Massachusetts AI Hub's $3.3 Million Second Round: What Five Winning Applied AI Models Reveal About the State's Priorities
November 19, 2026 · 5 min read
Granted Research Team · Editorial policy
The largest award in the Massachusetts AI Hub's second funding round did not go to Harvard or Boston University. It went to the Atlantic Resiliency Innovation Institute, a coastal-resilience nonprofit that won $958,550 to predict flooding at the level of individual pieces of energy infrastructure. The second-largest went to Harvard's Kempner Institute. The third went to Bristol Community College in Fall River.
That ranking is the clearest sign yet of how the state is spending its applied-AI money. On November 12, 2026, the Healey-Driscoll administration announced $3,295,792 for five projects through the Massachusetts Applied AI Models Innovation Challenge, run by the MassTech Collaborative's AI Hub. Recipients bring "more than $1.1 million in matching resources from industry and academic partners." The state counts about $6.1 million invested through the challenge across both rounds.
For universities, nonprofits and colleges in Massachusetts, these are the only two sets of winners so far, and together they are the best guide to what the program rewards.
The Five Winners
| Recipient | Award | Project |
|---|---|---|
| Atlantic Resiliency Innovation Institute (ARII) | $958,550 | AI coastal-resilience platform that combines weather models, sensor networks and infrastructure-impact modeling for hyperlocal flood-risk predictions at energy assets |
| Harvard University, Kempner Institute | $811,578 | Large Video Planner-Industrial (LVP-Industrial), a video-based "world model" that lets robots learn industrial tasks from demonstrations |
| Bristol Community College | $727,270 | AI system that checks workforce skills by assessing work products against industry skill frameworks |
| UMass Lowell Applied Research Corp. (UMLARC) | $500,000 | ARPO Sensor Fusion, reinforcement-learning coordination of multi-sensor counter-drone systems, built as a vendor-neutral interoperability layer |
| Boston University | $298,394 | PandemIQ Llama on the BEACON platform, a domain-specific language model for global disease surveillance, risk assessment and automated epidemiological reports |
The release says the projects are spread across Boston, Cambridge, Fall River, Marshfield and Lincoln. The program's ceiling is up to $1 million per project, covering "the full model lifecycle, data acquisition and real-world testing." No winner this round reached it. ARII came within about $41,000.
What Changed From Round One
The first round, announced October 16, 2025, gave $2,882,219 to seven projects chosen from 61 submissions by universities and nonprofits. Partners added more than $950,000 in matching funds. The winners were Boston Children's Hospital ($200,014, Crohn's disease care), EarthDNA ($1,000,000, footwear-waste AI), Northeastern University (two awards, $504,043 for real-time coastal risk and $16,500 for an open-source generative design model), Western New England University ($500,000, defect detection in metal additive manufacturing), and two WPI projects ($381,931 for machine-learning digital twins that turn waste into energy, and $279,731 for automated dataset generation in industrial recycling).
Comparing the two rounds shows three things.
The awards got bigger and fewer. Round one averaged about $412,000 per project. Round two averaged about $659,000. Total funding rose about 14 percent while the number of awards fell from seven to five. The state appears to be putting more money into fewer projects that can carry a model all the way to field testing.
Coastal resilience won both times. Northeastern's coastal risk work, piloted in Marshfield and Woods Hole, won in 2025. ARII's energy-infrastructure flood platform won the largest award in 2026. Flooding is a priority Massachusetts has funded through several state programs, and applicants should treat it as a standing theme, not a coincidence.
Defense and workforce are new. Round one leaned toward health, manufacturing and the circular economy. Round two added a counter-drone system and a community-college workforce-assessment tool. The counter-drone award fits the state's broader push to grow its defense technology base. The Bristol award shows that the program will fund applied AI at a two-year college when the problem fits regional economic needs. Bristol's tool targets skills checks tied to employers' own frameworks, which is exactly the "downstream application" the challenge asks for.
What the Program Rewards
The program page describes the target as "domain-specific AI models" that "develop, fine-tune or adapt AI models" with "potential for significant downstream applications within the next three years." Its listed sectors are healthcare, life sciences, financial services, robotics, advanced manufacturing, climate tech and education. In round one, selection weighed "impact potential as well as ethical and responsible AI practices."
Read against the ten funded projects, that language translates into a few practical tests.
A named domain and an actual user. Every winner names who will use the model: utility operators, epidemiologists, employers, factory robot integrators, counter-drone operators. None is a general-purpose model. "Fine-tuning a foundation model for climate" would not have made the list. A flood-risk model for specific energy assets in specific coastal towns did.
Data you can actually get. The award explicitly covers "data acquisition." Several winners, including ARII's sensor networks and the WPI recycling dataset project, are as much about building the training data as about training the model. A proposal that assumes clean data already exists is weaker than one that budgets for collecting it.
Real-world testing inside Massachusetts. Pilots in Marshfield, Woods Hole and Fall River are not decoration. The program is economic development money, and reviewers want results they can point to in the state.
Match, even though no ratio is published. Neither round published a required cost-share ratio. Winners still brought in roughly a third of the state award in partner match ($950,000 against $2.88 million in round one, $1.1 million against $3.3 million in round two). Treat match of about 30 percent as the competitive norm, ideally from an industry partner who will use the model.
Openness helps. The Kempner team told the Harvard Gazette it plans to "open source and release this model as well as provide some inference code." Open release fits the Hub's "responsible AI" language and its broader goal of building shared infrastructure for the state.
The Wider Hub and What to Watch
The challenge is one part of a larger state AI strategy. In May 2025 the governor announced a $31 million state grant for an AI computing resources environment at the Massachusetts Green High Performance Computing Center in Holyoke, and projected that the partnership with MGHPCC's member universities could reach about $120 million over five years. The Hub also runs the Data Commons Collaborative, the Pathmaker workforce-training program, The Open Accelerator (its first 16-week cohort, run with Red Hat, was announced November 10, 2026), and a paid AI internship program.
MassTech has not announced a third round of the Applied AI Models challenge, and none of the program materials commit to one. The pattern so far is one round a year, announced in mid-autumn, with about $3 million per round. Groups that want to compete if the challenge returns should line up the hardest parts now: an industry partner willing to commit match, a domain dataset strategy, and a Massachusetts test site.
Institutions that won earlier rounds also have an advantage they can use. The Hub's AI compute environment, data commons and accelerator are all designed to connect to each other, and a team that already has Hub-funded data or compute access can make a stronger case for its next proposal.
"From data gathering to deploying models in the field," AI Hub director Sabrina Mansur said, the challenge "provides support." The winners took that phrase literally: each of the five proposals reads like a deployment plan with a model attached.
When the next Massachusetts or federal applied-AI solicitation opens, Granted can help you shape a deployment-ready proposal that names the user, the data and the test site from the first page.