Dominic Williams, founder of DFINITY and the Internet Computer Protocol (ICP), has issued a detailed response to criticism from a NEAR supporter who argued that ICP’s current on-chain AI inference is too slow and expensive for practical use.
The exchange centres on a broader debate across decentralised computing networks about whether AI workloads can be run securely on-chain, or whether practical AI applications will continue to rely primarily on off-chain infrastructure.
Williams argued that the criticism focuses too narrowly on the current limitations of fully on-chain AI inference and overlooks what he describes as a wider technical strategy for AI on the Internet Computer network. He acknowledged that mainstream AI inference is not yet practical in its present on-chain form because of cost and performance constraints, while maintaining that the ability to run neural networks directly on-chain remains an important technical achievement.
He claimed that ICP has demonstrated the ability to run neural networks and basic language models directly on the network, and said this reflects a level of computational capability that differs from conventional blockchain architectures. Williams argued that the network is designed to support applications that operate directly on-chain, rather than simply attaching tokens to external services.
A central part of his response focused on what the Internet Computer ecosystem calls the Intelligence Gateway, an upcoming service intended to allow decentralised applications to access multiple AI models through a common interface while paying with the network’s native computational resource, known as cycles. Williams said applications can already use outcalls to connect to external AI providers, and that the gateway is intended to simplify access and add verification capabilities for supported inference workloads.
According to Williams, the verification system would allow applications to confirm that prompts were not altered, that inference was performed using the intended AI model, and that outputs were not modified. He suggested that selective verification could provide additional security without substantially increasing overall costs for many applications.
Williams also outlined plans involving cloud engines and specialised AI nodes designed for enterprises and governments that want sensitive AI workloads to run on hardware under their own control within their own jurisdictions. He argued that this approach could support sovereign AI infrastructure while retaining the security properties associated with decentralised systems.
Another part of his argument concerned what he called “AIware”, a model in which AI agents build, update and operate software directly on the Internet Computer. Williams said the network’s Motoko programming language and orthogonal persistence model are designed to make software development easier for AI agents by reducing complexity and allowing applications to run in persistent memory.
He argued that future software development will increasingly be driven by AI agents and that persistent on-chain applications could allow businesses to operate through AI-assisted workflows with direct access to application data and functionality. Williams presented this as a long-term vision for enterprise software rather than a claim about current mainstream deployment.
The debate reflects a wider divide within the blockchain and decentralised computing sector over the most practical path for AI integration. Many networks are pursuing hybrid models that combine decentralised coordination with off-chain AI infrastructure, while projects such as ICP are attempting to push more computation and application logic directly onto decentralised networks.
Whether ICP’s approach gains broader adoption will depend on performance, cost, developer uptake and enterprise demand as these technologies mature. For now, Williams’ response has shifted the discussion from the narrow question of current on-chain inference speed to a broader argument about how AI applications and enterprise software may be built and verified in decentralised environments.
Dear Reader,
Ledger Life is an independent platform dedicated to covering the Internet Computer (ICP) ecosystem and beyond. We focus on real stories, builder updates, project launches, and the quiet innovations that often get missed.
We’re not backed by sponsors. We rely on readers like you.
If you find value in what we publish—whether it’s deep dives into dApps, explainers on decentralised tech, or just keeping track of what’s moving in Web3—please consider making a donation. It helps us cover costs, stay consistent, and remain truly independent.
Your support goes a long way.
🧠 ICP Principal: ins6i-d53ug-zxmgh-qvum3-r3pvl-ufcvu-bdyon-ovzdy-d26k3-lgq2v-3qe
🧾 ICP Address: f8deb966878f8b83204b251d5d799e0345ea72b8e62e8cf9da8d8830e1b3b05f
Every contribution helps keep the lights on, the stories flowing, and the crypto clutter out.
Thank you for reading, sharing, and being part of this experiment in decentralised media.
—Team Ledger Life
Community Discussion