Google DeepMind picks 16 climate and conservation teams for first Asia-Pacific AI accelerator

Google DeepMind has selected 16 startups, nonprofits and research teams from across the Asia-Pacific region for the inaugural cohort of a new accelerator devoted to applying artificial intelligence to environmental problems, the company said in a post on its corporate blog this week. The program, called the Google DeepMind Accelerator: AI for the Planet, opened with a four-day, in-person bootcamp in Singapore that ran from September 7 through September 11. The announcement was co-written by Spencer Low, Google's head of regional sustainability for Asia-Pacific, and Sami Kizilbash, the company's head of developer ecosystems for the region, who described the selected teams as receiving "access to the latest Google AI stack (including specialized frontier models), tailored support, and mentorship" from Google's engineers over the following three months.

A three-month sprint from Singapore to December's demo day

The accelerator's official program page shows the company had originally planned to admit between 10 and 15 organizations; it ultimately selected 16. Applicants had to be headquartered in the Asia-Pacific region, have a working prototype or minimum viable product, treat AI as the core of their solution rather than an add-on, and commit at least two or three senior team members to the program. Applications closed on July 26.

What follows the bootcamp is three months of virtual mentorship, running through December, when the cohort will present its progress at an in-person Demo Day. Participation is equity-free: Google does not take a stake in exchange for admission. Instead, teams get technical support from Google engineers, potential cloud credit grants, and access to a stack of proprietary and general-purpose AI models, including Gemini and Gemma alongside more specialized tools such as AlphaEarth Foundations, ForestCast, SpeciesNet, Perch and an agricultural model referred to internally as AnthroKrishi.

Some of those tools have their own published research behind them. AlphaEarth Foundations, the satellite-embedding system several cohort members plan to use for land and water monitoring, is described in a DeepMind research paper as producing 10-by-10-meter geospatial embeddings with roughly a 24 percent lower error rate than comparable mapping methods when reference data is scarce. SpeciesNet, used for camera-trap identification, is described in the same body of DeepMind research as having been trained on more than 65 million labeled images spanning 2,498 animal categories.

Eight countries, three problem areas

The 16 selected organizations are drawn from India, New Zealand, Singapore, Indonesia, Australia, Japan, South Korea and Thailand — with India contributing four teams and New Zealand and Singapore three apiece. Their work clusters into three broad categories: biodiversity and climate-resilience monitoring, sustainable agriculture, and carbon or climate-credit verification.

Among the named participants are New Zealand's 800 Trust and Listening Lab, both of which use acoustic sensors to detect environmental threats or track species by sound, and Wildlife.ai, which builds open-source AI-powered camera systems for conservation groups. South Korea's TelePIX is using satellite imagery for mangrove monitoring, while India's Terrastack, Farmers for Forests and Varaha Climate are working on carbon measurement and agricultural applications. Google has described the cohort as joining a broader alumni network of more than 2,000 companies and nonprofits that have passed through its various Google for Startups accelerator programs worldwide, though this particular track — aimed specifically at environmental applications in the Asia-Pacific region — is new.

Independent reviewers flag validation gaps

Trade coverage of the announcement has been broadly positive about the program's reach but pointed in places about what it does not say. An analysis published by eWeek found that 13 of the 16 selected projects are primarily focused on measuring, monitoring or predicting environmental conditions — tracking biodiversity through sound, cameras or satellite imagery, or verifying carbon output — rather than on deploying direct interventions. The outlet noted that Google's announcement includes no accuracy benchmarks, no field-validation comparisons against existing methods, and no individual deployment timelines for any of the 16 teams, a gap it called significant given that measurement and prediction tools generally require independent verification before governments, insurers or carbon markets can rely on them.

A separate rundown of the cohort from CompleteAITraining likewise catalogued the full roster of organizations and their focus areas without turning up disclosed funding amounts or application statistics — Google has not said how many teams applied for the 16 spots, nor whether direct cash grants, as opposed to cloud credits and mentorship, are part of the package. Google's own materials describe the goal in general terms, saying the program exists to help participants "scale the next generation of solutions for our planet," without elaborating on how success will be measured at the December demo day.

For the participating organizations, most of which are early-stage and resource-constrained, the immediate benefit is access to computing infrastructure and modeling tools that would otherwise be costly or technically out of reach — AlphaEarth's global satellite embeddings and SpeciesNet's image-recognition training set, in particular, represent years of DeepMind engineering that a small nonprofit could not replicate independently. Whether that translates into measurable on-the-ground results for biodiversity, agriculture or carbon accounting will not be clear until the cohort's work is presented and, ideally, independently checked against the kind of field data that outside reviewers say has so far been absent from Google's public description of the program.

The accelerator's next visible milestone is the December Demo Day, when the 16 teams are expected to show what they built during the mentorship phase. Google has not said whether it plans to run additional Asia-Pacific cohorts or expand the model to other regions, though the company has run similar Google for Startups accelerator tracks — focused on different themes — in other markets for several years.