Venice: The Unicorn That Refuses to Remember You
Erik Voorhees built a profitable, private AI platform to 3.5 million users with his own money.
Erik Voorhees built a profitable, private AI platform to 3.5 million users with his own money. Then crypto's smartest investors paid $1 billion to get in. Here's the bull case, the bear case, and the one number that decides it
Company Overview
Start with the thing that makes Venice unusual: it was profitable before it raised a dollar.
On July 1, Venice closed a $65 million Series A at a $1 billion post-money equity valuation. It was the company’s first outside capital, roughly two years after launch. Founder and CEO Erik Voorhees put up the seed money himself and ran the company to profitability, which it reached in Q1 2026, before letting anyone else in.
In plain English, here’s what Venice does. It’s a private AI app. Instead of building its own model, it gives you one interface to more than 200 AI models across text, image, video, and audio, most of them open-source. The pitch has two parts. First, it doesn’t log or store your prompts on its servers; your conversations stay on your device. Second, it strips out many of the content filters that competing tools apply. Private, and unrestricted.
The traction is real, and it’s company-reported: over 3.5 million registered users, roughly 1.3 trillion tokens processed per month, more than 1.7 million daily API calls, and annualized run-rate revenue above $70 million. Profitable at that scale, before the raise.
The round was led by Dragonfly, with Coinbase Ventures, North Island Ventures, F-Prime, Archetype, Morgan Creek, and Liquid2 Ventures joining. Look closely at that list. It’s almost entirely crypto-native capital. Not Sequoia, not Andreessen’s AI fund, not the usual names writing $65 million checks into AI. That’s not an accident, and we’ll come back to it.
The founder explains it. Voorhees is one of crypto’s earliest builders. He launched Satoshi Dice in 2012 and later founded ShapeShift, the non-custodial exchange built so users could trade without handing over their identity or their coins. He also settled with the SEC in 2014 over unregistered securities offerings tied to those early ventures, which matters here because it means the man structuring Venice’s capital stack has personally learned where the regulatory tripwires are. President and CTO Jesse Proudman, based in Seattle, previously built and sold a cloud company to IBM. This is not a first-time team.
The Market
Bloomberg Intelligence projects the generative AI market reaches roughly $2.3 trillion by 2032, and it flagged a specific shift: inference, the cost of actually running models, is overtaking training as the dominant revenue driver. Venice is a pure bet on that shift. It doesn’t spend to train frontier models. It spends to serve them.
The gap Venice attacks is the one the giants can’t comfortably close. OpenAI, Google, and the rest log conversations, train on user data, and wrap their products in heavy content moderation. That’s not a bug in their model, it’s central to it: the data improves the product, and the moderation protects the brand and the enterprise contracts. A meaningful minority of users wants the opposite, and history says that minority is durable. Signal, ProtonMail, and DuckDuckGo all proved that a low but persistent single-digit share of a huge market will choose privacy, and that some of them will pay for it.
The unit economics follow from the strategy. Because Venice rents and aggregates open models rather than building its own, its cost base is inference plus interface, not multibillion-dollar research. That’s how a 3.5-million-user platform gets to profitability on a self-funded budget. The new capital goes toward building proprietary data centers to cut its reliance on leased GPUs, which the company says improves margins and locks in capacity before the next compute crunch.
Timing is the last piece. Open-source models are finally good enough that you no longer need your own frontier lab to ship a credible product. Privacy anxiety is rising as AI moves into how people think, not just what they search. And the agent economy is arriving: Voorhees frames Venice’s target as a few hundred million people and, eventually, several billion AI agents, all needing access to intelligence that isn’t logged or gated. Whether that agent number materializes is a thesis, not a fact, but the direction is real.
Business Model and Moat
Venice makes money like a consumer SaaS company: a free tier to start, paid subscriptions above it, and API revenue from developers. Notably, only about 8% of users pay with cryptocurrency, so despite the crypto roots, most of the revenue base behaves like ordinary software revenue.
There is a token, VVV, and it needs to be described carefully and factually. Venice ties platform usage to VVV through a burn mechanism; the company reports that roughly 42% of supply has been burned to date and that annual emissions are stepping down. Importantly, in this raise Venice chose to sell equity rather than its treasury tokens, and structured investor upside as a mix of equity plus warrants that vest slowly over years. The token functions as an incentive and capacity mechanism in the business, not as the subject of this analysis. It is also a volatile, speculative asset, and nothing here is a view on it.
The competitive set is brutal at the top: OpenAI, Anthropic, and Google have vastly more compute and capital. But Venice isn’t trying to out-model them. Its nearer competitors are other aggregators and privacy-forward tools, and against those its edge is the combination of no-logging, fewer restrictions, a single interface to hundreds of models, and a crypto-native community that markets the product for it.
So where’s the moat? Three places. The first is trust: a privacy brand is hard to fake and harder for an incumbent to copy, because matching it would mean breaking the data-and-moderation model that funds their business. The second is the user relationship: Venice owns the interface while the models underneath commoditize. The third is coming compute ownership, which would deepen the cost advantage. The honest weakness is that Venice doesn’t own a model. It’s an interface, and interfaces can be thinner moats than they appear when the thing they wrap is a swappable commodity.
