Simile AI Valuation 2026: $2B After a 20x Jump in 5 Months

Simile is worth $2 billion as of July 30, 2026 — a roughly 20x markup in five months on a startup that sells companies access to AI-simulated copies of their own customers instead of real focus groups.

Joon Sung Park started Simile in Palo Alto, California in 2025, building on Stanford research he led two years earlier: the 2023 “Generative Agents” paper — nicknamed “Smallville” for the simulated town of AI characters it ran — and a 2024 follow-up that modeled 1,000 real people as language-model agents. Fifteen months after that follow-up, Fortune 100 companies including CVS Health, Deloitte, Gallup, and Wealthfront are paying six and seven figures a year to run experiments against Simile's simulated populations before touching a real customer.

Simile AI Valuation in 2026: How a $100M Round Became $2B in Five Months

Simile is valued at $2 billion after a Series B of more than $200 million closed July 30, 2026, led by Greenoaks. That mark is roughly 20 times where the company stood just five months earlier at its February 2026 Series A, according to Tech Funding News and reporting on the round. Revenue grew 5x over the same five months, and headcount passed 50 employees.

The Funding Timeline: Two Rounds, Seven Months, One Stanford Lab

Simile's entire disclosed funding history sits inside a single calendar year. The research it commercializes is older — Park's original generative-agents work dates to 2023 — but the company itself, the Series A, and the Series B all landed between February and July 2026.

Sources: Menlo Times on the February 2026 Series A; TechCrunch and Dealroom on the July 2026 Series B close. All figures as reported through August 29, 2026.

Who Backed the Round, and Why Greenoaks Led

Index Ventures led the February 2026 Series A and came back to increase its position five months later, a pattern that tends to signal an investor is more convinced by internal metrics than by the round's optics. The Series A also drew individual checks from three of AI's highest-profile operators: former Tesla AI head Andrej Karpathy, World Labs CEO Fei-Fei Li, and Adam D'Angelo, the Quora and Poe founder — the kind of angel list a company gets when the underlying research is already well known inside the field rather than newly discovered.

Greenoaks stepped in to lead the Series B, joined by returning backers Hanabi, Bain Capital Ventures, A*, Factory, and Definition, plus CVS Health Ventures — notable because CVS Health is also a paying customer. A corporate venture arm investing in a vendor it already uses is not unusual, but it does mean CVS Health's enthusiasm for the product and its financial stake in the company's valuation are no longer fully separable facts.

The Founding Team: One Stanford Lab, Four Co-Founders

Simile has four named co-founders, and three of them held faculty or senior research roles at Stanford before spinning the company out in 2025. Joon Sung Park is CEO and led the original research. Michael Bernstein, a Stanford human-computer-interaction professor, is Chief Product Officer. Percy Liang, director of Stanford's Center for Research on Foundation Models, is Chief Scientist. Elaina “Lainie” Yallen rounds out the founding group as Chief Customer Officer. That lineup matters for the pitch to enterprise buyers: the underlying method isn't a black box a startup invented to raise money, it's a published, peer-reviewed research line the founders have been developing since the original 2023 paper — which is also why angel investors like Fei-Fei Li and Andrej Karpathy, who move in the same academic circles, wrote checks before most outsiders had heard of the company.

What an “Agentic Twin” Actually Is

Simile's product is built from real human input, not invented personas. The company runs roughly two-hour qualitative interviews with about 1,000 representative participants, then layers in a much larger training corpus — more than 400,000 consented respondents and 2.9 million survey and behavioral responses spanning over 200 scenarios, according to reporting on the methodology. Those interview transcripts, transaction histories, and behavioral priors get fused into goal-driven LLM agents that Simile recalibrates on a rolling basis, so a twin's answers keep tracking the live opinions of the person it's modeled on rather than freezing at interview time.

The company's own validation testing, per the same reporting, found the twins reproduce a real person's stated attitudes and choices with roughly 85% accuracy, and back-tests of aggregate behavior distributions came in as high as 95% replication against previously observed outcomes. Those are Simile-reported numbers rather than independently audited ones — worth flagging given how much of the company's pitch rests on them.

Pricing scales with how many agents a customer wants modeled, from hundreds of thousands of twins up to multiple millions, and runs from roughly $150,000 to several million dollars a year per account. CVS Health has used the twins alongside a partnership with Gallup to pressure-test consumer-facing healthcare products before launch, one of the clearer examples of the tens of millions of cumulative simulations Simile says it has now run for Fortune 100 clients.

How Simile Compares to Aaru and Evidenza

Simile is not the only company pricing itself on the promise of simulated humans, and it is not obviously the most commercially proven one either. Aaru raised more than $50 million in a Series A led by Redpoint Ventures in December 2025 at a roughly $1 billion headline valuation, built on a multi-tier structure rather than a single clean price. Aaru leans toward outcome forecasting — it has been used to model election results and other discrete predictions — with clients including Accenture, EY, and Interpublic Group. Evidenza, founded by two former LinkedIn B2B Institute leaders, targets enterprise B2B marketing with “synthetic CMO” personas for clients like BlackRock, Microsoft, and JPMorgan, and reported a 0.87 correlation with traditional research methods in a June 2025 pilot with Dentsu.

