Simile Raises $200M to Simulate all 8 Billion People on Earth

Simile, the Palo Alto-based startup building what it calls "the simulation company," just closed a $200M Series B at a $2 billion post-money valuation. The round was led by Greenoaks, with Index Ventures doubling down and returning and new investors joining in. The stated mission: simulate all eight billion people on earth, accurately and honestly.

From a Stanford dorm sim to a $2B company in under a year

Joon Sung Park, founder and CEO of Simile, is also the creator of Stanford's "Smallville" generative agents study, a 2023 research project that placed 25 LLM-driven agents in a Sims-like virtual town. In that paper, the agents woke up, cooked breakfast, went to work, formed opinions, and gossiped. One agent organized a Valentine's Day party; others coordinated invitations and showed up at the right time. The paper won best paper at UIST '23.

Park's academic work produced two landmark papers: Social Simulacra (2022), which used GPT-3 to simulate entire subreddits, and Generative Agents (2023), which created that town of 25 AI agents. His academic path runs through Swarthmore, UIUC, and a Stanford PhD co-advised by Michael Bernstein (HCI) and Percy Liang (NLP/foundation models). All three are now Simile co-founders, alongside Lainie Yallen.

In the five months since launch, Simile has grown revenue 5x, built a new foundation model for human behavior that has run tens of millions of simulations for Fortune 100 enterprises, trained a confidence model that predicts the accuracy of individual simulations, and expanded to a global team of 50+.

The deal terms

• $200M Series B at a $2 billion post-money valuation
• Led by Greenoaks, with returning investors Index Ventures, Hanabi, Bain Capital Ventures, A*, Factory, and CVS Health Ventures, plus new investor Definition
• Capital will fund training of core human behavior foundation models, expanded simulation compute, and commercial engineering across enterprise healthcare, financial services, and media

This follows a $100M Series A led by Index Ventures earlier this year, bringing total raised to $300M in under 12 months. Series A backers included Fei-Fei Li and Andrej Karpathy.

What Simile actually builds

Simile's core product is a human behavior foundation model, a purpose-built AI trained to replicate how real, irrational, diverse people actually think and decide rather than to be smarter or more rational than them. Park frames today's frontier models as the "CPU of intelligence," rational and superhuman at problems with right answers, and Simile as the "GPU of intelligence," built to encode the diversity of people's values, preferences, and tastes.

Simile grounds its simulations in real behavioral data collected from representative populations via Gallup partnerships, producing generative agents: AI actors equipped with memory, reflection, and planning capabilities, designed to behave like specific people rather than like an average or an archetype. In a validation study, Simile interviewed 1,052 people, built agents from those interviews, and tested whether the agents could reproduce participant responses across survey and experiment tasks. On the General Social Survey, agents reached 85% of the accuracy humans achieved when retaking the same survey.

Simile also ships a confidence model, a meta-layer that predicts how reliable each individual simulation is before you act on it. Rather than giving you a single system-wide accuracy number, it flags which specific predictions to trust, a meaningful distinction when the outputs are feeding real business decisions.

Who is already using it

Enterprises including CVS Health, Wealthfront, Deloitte, and Gallup are using Simile to strategize product launches, optimize customer experiences, and enter new markets. Use cases span a wide range:

• CVS simulated customer behavior across 9,000+ stores for product placement optimization
• Companies rehearse earnings calls to predict analyst questions
• Organizations model litigation outcomes, test policy changes, and validate product concepts before launch

Why the timing makes sense

The global market research industry is worth roughly $80 billion. Traditional focus groups cost $5K–$20K per session and take weeks to yield results. Simile promises comparable signal in hours, at lower cost, and at a scale no human panel could match. Already, 69% of market researchers report incorporating synthetic data into their work.

The broader shift Simile is betting on: generative AI spent 2024–2025 focused on creating content; 2026 is about predicting behavior. In a world saturated with AI-generated output, the bottleneck is no longer creation. It's knowing what to create, for whom, and whether it will work.

Simile's academic roots at Stanford give it a credibility advantage in that environment. The team treats simulation fidelity as a scientific discipline, publishing validation methodology rather than just claiming accuracy scores. That matters because enterprise buyers are now demanding proof, not just demos.

Who wins and who loses

Enterprises that need fast, large-scale decision intelligence gain the most. Simile's pitch is a pre-flight simulator for consequential decisions: rehearse a product launch, a pricing change, or a policy shift before exposing real customers, employees, investors, or voters to it. Traditional enterprise market research studies cost $50K–$500K each; synthetic approaches promise 10–100x cost reduction at scale.

Traditional market research firms face the clearest pressure. Gallup is both a Simile partner and a potential disruption target, a tension worth watching as the relationship deepens.

Park's larger ambition is a "CERN of human society" capable of modeling bank runs, climate cooperation, or early signals of democratic collapse. Nearer term, the more immediate applications are in polling, policy testing, and drug trial design, anywhere that running a real human study is slow, expensive, or ethically constrained.

The real risks

The 85% accuracy result is meaningful but bounded. It shows that interview-trained agents can reproduce survey responses with real fidelity. It does not prove Simile can predict product choices, retail behavior, policy reactions, ad responses, or buying-committee decisions across every domain. Cross-domain validation remains limited, and the gap between "reproduces survey answers" and "predicts consequential decisions" is where the scientific critique lands hardest.

Regulatory exposure is also building. EU AI Act high-risk provisions and California's CPPA automated decision-making rules are already in effect or imminent for exactly this class of system. Customers running Simile outputs through pricing, product access, or healthcare-adjacent decisions will demand auditability, bias documentation, and compliance architecture.

Competition is crowding in fast. Aaru raised at a $1 billion headline valuation in December 2025, led by Redpoint Ventures. Where competitors are racing to replace human feedback with pure AI, Simile's differentiator is its tether to real behavioral data. That grounding is also a constraint on how fast it can scale, since the data collection pipeline is harder to replicate than a model architecture.

With $200M in fresh capital, a 5x revenue run in five months, and Fortune 100 customers running production simulations, Simile has moved well past proof-of-concept. The harder question, whether an 85% accurate simulation of a person is close enough to stake billion-dollar decisions on, is one the market is now actively answering.