You have roughly a 0.00003% chance of investing at pre-seed or seed in a company that eventually becomes worth $1 trillion.
Now imagine something different.
What if, when a company was still at seed, you could simulate its possible futures and discover that it had an 86% probability of following the characteristics of an extreme-outlier trajectory?
Not 86% certainty that it would become exactly a trillion-dollar company. That would be unrealistic. But 86% confidence that the company was entering the right combination of market, adoption, technology, team, competitive and capital conditions that historically precede an exceptional outcome.
That would completely change venture capital.
Because today, the hardest problem in VC is not finding startups. There are databases for that. It is not finding market information. There is more information available than ever. The hard problem is determining, while everybody is looking at roughly the same company, which version of its future is most likely to happen.
And the last two years have shown just how difficult that has become.
Cursor is what happens when the future arrives faster than your spreadsheet
In April 2022, Anysphere reportedly raised around $400,000 in pre-seed funding.
Four young MIT founders. No famous Big Tech executive founding team. No huge revenue. And more importantly, the final product was not obvious yet.
Before Cursor, the founders spent months experimenting with AI for computer-aided design. Eventually, they moved back toward programming and built what became Cursor.
That makes the investment problem much harder.
At pre-seed, you are not always predicting a product.
You are predicting a team navigating an unknown future.
By October 2023, Anysphere raised an $8 million seed round led by the OpenAI Startup Fund. Less than a year later, it raised at roughly a $400 million valuation. The company then moved through multi-billion-dollar valuations at extraordinary speed before eventually reaching a $60 billion acquisition outcome.
Try modelling that in a traditional seed-stage spreadsheet.
You would have needed to predict not only revenue growth, but a pivot, rapidly improving foundation models, a massive change in developer behavior, the emergence of AI coding as a new category, and Cursor surviving competition from Microsoft and GitHub.
A spreadsheet is good at extrapolating an existing business.
But sometimes the business responsible for the return doesn’t exist yet when you invest.
OpenRouter and Etched show the same problem
OpenRouter was valued at roughly $1.3 billion in May 2026 and was acquired only months later in a deal reported above $8 billion.
Why could its value change so quickly?
Because the world around it was changing. More AI models were appearing, applications were becoming multi-model, inference spending was growing and developers increasingly needed infrastructure to route between providers.
If you evaluated OpenRouter only as an API layer, you could miss the opportunity.
If you simulated a future with hundreds of models and billions of autonomous agent requests choosing models dynamically, OpenRouter starts looking very different.
Etched is another extreme example. The company went from an early reported valuation of around $34 million to approximately $21 billion, while securing more than $1 billion in customer contracts.
But early investors had to make a chain of uncertain bets: transformers would remain important, inference demand would explode, specialized chips could outperform general-purpose GPUs economically, manufacturing would work and customers would trust a startup with critical infrastructure.
Every one of those assumptions had a probability. And those probabilities interacted.
That isn’t just forecasting. That’s simulation.
Humans are good at conviction. Machines can be better at exploring probability.
Great VCs can meet a founder and recognize ambition, obsession, technical ability or something unusual that isn’t visible in the numbers.
That human judgment matters.
But humans are less reliable when asked to combine dozens of uncertain variables at once. We anchor on familiar patterns. We like recognizable signals. Four ex-OpenAI executives can feel safer than four students. A startup with $10 million ARR is easier to defend than a team still searching for the right product.
The problem is that extreme venture outcomes frequently come from companies that initially look abnormal.
So the future shouldn’t be AI replacing investors.
It should be human judgment + simulation.
The human asks: “Do I believe this team is exceptional?”
The simulation asks: “What happens if they are?”
In many worlds, a company fails. In some, it becomes a solid $1 billion company. In others, competitors destroy it.
But perhaps a small group of simulations shows something unusual: the market expands faster than expected, the founders pivot at the right time, product adoption accelerates, talent concentrates around the company and a new category emerges.
The system doesn’t need to predict: “SpaceX will acquire Cursor for exactly $60 billion.”
It needs to say: “This team, market and product have an unusually large number of credible paths toward an extreme outcome.”
That is far more useful.
And then you can stress-test it.
- What happens if OpenAI launches a competitor? Run it again.
- What if inference becomes 90% cheaper? Run it again.
- What if the founders pivot? Let them pivot inside the simulation.
- What if transformers disappear? Run Etched again.
- What if the AI model market consolidates to three providers? Run OpenRouter again.
Prediction gives you one answer. Simulation lets you explore the worlds behind the answer.
The next edge in VC might be simulation
The first generation of venture was powered by networks. Then came data.
Now investors are using AI to source companies, analyze markets and accelerate due diligence. But most AI today still helps investors understand what already happened.
Simulation could help answer a different question: what could happen next?
The goal isn’t to perfectly predict the next trillion-dollar company. It is to become slightly better at identifying the tiny number of companies with asymmetric futures.
Because in venture, being slightly better at finding the extreme tail can change an entire fund.
What we are building
We are building a simulation layer for VC, private equity, trading, banking and investors to model not just what an asset or company looks like today, but the different futures it could move into.
For venture, that means simulating founders, markets, competitors, technology shifts, financing and exit paths before writing the check. For PE, it means testing operational decisions and value-creation plans before executing them. For trading and banking, it means simulating how markets, participants and external shocks could interact before capital is deployed.
The goal is not to claim that we can predict the future with certainty. It is to make capital allocation less static, more probabilistic, and better prepared for worlds that have not happened yet.
Because the future of investing may not be about having more data. It may be about being able to simulate what happens next.