Economic superintelligence.
For everyone.

The twenty-first century economy generates more structured, learnable data than any system in human history. Every transaction, every shipment, every price change, every weather observation, every policy signal—the economic surface of the planet is already fully instrumented and growing richer every year. At the same time, AI has proven that when you feed structured data into the right architecture at sufficient scale, capabilities emerge that no one predicted. Language models went from autocomplete to reasoning in five years. Protein folding went from unsolved to solved in one. The pattern is clear: given enough data and the right model, hard problems yield.

Economics is next—and it may be the most consequential domain of all.

What makes this moment extraordinary is the convergence. The data has always been there in fragments—transactions here, weather there, supply chains somewhere else, policy in a different silo entirely. But the infrastructure to unify it at scale is new. The model architectures that can learn temporal, causal, multi-agent dynamics are new. The scaling laws that tell us exactly how capability improves with data and compute are new. All three arrived in the same decade. That is not a coincidence—it is the signature of a technology whose time has come.

Economic superintelligence means a model of reality that sees the whole board. Not a dashboard. Not a spreadsheet with better formatting. A continuously learning model that understands how weather drives demand, how demand drives supply chains, how policy reshapes incentives, how consumer behavior shifts in response to all of it at once—and can reason about what comes next. When that model exists and is available to everyone, the quality of economic decision-making across the entire planet improves. Not incrementally. Structurally.

That is what we are building.

Three things had to become true simultaneously. First, AI had to prove it could learn the kind of complex, noisy, multi-scale temporal dynamics that economic systems generate—not just language, but structured causal data where the rules change and the agents adapt. That happened. The scaling laws that drove language models from GPT-1 to GPT-4 apply to any learnable domain with sufficient data and structure. These scaling laws—discovered between 2020 and 2022 by researchers at OpenAI and DeepMind (Kaplan et al., 2020; Hoffmann et al., 2022)—showed that AI performance improves as a precise mathematical function of data, model size, and compute. The relationship is a clean power law spanning orders of magnitude: double the data, get a predictable improvement. This means you can calculate in advance how capable a model will be before building it—which turns AI development from art into engineering. In 2024, researchers formally proved at NeurIPS that time series models follow the same scaling laws—the bridge between language AI and economic forecasting is now mathematically established. Economics has both the data and the structure.

Second, the data infrastructure had to mature enough to pipe weather observations, transaction records, supply chain signals, commodity prices, policy changes, and consumer behavior into a single training pipeline at scale. That happened. The economic data surface is no longer fragmented across incompatible silos—it is integrable, and its volume dwarfs the training sets of the largest models ever built.

Third, the machine learning research community and the economics research community had to start reading each other’s papers—and someone had to build at the intersection, treating them not as separate disciplines but as two halves of the same unsolved problem. The path from here to universal economic intelligence is not speculative. It is an engineering problem with a known trajectory. We are on it.

The history of transformative technologies follows a pattern: they begin as capabilities available only to the largest, most sophisticated institutions, and they become infrastructure available to everyone. Weather models began in military research laboratories and now run on every phone on Earth. Financial data was once the exclusive domain of trading floors and now sits in every brokerage app. The question is never whether the capability will spread. The question is how long it takes, and whether the architecture was designed for universality from the start.

We are designing for universality from the start. The same model that serves a Fortune 500’s supply chain optimization serves a ten-person company trying to understand why demand shifted. The same intelligence that helps a government see economic stress in real time helps a regional employer plan next quarter’s hiring. The model does not dilute as it scales—it improves, because every participant’s data makes it more complete.

Every platform that reshaped its industry—Stripe for payments, AWS for computing, Bloomberg for financial data—was built on the same insight: when you make something universally accessible that was previously scarce or nonexistent, you don’t capture a market—you create one. The economic intelligence market does not exist yet. When it does, it will be one of the largest markets in the world—because every decision in the economy is an economic decision, and every one of them gets better with better inputs.

The first version helps enterprises understand demand. The next version helps governments see policy impact in real time. The version after that gives every person on Earth access to economic intelligence that, right now, does not exist at any price.