Climate risk has rarely mattered more to global food companies.
Weather volatility is disrupting harvests, agricultural sourcing is becoming more complicated, supply chains are under pressure, and companies are being asked to make investment decisions about crops and sourcing regions years – sometimes decades – into the future.
Yet one of the companies built specifically to help them make those decisions has just disappeared.
San Francisco-based ClimateAi has ceased operations after eight years, announcing on 7 August that it would wind down the company and return remaining capital to investors.
According to AGFunderNews, co-founder and CEO Himanshu Gupta attributed the decision to unspecified ‘geopolitical and climate headwinds’ that had made it difficult to continue the company’s mission.
ClimateAi had raised approximately $38 million from investors including Four Rivers Group, Radical Ventures and Robert Downey Jr’s FootPrint Coalition.
On its own, the demise of another climate-tech startup would be noteworthy. But in food and agriculture, there is an uncomfortable precedent.
In May 2024, Gro Intelligence, one of the most ambitious agricultural-data businesses of the previous decade, also closed after failing to secure sufficient additional funding. Gro had raised more than $125 million before its collapse. Its intellectual property was subsequently acquired by agricultural technology company Almanac.
The two companies were different. Their failures should not be conflated.
But together they raise a question that the agricultural sector – and particularly cocoa – should take seriously: If better data is becoming indispensable to managing agricultural risk, why is it proving so difficult to build durable independent businesses selling that intelligence?
Climateai Was Not Solving An Imaginary Problem
There is little evidence that ClimateAi failed because climate risk suddenly became irrelevant.
Quite the reverse.
The company built its ClimateLens platform around a straightforward proposition: conventional weather forecasts become much less useful beyond relatively short horizons, while agricultural businesses increasingly need to make procurement, planting, investment and sourcing decisions months or years ahead.
ClimateAi combined information from satellites, weather stations, radar, ocean buoys and established meteorological models with machine learning, attempting to translate environmental signals into crop- and business-specific intelligence.
Its customer list included businesses such as Dole, Driscoll’s, AB InBev and Suntory, alongside seed companies, agricultural processors and financial institutions including Rabobank. ClimateAi said in late 2024 that it had worked with more than 56 food, beverage and agricultural partners across more than 40 crops and 60 countries.
Earlier growth had appeared strong.
When announcing its $22 million Series B in April 2023, ClimateAi said annual recurring revenue had increased fivefold over the preceding 18 months and its customer base had quadrupled. The company employed around 60 people at that stage.
It also claimed demonstrable commercial benefits. ClimateAi said seed company Advanta had captured additional sales after using forecasts to anticipate weather-driven changes in demand, while its case-study portfolio included supply-chain modelling, agricultural investment risk, crop-yield outlooks and commodity hedging.
This is what makes the shutdown significant.
ClimateAi was not simply selling carbon-accounting software into an evaporating ESG fashion. It was trying to address one of the most tangible problems facing agriculture: what happens next, where will it happen, and what should a company do about it?
The Warning Signs Were Hidden In The Business Model
ClimateAi itself had already articulated the commercial difficulty.
In an extensive December 2024 interview, its executives identified three barriers to adoption.
First, climate intelligence often did not have an existing corporate budget line.
Second, customers had to change established decision-making behaviour.
Third, ClimateAi had to persuade companies that its product was not simply a sustainability tool, but something capable of affecting procurement, risk and, ultimately, the bottom line.
That distinction is crucial.
A company may believe climate volatility represents a major strategic threat without having a procurement process capable of spending hundreds of thousands of dollars on an independent intelligence platform.
The person who needs the information may work in sustainability.
The person who can act on it may work in procurement.
The financial benefit may eventually appear in agricultural sourcing, inventory, insurance, trading or capital expenditure.
And the purchasing decision may belong somewhere else entirely.
ClimateAi was seeing signs that those silos were beginning to converge. In January this year, COO Will Kletter described procurement and sustainability teams increasingly coming together as companies treated climate as a core business variable.
