When Google DeepMind opened applications for its first Asia-Pacific accelerator this summer, it said it would pick 10 to 15 organizations. On Monday it named 16 — an overshoot small enough to be a rounding error, and telling enough to suggest the region has more environmental AI work in the pipeline than Google budgeted for.
The inaugural cohort of the Google DeepMind Accelerator: AI for the Planet (APAC) was announced Sept. 7 by Spencer Low, Google’s head of regional sustainability for Asia-Pacific, and Sami Kizilbash, its head of developer ecosystems. The 16 startups, nonprofits and research teams come from eight countries — Australia, India, Indonesia, Japan, New Zealand, Singapore, South Korea and Thailand — and began a week-long bootcamp in Singapore on Sept. 7 that runs through Sept. 11. Three months of virtual mentorship follow, ending in an in-person Demo Day in December.
Notably, there is no headline funding number, because there is no fund. Support is equity-free: model access, engineering time, and cloud credits subject to eligibility. Graduates join Google’s accelerator alumni network, which the company puts at more than 2,000 startups and nonprofits. Applications closed July 26.
What the 16 teams are actually building
Google sorts the cohort into three buckets. Six teams work on nature protection and climate resilience. Three are from New Zealand alone: 800 Trust and Listening Lab are both building bioacoustic monitoring — using recorded sound to track biodiversity and detect environmental threats — while Wildlife.ai builds open-source, AI-powered camera traps. Singapore’s Kumi Analytics pairs remote sensing with deep learning to establish conservation baselines, South Korea’s TelePIX turns satellite feeds into global mangrove monitoring, and Indonesia’s Yayasan Ekosistem Lestari is building a predictive platform that links ecosystem degradation to disaster risk.
Five teams target agriculture. Indonesia’s Edufarmers pushes near real-time pest, disease and weather guidance to smallholders through the messaging apps they already use — a distribution choice that matters more than the model in a region where the average farm is a few hectares. Thailand’s Living Roots designs biological fertilizers tuned to specific crops. Singapore’s SIGMA, housed at an Illinois-affiliated research center, estimates crop yields from satellite imagery. India’s Terrastack fuses satellite and agronomic data into plot-level land intelligence. Australia’s X-Centric is the outlier: portable, AI-enabled X-ray hardware meant to replace the soil laboratory entirely.
The remaining five sit on the carbon side, and four of them are, in effect, measurement companies. Japan’s Archeda uses satellite data to turn nature-based carbon credits into auditable assets. India’s Climitra Carbon uses geospatial AI to verify invasive-species removal and biochar conversion; Farmers for Forests measures smallholder agroforestry with drones; and Varaha Climate verifies regenerative agriculture so farmers can earn credits. Singapore’s City Syntax Lab is building an agentic AI platform to optimize energy and carbon across city districts.
Teams get Google’s specialized environmental models — AnthroKrishi for agriculture, ForestCast for deforestation risk, AlphaEarth Foundations for planetary mapping, and SpeciesNet and Perch for identifying wildlife from images and audio — alongside Gemini and Gemma.
Why It Matters
Strip away the categories and a single pattern dominates. By eWeek’s count, 13 of the 16 projects primarily monitor, measure, estimate, verify or predict something about the physical world. Only three — Edufarmers, Living Roots and City Syntax Lab — are principally about taking action on the result.
That is not an accident. It reflects where the money is. Carbon markets, sustainability disclosure regimes and climate-risk underwriting all run on measurement, and Asia-Pacific is where the measurement gap is widest. Low has been blunt about why the region is a different problem than Europe or North America. Speaking on the Insignia Business Review podcast in July, he argued that decarbonization framing does not fit economies still building out.
Economic progress is often correlated with increased energy consumption. But the challenge is: how do you decouple increased energy consumption from increased emissions? That is the core aspect.
He has also made the resilience case for AI in concrete terms: AI, he said, can give you extra days of advance certainty in terms of how strong a storm is going to be and what its likely track is, allowing local communities to prepare — the same argument that underpins Yayasan Ekosistem Lestari’s disaster-risk work.
The uncomfortable part is verification. Google’s announcement contains no project-level accuracy benchmarks, no independent field comparisons, no carbon-registry approvals and no named customer deployments, and it does not map individual teams to specific Google models. The accelerator’s own criteria required a working prototype, early validation and proven traction, so that evidence presumably exists somewhere. It just is not public. Compare that with Google’s WeatherNext cyclone research, which shipped with published performance results and head-to-head comparisons against existing forecasting systems.
For a biodiversity monitor, the standard is whether it identifies species reliably. For a crop-yield model, whether it matches the harvest. For a carbon-verification tool whose output may back a financial instrument, the bar is higher still. Accelerator selection tells a buyer that a team cleared Google’s screen. It does not tell them the system works.
There is also the awkward backdrop: Google, like its peers, has reported rising carbon emissions as AI infrastructure expands. Programs like this one partly demonstrate that the same technology can be pointed the other way — which makes independent evidence of impact more important, not less.
What to watch
December’s Demo Day is the first real test. Watch for three things: whether any team publishes accuracy benchmarks or field-validation results rather than pilot anecdotes; whether the carbon quartet — Archeda, Climitra, Farmers for Forests and Varaha — can get methodologies recognized by a registry, which is the only thing that converts a model output into a tradeable credit; and whether Google names a second cohort, and how many it takes. Low and Kizilbash closed their post saying they are eager to see how these teams develop, deploy and scale AI to solve environmental challenges in Asia Pacific. Three months from now, eagerness will not be the measure. Evidence will.
“Economic progress is often correlated with increased energy consumption. But the challenge is: how do you decouple increased energy consumption from increased emissions? That is the core aspect.”— Spencer Low, Head of Regional Sustainability APAC, Google