Exploit Summit Concludes with Advances in Distributed AI

Distributed artificial intelligence took center stage at the Exploit Summit, which concluded on September 29 at New City Gas in Montreal. The two-day event, held from September 28-29, 2026, brought together 400 attendees and more than 70 subnet teams from across the Bittensor network. Over 10,000 concurrent viewers followed the sessions on X Live broadcasts.

Chutes announced an 8-billion-parameter Paralex training run, which cost $6,500 and was demonstrated on a phone. IOTA set out plans for an SDK and Liquid Compute platform, which will allow builders to pre-train, fine-tune, and post-train models on its GPU network. Metanova Labs reported that molecules produced through its drug-discovery subnet are being tested and synthesized in a real-world laboratory. Carbon won the summit's Pitchtensor competition, while OpenRoboto demonstrated Shift, its hardware for converting real-world robotic work into training data. Green Compute announced a $4 million data center planned for the end of October.

Founders and senior contributors from Chutes, Macrocosmos, Metanova Labs, Swarm, 404-Gen, and other teams attended, alongside researchers, developers, and operators from across the Bittensor ecosystem.

Commercialization was one of the main themes at Exploit Summit. Sessions examined how subnets are moving from experiments and proofs of concept toward products used by customers both inside and outside the Bittensor ecosystem, across inference, compute, enterprise AI, and other services. Gamma, a proposal presented at the summit, would let subnets use part of their network emissions to pay for compute, inference, and storage from infrastructure subnets on the network.

The summit also addressed practical questions facing teams building outside conventional technology-company models, including how resources are allocated and how independent contributors are organized. Other debates asked whether decentralized networks can compete with organizations spending at industry scale on infrastructure and how open networks approach safety and quality control.

Choosing Montreal as the venue provided an additional point of reference. The city has played an important role in the history of modern AI and deep-learning research, making it a fitting setting for a conference examining different ways of organizing AI development. The connection is historical and thematic; Montreal's established academic AI institutions were not participants in or endorsers of the Bittensor network.