Inference time cut from minutes to seconds
New models deployed in 1.5 weeks
Platform delivered months ahead of original timeline
Bold built agricultural financing AI infrastructure on Google Cloud, cutting credit analysis from weeks to hours for BBVA Argentina.
Bold built agricultural financing AI infrastructure on Google Cloud, cutting credit analysis from weeks to hours for BBVA Argentina.
Building the infrastructure behind
agricultural credit at scale
Bold is an Argentine deep-tech startup building the agricultural financing AI infrastructure that turns Earth observations data into credit intelligence for financial institutions across Latin America.
Founded in 2024, the company processes real-time data from climate, soil, radar, laser, and multispectral sensors covering 600 million hectares in Latin America, detecting planting and harvesting dates of crops, rotation patterns, yield estimates, water stress, carbon capture, and deforestation at the level of each hectare across the continent.
Their platform runs on two internal layers: Daredevil, the Earth-sensing layer that ingests and processes signals, and Thot, the AI agentic layer that turns those signals into actionable economic intelligence.
When BBVA Argentina, one of the country's leading agricultural lenders and part of the group named the world's most innovative bank in 2026 by Global Finance, needed real-time visibility into its agricultural lending portfolio, the challenge was clear.
Traditional methods like physical field inspections and farmer self-reporting were slow, incomplete, and impossible to scale across millions of geographically distributed hectares. Without reliable field-level data, every credit decision for an agricultural producer carried more uncertainty than it should.
"When you are building something nobody has built before, there is no manual," says Kevin Canova, CTO and co-founder of Bold. "You figure out the architecture, you test it, and then you need infrastructure that does not collapse when the data starts coming in at scale."
To match that ambition, Bold evaluated other cloud providers before choosing Google Cloud for its tight product integration, lower latency for real-time inference, and confidence that the infrastructure could grow with them across the continent.
Partner Maplink helped homologate the architecture to ensure it could perform in production from day one. "Latin America has 600 million hectares of agricultural land," Canova says. "We have made 40 million visible so far in Argentina. Google Cloud is the foundation that lets us scale to the rest."
Turning Earth observation data into
real-time lending decisions
With Google Cloud in place, Bold built a pipeline that takes raw signals and turns them into credit-ready intelligence in near real time.
The process starts with Daredevil ingesting signals from satellites and data streams across millions of hectares. That data flows into BigQuery, which serves two roles simultaneously: a data warehouse that stores and organizes the geospatial information, and a feature store that exposes the right data the moment a request comes in. When a financial institution submits a polygon representing a field, BigQuery surfaces everything Bold knows about that land instantly.
Gemini Enterprise Agent Platform is where Thot does its work, training and deploying the models that power real-time insights. Thot, powered by Gemini, adds a layer of contextual reasoning on top of the signals, interpreting complex spectral patterns that traditional computer vision would miss, like identifying whether a field is growing a second-crop soybean based on multispectral imagery alone. Gemini Enterprise Agent Platform Batch Inference scales the inference API automatically, handling requests independently of their size, whether the query covers 1,000 or 1 million hectares.
What that looks like in practice for BBVA Argentina is striking. A loan officer reviewing a producer's credit application no longer needs to wait for a field visit. Bold's platform identifies the sowing and harvest dates for that field, estimates the yield, and infers the producer's expected cash flow. The bank can then align loan maturity dates to match when the producer will actually have money.
"What used to be a slow, manual, fragmented process is now a complete flow," says Juan Pedro Gazzotti, head of agribusiness at BBVA Argentina. "With minimal information, we can infer everything we need and finish a credit analysis in 48 hours."
We process billions of data points for millions of hectares that arrive at different times of the year. The peaks are brutal. Google Cloud meant we never had to choose between speed and scale.
Kevin Canova
CTO and Co-founder, Bold
Expanding across Latin America
to close the credit gap
The results speak for themselves. Bold's platform now powers $1.2 billion in annual agricultural loans at BBVA Argentina. Inference time dropped from minutes to seconds per 1,000 hectares, while API calls to fetch field data were reduced by 25%. New models go from ideation to production in a maximum of three weeks, which helped Bold reach this point two years ahead of their original project timeline.
For Canova, the numbers are only the beginning. Latin America's agricultural sector has long suffered from a fundamental information gap: financial institutions betting on producers they couldn’t actually see, and producers underserved because the data to assess them didn’t exist.
"Every credit decision used to be a bet," adds Canova. "What we are building closes that gap. It turns an uninformed bet into a decision backed by what is actually happening in the field."
The next step is continental. Bold is now expanding into Uruguay and Colombia, two markets with enormous agricultural economies and the same structural visibility problem that drove the BBVA partnership in Argentina. Each market has its own financial logic and crop calendar, and Bold's platform is built to adapt to that complexity region by region.
We were not even finished when the market started asking for it. That tells you everything about the size of the problem. With Google Cloud, we could move fast enough to actually answer that demand.
Kevin Canova
CTO and Co-founder, Bold
For other deep-tech startups considering a similar path, Canova's advice is direct. Building something this hard means there is no manual, no blueprint to follow. The architecture has to be right from the start, because if it doesn’t scale, nothing else matters. "The only way a startup wins is by being fast," he says. "Google Cloud is what lets us move at the speed this market needs."
Bold is an Argentine deep-tech startup building the AI infrastructure that turns Earth observations data into economic intelligence for financial institutions across Latin America.
Industry: Agriculture
Location: Argentina
Products: Google Cloud, BigQuery, Cloud Run, Gemini, Gemini Enterprise Agent Platform, Apigee