But the company was still investing heavily.
Asked in January whether ClimateAi was profitable, Kletter said remaining competitive in AI was more important than immediate profitability, provided the company’s unit economics remained favourable.
Just over six months later, ClimateAi was gone.
And Only 20 Months Earlier, Profitability Was The Target
The chronology makes the closure more striking.
In December 2024, Gupta said ClimateAi expected to become profitable around the third or fourth quarter of the following year and described the business as a scalable SaaS model with low marginal costs for additional customers.
The company hoped first to demonstrate value across roughly 200 partners and then potentially raise a Series C, according to AgFunderNews.
By January 2026, it was releasing an agentic-AI tool intended to turn Growing Degree Day calculations and agricultural forecasts into operational decisions.
Instead of simply providing graphs, the company’s ambition was to embed intelligence directly into customer workflows: predicting crop development, processing volumes, weather risks and potential agronomic responses.
That is a sophisticated evolution from selling data.
It is also revealing.
The competitive benchmark was no longer simply whether ClimateAi had better forecasts. The platform increasingly needed to integrate those forecasts into the customer’s everyday operations and tell users what to do next.
Information alone was becoming insufficient.
The Gro Intelligence Comparison
There are obvious parallels with Gro Intelligence, but important differences.
Founded by former commodities trader Sara Menker, Gro attempted to create an enormous agricultural intelligence platform by bringing together government statistics, satellite imagery, trade data, weather information and other sources.
It attracted major investors and an $85 million Series B in 2021.
But its commercial model struggled.
According to former employees interviewed by AgFunderNews, the company generated a substantial portion of its revenue from a limited number of large customers, while some prospective projects became closer to bespoke consultancy than repeatable software revenue.
One former employee described a ‘fundamental mismatch between the product and the market’.
Gro also experienced serious operational and financial turmoil before closing, including layoffs and an inability to make payroll. These circumstances distinguish it materially from ClimateAi, which says it is returning remaining capital to investors.
Indeed, ClimateAi’s leadership had explicitly studied the Gro problem.
In 2024, Gupta argued that agricultural-data startups could become trapped by venture-capital pressure: chasing almost any available revenue, taking on increasingly bespoke customer requirements and gradually becoming low-margin consultancy businesses rather than scalable software companies.
ClimateAi, he said, intended to avoid that trap by concentrating on repeatable use cases.
Its subsequent closure does not prove that analysis was wrong.
It suggests that avoiding the consultancy trap may not, on its own, solve the underlying economics.
Cocoa Demonstrates Why The Problem Is Difficult
ClimateAi’s cocoa work provides an unusually good illustration.
The company told AgFunderNews that traditional yield modelling was difficult in cocoa because historical datasets were insufficient.
Instead, ClimateAi developed a cocoa risk outlook, using indicators such as heat stress during pollination or drought during harvesting, then examining how such conditions had historically influenced markets.
That is an important acknowledgement.
Artificial intelligence cannot manufacture observations that do not exist.
Agricultural intelligence depends on the quality, continuity, granularity and comparability of the underlying information.
This matters particularly in cocoa, where production is concentrated among millions of smallholders and data can be fragmented across governments, traders, certification systems, cooperatives, companies, satellites, weather networks and research institutions.
ClimateAi was nevertheless pushing further into the sector.
In August 2025, it announced a collaboration with Japanese technology group NEC specifically examining climate adaptation for cocoa and rice in Africa.
The project sought to calculate the potential economic return from interventions including irrigation, changes in planting time and the adoption of climate-adapted varieties.
The ambition was not merely to predict climate change, but to answer the financially relevant question: which adaptation intervention is worth paying for?
That distinction may eventually determine whether agricultural intelligence becomes economically sustainable.
The Climate Adaptation Paradox
There is a broader contradiction.
According to UNEP’s 2025 Adaptation Gap Report, developing countries could require around $310–365 billion annually for adaptation by 2035, while international public adaptation finance flows were just $26 billion in 2023.
