Why Japan's largest bank is betting on Fundamental
Gabriel Suissa
CHIEF PRODUCT OFFICER & CO-FOUNDER, FUNDAMENTAL
Walk into any global bank today and you'll find two versions of AI. The first is focused on efficiency: language models drafting emails, summarizing documents, answering customer queries. Doing the same work faster and at a lower cost.
The second version is hidden yet transforming business performance. Not AI that writes, but AI that predicts. Which trade to make. Which customer will churn. Which loan is likely to default. The decisions that determine whether a financial institution wins or falls behind.
Today, Mitsubishi UFJ Financial Group, Japan's largest banking group and one of the biggest financial institutions in the world, placed a bet on that second future.
MUFG will deploy NEXUS, our Large Tabular Model, across key areas of its business. From markets and trading to digital banking, the ambition is simple: put more accurate predictions behind more of the organization’s decisions.
"We're excited to be amongst the first to adopt this groundbreaking technology and deploy it into real-world operations. From markets and trading to digital banking, we see enormous long-term potential in what we can build together."
The data nobody talks about
For all the excitement around AI, one reality often gets overlooked: most enterprise data isn't text.
Banks don't run on prose. They run on tables.
Transactions, positions, ledgers, customer records, market data. Decades of structured information that institutions have invested billions of dollars to collect, govern, and maintain.
As our CEO, Jeremy Fraenkel, puts it:
"Trillions of dollars of value remains locked in the huge structured datasets that every enterprise is sitting on," he says. "Financial institutions have spent decades building systems around that data, but traditional machine learning workflows remain manual, fragmented, and difficult to scale. MUFG understands that the next wave of enterprise AI will be driven not just by language models, but by systems purpose-built to predict outcomes from tabular data."
That distinction matters. Language models were built to understand language. Predicting outcomes from structured enterprise data is a fundamentally different problem. It demands a different kind of model.
That's exactly what NEXUS was built to do.
From pilot projects to real operations
What makes this relationship notable isn't the intent to experiment. Enterprises have been experimenting with AI for years. What’s different is the intent to operationalize it.
The next generation of AI won’t live on the sidelines. It will increasingly sit inside the decisions that determine revenue, risk, customer growth and operational performance.
As part of the partnership, MUIP has also made an equity investment in Fundamental, reinforcing its long-term commitment to bringing predictive AI into the core of the business.
The efficiency era of enterprise AI cut costs. The prediction era will create alpha. And the trillions of dollars of value still locked inside structured enterprise data will belong to the organizations – like MUFG – that learn to predict what's coming next.
