Modulus Labs

Modulus was built by a credible Stanford-trained team — founders Daniel Shorr and Ryan Cao — that shipped real demonstrations like the on-chain Rockybot agent and published serious zkML cost research. Execution was research-strong but commercially inconclusive: the team produced knowledge and prototypes more than a self-sustaining product, and ultimately judged that joining Tools for Humanity in December 2024 was the better path. Against EZKL, which kept an independent library alive, Modulus chose absorption over standalone survival. The neutral signal reflects genuine talent paired with the failure to establish an independent business.

Modulus did legitimately differentiated work, including its 'The Cost of Intelligence' research that benchmarked the feasibility of proving large models and pushed on the hardest part of the zkML problem. That ambition set it apart from tool-focused peers, but it also surfaced the uncomfortable conclusion that proving costs were far from practical at scale. Versus EZKL's pragmatic ONNX-to-Halo2 library, Modulus aimed higher and shipped less usable product. The differentiation was intellectual rather than commercial, and it now lives on inside World's applied-research roadmap rather than an open market.

Modulus never launched a token and now never will as an independent entity, so there is no economic mechanism to assess and no value-accrual path for outside participants. Any future value from the team's work accrues to World and the WLD ecosystem, not to a Modulus instrument. Compared with token-live peers like Giza, there is simply nothing to own here. Lookout marks this weak because the absorption foreclosed any independent economic future, not because of a flawed design.

Adoption never reached production scale: Modulus's outputs were demos, research, and proofs-of-concept rather than widely-used infrastructure with sustained demand. The most telling adoption signal is negative — the team concluded its work was best continued inside a larger organization rather than as a standalone network with users. Against RISC Zero's live Bonsai and Boundless usage, Modulus had no comparable deployed footprint. This axis is weak because there is no live, growing usage to point to, and the entity that would have built it has been folded in.

On financing pedigree alone Modulus was strong: a $6.3M seed co-led by Variant and 1kx, with backing tied to the Ethereum Foundation, Solana, Polygon, Celestia and Worldcoin ecosystems, is a genuinely high-quality cap table for the stage. That backer roster validated the thesis even though the company did not convert it into a standalone outcome. The irony is that one of those ecosystem relationships — Worldcoin — became the acquirer. The strong rating reflects the quality of capital raised, decoupled from the eventual wind-down.

Modulus rode the same resonant 'verifiable AI on-chain' narrative as EZKL and Giza, and 'bring AI on-chain' was an evocative framing at the height of the zkML cycle. But narrative fit without a sustainable product is precisely how a project ends up absorbed rather than scaled, and the market ultimately did not support an independent business around the pitch. Against peers that still carry the narrative as live entities, Modulus is now a historical reference. The fit was real; the durability was not.

For an outside observer the risk is total and already realized: Modulus no longer exists as an independent, investable entity, so there is no forward exposure to underwrite. Its own research crystallized the category's core technical risk — zkML proving costs being prohibitive for large models — which helps explain the outcome. Relative to peers still in the arena, Modulus's 'risk' is simply that the bet has resolved, into an acquisition rather than a network. Lookout rates this weak and treats the name as closed for deal purposes, tracked only for category honesty.