At the same time, businesses still struggle to obtain or interpret the very information needed to decide where adaptation capital should go.
The OECD reported this year that private companies face barriers including limited access to relevant weather data, insufficient technical capacity and difficulty quantifying the costs and benefits of adaptation.
In other words:
The demand for better decisions is enormous. The need for data is enormous. The financial exposure is enormous.
But the willingness – or organisational ability – to pay standalone intelligence providers enough to sustain expensive modelling, data science and AI teams may be considerably smaller.
That is the paradox exposed by ClimateAi.
Data Has Value. But Where Is That Value Captured?
Gro’s technology did not vanish when Gro disappeared.
Almanac acquired its assets and incorporated Gro’s AI models into a broader agricultural technology platform already serving more than 100 million acres.
That may offer a clue.
Agricultural intelligence may ultimately be most valuable inside another commercial activity, rather than as an isolated product.
A trader can use intelligence to improve procurement.
An insurer can use it to price risk.
A bank can use it to assess lending.
A processor can use it to manage inventories.
A seed business can use it to determine where varieties should be developed or sold.
A farmer platform can use it to improve agronomic decisions.
The intelligence can create considerable economic value without the intelligence provider necessarily being able to capture that value through subscriptions alone.
That distinction matters.
The failure of an intelligence company does not necessarily mean its intelligence had no value.
It may instead mean that the value accrued elsewhere in the system.
The Lesson For Cocoa
For cocoa, there is a temptation to believe that more data automatically produces more intelligence.
It does not.
The ClimateAi experience suggests several additional requirements.
Data needs context. Forecasting needs interpretation. Models need validation. Intelligence needs to reach the person able to act on it. And, crucially, the decision resulting from that intelligence needs to create enough measurable economic value to justify the cost of producing it.
This is particularly relevant as cocoa companies invest in farm mapping, traceability, satellite monitoring, climate modelling, deforestation assessment, crop forecasting and regulatory compliance.
Each creates another layer of information.
The danger is creating an increasingly sophisticated collection of disconnected data systems without a common mechanism for answering the question that ultimately matters:
What decision should change because we know this?
ClimateAi called this ‘decision accuracy’ rather than simply forecast accuracy.
A forecast did not have to predict rainfall perfectly, the company argued. If it gave a producer enough warning to alter a planting decision and prevent a significant loss, then the intelligence had delivered value.
That principle extends well beyond weather forecasting.
It applies to cocoa supply, disease, farmer economics, regulation, sustainability and markets.
From Data Platforms To Decision Infrastructure
ClimateAi’s closure should therefore not be interpreted as an argument against agricultural AI.
Nor does the demise of Gro Intelligence mean the industry should abandon ambitious attempts to connect agricultural datasets.
The opposite conclusion may be more useful.
Both cases show how difficult – and expensive – it is to build an independent intelligence layer between fragmented agricultural data and commercial decision-making.
The next generation of platforms may consequently need to be narrower, more specialised and more deeply integrated with existing workflows.
Rather than attempting to become universal warehouses of agricultural knowledge, successful systems may concentrate on connecting selected high-value signals, explaining why they matter and making their implications immediately usable.
The commercial test is unlikely to be:
How much data does the platform possess?
It will be:
Does the platform help someone make a better decision?
And can that improvement be demonstrated strongly enough that somebody will consistently pay for it?
ClimateAi spent eight years trying to answer that question.
Its technology helped establish climate adaptation as a real business category. Its customers apparently found real applications for the intelligence. Its work reached directly into cocoa and the economics of climate adaptation.
And yet ClimateAi itself did not survive.
That should make the food and agriculture sector pay attention.
Because in an industry producing more information than ever before, the lesson from ClimateAi and Gro Intelligence may ultimately be uncomfortable but valuable:
Data becomes gold only when somebody can connect it to a decision. Otherwise, however sophisticated the technology behind it, it risks becoming dust.