Spence’s Take
Two models frame this one, and they land on opposite sides.
The bull case runs through Aggregation Theory. In aggregation, the winner isn’t whoever builds the best supply, it’s whoever owns the demand and the interface, because that forces the suppliers to commoditize beneath them. Venice aggregates a specific slice of demand, the users who want privacy and fewer restrictions, and modularizes supply into 200-plus interchangeable models. If it genuinely owns the “private AI” relationship, the model underneath becomes a component it can swap for whatever is cheapest or best, and Venice keeps the margin. The profitability and the $70 million run rate suggest the flywheel is already turning. That’s a strong hand.
What could kill it runs through Revealed Preferences. Privacy is the graveyard of stated preferences. People tell every survey they want it, then trade it away for convenience and a marginally better product, every single time. DuckDuckGo has spent more than fifteen years on exactly this pitch and still sits in the low single digits against Google. The tell is already in Venice’s own numbers: only about 8% of users opt into the crypto payment rail that reflects the privacy-maximalist worldview. The risk isn’t that Venice is wrong that privacy matters. It’s that the set of people who will actually switch and pay for it is smaller than a billion-dollar valuation assumes, and the day a major lab ships a credible “private mode,” the wedge gets narrower fast.
Put simply: the bull case is that Venice owns a durable, defensible niche the giants are structurally barred from serving. The bear case is that the niche is real but small, and that “structurally barred” turns into “hasn’t bothered yet.”
Why It Matters
For investors and VCs. This is a deal worth understanding even if you never touch it, because it’s a clean test of whether a privacy wedge scales into a mass market. The signals to track are specific: paid-conversion rate, the split of revenue across subscriptions, API, and token mechanics, whether the owned data centers actually lift gross margin, and net user retention, since privacy products often acquire well but churn quietly. Watch, too, for any large lab shipping a private mode, which is the single fastest way this thesis compresses. One structural caveat: the crypto-native cap table and the token warrants make this a non-standard equity story, with a volatile asset sitting alongside the equity. All figures here are company-reported, unaudited, and private, and none of this is investment advice.
For potential customers. What’s coming is one app and one API that reach 200-plus models, don’t log your prompts, and apply fewer restrictions, with access designed for both people and automated agents. It solves a real and growing problem: the unease of knowing your prompts may be retained or used to train someone else’s model. Two honest caveats. “Unrestricted” cuts both ways, because fewer guardrails means more responsibility falls on you. And privacy claims should be judged on their technical specifics, not their slogans, so read how the client-side storage actually works before you rely on it.
For competitors and builders. The lesson is bigger than AI. You do not need to own the model to build a real AI business. You can win the interface and the trust layer while the models commoditize beneath you, and Venice is proof that a single operator can reach profitability doing it. There’s a second lesson in the capital structure. Venice raised equity while sitting on a token, deliberately declined to sell its treasury tokens, and structured investor upside as slow-vesting warrants. That’s a hard regulated-markets design problem, how to run both an equity and a token economy without tripping securities law, and it’s the exact terrain we build on at /mkt, where offerings run under Reg A+ with tZERO as the trading infrastructure. If you’re building anything tokenized in a regulated market, study how Venice sequenced equity first and token second.
The Bottom Line
Venice is the rare AI unicorn that earned its valuation with revenue before it raised, built by an operator who understands both privacy and the regulatory edges better than almost anyone in the room. The bull case is genuinely strong: own the private-AI interface, let the models commoditize, keep the margin. The whole thing rests on one number, though, and it isn’t in the press release. It’s the share of users who will actually pay to be forgotten. Watch that, not the valuation.
This is Startup Spotlight. Every week I break down a company worth watching. Subscribe for $7.99/mo to get the full analysis.
This is Startup Spotlight. Every week I break down a company worth watching. Subscribe for $7.99/mo to get the full analysis.
If you want the mental models behind breakdowns like this, my book, Mental Models: How to Think, Act, and Win, is on Amazon now.
Disclosures and sourcing. This post is for informational and educational purposes only. It is not investment advice, a recommendation to buy or sell any security or digital asset, an offer or solicitation, or a suggestion that any product, token, or strategy is suitable for any individual. Venice is a private company and its securities are not offered here.
VVV is a volatile, speculative digital asset. Nothing in this post is a view, forecast, or recommendation regarding VVV or any token, and token mechanics are described solely as reported by the company. Funding figures, valuation, revenue, user counts, token-supply data, and operational claims are as reported by the company, its investors, or named outlets including TechCrunch, The Block, and GeekWire, and have not been independently verified. Private-company figures are point-in-time snapshots and may be stale. Market-size projections are attributed to Bloomberg Intelligence as cited. The reference to the founder’s 2014 SEC settlement reflects the public record and is included as neutral regulatory context.
The author is an officer at /mkt, which operates in regulated securities markets and is referenced here solely as an operating example of building tokenized offerings in regulated markets, not as an investment opportunity or offering of any kind.