Simile has raised roughly 6x what Aaru disclosed and carries double the headline valuation, but Aaru got to a $1B mark on a fraction of the capital.

The category is broader than these three names. Ditto lets product teams test Figma prototypes against 100 synthetic users without leaving the design tool, a narrower and cheaper wedge than Simile's enterprise panels. Artificial Societies takes a network-simulation approach, letting personas interact with and influence each other to model how a message spreads rather than just how individuals respond to it. None of these approaches has converged on a standard yet, which is itself informative: research firm estimates put the broader synthetic data market at $1.8 billion in 2024, growing to a projected $8.2 billion by 2029, and a half-dozen venture-backed companies are each betting their approach becomes the default.

What the headline misses

“$2 billion for simulated customers” reads as a straightforward AI-hype story, but the research community that would actually use these tools is far less sold than the cap table suggests. A survey covered by User Interviews found 47% of research professionals skeptical of synthetic users and wanting more evidence before trusting them, with another 17% actively opposed — and not one of the 150 respondents said they had zero concerns. A separate 2026 survey cited by Development Corporate found that while 97% of researchers now use AI somewhere in their workflow, only 8% actually trust AI-generated participants as a stand-in for real ones.

The academic critique is sharper than a trust gap: synthetic agents can reproduce the coherence and emotional tone of a real answer without the lived experience behind it, which risks manufacturing an illusion of consensus rather than surfacing a genuine one. Even Evidenza's own headline number — a 0.87 correlation with traditional research in its Dentsu pilot — is being marketed as a win, but it also means roughly 13% of the signal doesn't match what real humans said, in the industry's own best-case public example. This likely means Simile's Fortune 100 customers are treating agentic twins as a faster first pass that narrows down what to test on real people next, not as a full replacement for it — though none of the reporting on Simile's deployments confirms that's how every customer is actually using the product.

What Would Reprice Simile From Here

Three things would move the number next. The first is independent, third-party validation of the 85%-accuracy and 95%-replication figures — every number cited on that front so far traces back to Simile's own reporting, and a company selling “we know what your customers will do” has an obvious incentive to round those numbers up. The second is data provenance: Simile's twins are built from consented interviews and behavioral logs on real, named individuals, and as more regulators scrutinize how AI companies collect and reuse personal data, a synthetic-user vendor is a more exposed target than a typical B2B SaaS company — a lawsuit or regulatory inquiry over consent terms would hit the valuation harder than it would hit a company with no human data at its core. The third is whether Aaru, Evidenza, or a hyperscaler entrant undercuts Simile on price before its enterprise contracts lock in; at $150,000 to several million dollars a year per account, the switching cost today is mostly the sunk cost of onboarding a panel, not a structural moat.

Watch for a disclosed churn or renewal-rate figure as the next real signal. Simile has been happy to publicize revenue growth and simulation volume; it has not published anything about how many of its early customers renew after a first contract ends, which is the number that would actually confirm whether agentic twins are becoming a permanent part of how these companies make decisions.

Why VCs Are Chasing a 20x Markup in Five Months

A 20x valuation increase in five months is rare even by 2026 AI-funding standards, and it says as much about investor positioning as it does about Simile's product. When Index Ventures returned to increase its stake at the Series B rather than let a new lead set the price on its own, that is a signal existing investors wanted more exposure at almost any markup rather than risk being squeezed out of a company they believed was still under-priced. Greenoaks, which has built a reputation for writing large, concentrated checks into companies already showing traction rather than seed-stage bets, taking the lead is a similar tell: this was priced as a growth-equity round on AI-software multiples, not an early-stage venture bet on an unproven thesis.

The 5x revenue growth Simile reported over the same five months partially justifies the jump on its own — a company growing revenue that fast can outrun even an aggressive markup within a year or two if the growth rate holds. But five months of data is not a track record, and the synthetic-user-research category itself is still unsettled: the same reporting that covers Simile's round also covers Aaru, Evidenza, Ditto, and Artificial Societies raising money on overlapping pitches, which means today's $2 billion mark is being set in a market where the winning approach, and the winning company, has not yet been determined.

For LPs and other investors watching from outside the round, the more durable signal is probably the customer list rather than the multiple: CVS Health, Deloitte, Gallup, and Wealthfront are the kind of logos that take procurement cycles measured in quarters, not weeks, which means Simile had real enterprise traction before the hype cycle caught up to it. That is a different, and generally more reliable, kind of evidence than a headline valuation alone.

The Bottom Line

Simile enters late August 2026 with $300 million-plus raised in a single seven-month stretch, a $2 billion valuation set July 30, 2026, and a customer list — CVS Health, Deloitte, Gallup, Wealthfront — that reads as genuine enterprise traction rather than pilot-stage interest. What it doesn't have yet is independent validation of its 85%-accuracy and 95%-replication claims, or a research community that has stopped arguing about whether simulated humans belong anywhere near real decisions. Both of those gaps, not a new funding round, are what will determine whether the next disclosed number is higher or lower than $2 billion.