 05 OpenRouter statistics, 12 Acquiry calculation
Metrics	05 OpenRouter statistics, 06 Sacra, 12 Acquiry assessment
Precedents	10 precedent set 2023-26
Comparables	10 precedent set, 12 Acquiry assessment
Valuation path	02 NYT, 07 Series B announcement, 12 Acquiry calculation
Reference points	02 NYT, 06 Sacra, 07 Series B announcement
Capital history	07 Series B announcement, 02 NYT via PitchBook
Model mix	05 OpenRouter rankings, 06 Sacra
Where the value goes	02 NYT, 12 Acquiry calculation
Market context	08 Stripe disclosures, 09 The Information, 10 market projections
Competitive	12 Acquiry assessment
Value levers	01 Stripe announcement, 12 Acquiry assessment
Risk register	12 Acquiry assessment
The deflation question	04 OpenRouter pricing, 06 Sacra
Integration	11 acquisition history, 12 Acquiry assessment
Timeline	01 Stripe announcement, 03 Bloomberg and WSJ, 07 Series B
Consequences	02 NYT, 12 Acquiry assessment
The acquirer	11 acquisition history
Verdict	12 Acquiry assessment
What to watch	12 Acquiry assessment
Sources	Ledger, 12 entries

Field	Value
Acquirer	Stripe, Inc. · San Francisco and Dublin · private
Target	OpenRouter, Inc. · New York, USA · private, founded 2023
Transaction type	Acquisition, agreement announced
Reported consideration	USD ~7,500,000,000 · reported, not confirmed
Confirmed consideration	Not disclosed by either party
Consideration mix	Not disclosed
Announced	19 August 2026
Expected close	Coming weeks, subject to customary closing conditions
Status	Announced, price undisclosed
Sector	AI infrastructure, model gateway and inference routing

Stripe Acquires OpenRouter and Moves Into the Layer That Decides Which Model Gets the Request

Stripe buys the layer that decides which model gets the request

What is in the announcement, and what is not

Terms as reported

What OpenRouter actually is, stripped of the gateway marketing

The three-layer read

The single most important thing to understand about how OpenRouter earns money

Why a payments company writes its largest cheque for an inference router

The four rationales, ranked by how well the evidence supports them

Eight million developers, four hundred models, fifty employees

What the buyer is acquiring, quantified

What is published, and what a fuller picture would add

How the implied multiple sits against AI transactions of the last three years

The 2026 window

Where the implied multiple sits against the gateway peer set

From $1.3bn to a reported $7.5bn in eighty-three days

Four numbers attached to the same company, on four different bases

Four rounds, $164m, and one that dominates the total

Chinese open-weight models now carry more traffic than American ones

Where the reported $7.5bn lands, to the limit of the public record

Why this transaction is happening now

Where OpenRouter leads, and where the headroom sits

Five ways the buyer creates value, rated on confidence of delivery

Eight items to monitor, scored on materiality and likelihood

A flat percentage fee on a unit price that keeps falling

Fifty people joining a company of nine thousand

What the precedent suggests

How the transaction got here, and what remains

Who benefits, and who now has a harder conversation

Eight years of small, disciplined acquisitions, then one that is six times larger

Strategically coherent, priced on growth, and reported rather than confirmed

Eight resolvable questions, and where each one gets answered

Source ledger

References, citation and open questions

APA, 7th edition

Chicago, author-date

Harvard

BibTeX

How much is Stripe paying for OpenRouter?

Why are several different prices circulating?

Is the deal cash or stock?

How is the reported price split between founders and investors?

When is it expected to close?

What is the strategic rationale?

Who advised on the transaction?

Does OpenRouter continue as a standalone product?

What multiple does the deal imply?

Which multiple should be used for comparison?

Is 53.6 times high for a software business?

What justifies a multiple at that level?

How does this compare with the May 2026 valuation?

How does the price compare with capital raised?

What is the revenue estimate based on?

What does OpenRouter actually do?

How does OpenRouter make money?

How large is OpenRouter?

Who founded OpenRouter?

How large is Stripe?

Is this Stripe's largest acquisition?

What is the main economic question in the transaction?

How defensible is the position?

Does neutrality survive corporate ownership?

Are there regulatory considerations?

Does this set a benchmark for other AI infrastructure companies?

Editorial independence

Corrections and updates

Not investment advice

Positions

Contact

Related Acquiry research

Acquiring or selling a software business between $1m and $500m?