Tee-rollups Enable 0.07 Second Blockchain Inference for Large Language Models

The increasing demand for scalable and trustworthy artificial intelligence within decentralized networks currently faces a fundamental challenge, often described as the Verifiability Trilemma, which limits the ability to simultaneously ensure computational integrity, low latency, and low cost. Aaron Chan, Alex Ding, and Frank Chen, from DGrid AI, alongside Alan Wu, Bruce Zhang, and Arther Tian, present a novel solution in the form of Optimistic TEE-Rollups, a hybrid verification protocol designed to overcome these limitations. This research introduces a system that combines the speed of optimistic approaches with the security of cryptographic proofs, leveraging Trusted Execution Environments to deliver near-instantaneous results while maintaining a high degree of trustworthiness. By formally defining a new consensus mechanism, Proof of Efficient Attribution, and employing stochastic Zero-Knowledge spot-checks, the team demonstrates a system that achieves performance comparable to centralized systems, with a minimal cost increase, and robust protection against malicious actors and hardware vulnerabilities.

Optimistic TEE-Rollups for Verifiable AI

Researchers have developed Optimistic TEE-Rollups (OTR), a novel protocol addressing the Verifiability Trilemma that hinders decentralized AI inference networks. This trilemma states that systems struggle to simultaneously achieve high computational integrity, low latency, and cost efficiency, and OTR successfully harmonizes these constraints. OTR combines the strengths of existing approaches by utilizing TEEs for efficient off-chain computation, optimistic rollups for immediate provisional finality, and cryptographic fraud proofs and stochastic zero-knowledge checks to enhance security.

The TEE performs the AI inference, and the result is posted on-chain with the assumption that the computation is correct unless challenged. A key innovation, Proof of Efficient Attribution (PoEA), cryptographically binds the execution trace to the hardware attestation, preventing reward hacking attacks where someone claims a reward for a complex model while actually running a cheaper one. Experiments demonstrate that OTR achieves 99% of the throughput of centralized systems, with a marginal cost overhead of only $0.07 per query. Compared to Zero-Knowledge Machine Learning (ZKML), OTR delivers a 1400x speedup, and in contrast to Optimistic Machine Learning (opML), it achieves a 99% reduction in latency.

The system maintains Byzantine fault tolerance, even in the presence of transient hardware vulnerabilities, ensuring robust and reliable operation against rational adversaries. This research establishes a foundational infrastructure for transitioning decentralized AI from a theoretical concept to a practical, production-ready reality, enabling verifiable and censorship-resistant inference at scale. Future work will focus on refining the multi-prover consensus mechanism to further reduce reliance on single-manufacturer enclaves.

Trusted Inference and Binding with TEE-Rollups

The research team addressed the limitations of existing decentralized inference systems by developing Optimistic TEE-Rollups (OTR), a novel protocol that achieves high throughput, low cost, and strong security guarantees. The protocol operates through a three-phase process, beginning with Trusted Inference and Binding. A Sequencer performs the inference within the TEE, and Proof of Efficient Attribution (PoEA) cryptographically binds the execution to a specific model through the enclave’s measurement.

To ensure privacy, the user encrypts the input query before sending it to the Sequencer, who decrypts it within the secure enclave and computes the response. To defend against compromised hardware or side-channel attacks, OTR incorporates a probabilistic verification layer during an optimistic window. A system-defined security parameter determines the probability of triggering a ZK-Spot-Check, requiring the Sequencer to generate a succinct proof of the computation. By carefully tuning this parameter, the scientists demonstrate that the system achieves 99% of the throughput of native execution while maintaining a credible threat against compromised hardware.

Optimistic Rollups Solve Verifiability Trilemma

Researchers have introduced Optimistic TEE-Rollups (OTR), a new architecture designed to address the Verifiability Trilemma that hinders decentralized AI inference networks. This trilemma states that systems struggle to simultaneously achieve high computational integrity, low latency, and cost efficiency, and OTR successfully harmonizes these constraints. OTR combines the strengths of existing approaches by utilizing TEEs for efficient off-chain computation, optimistic rollups for immediate provisional finality, and cryptographic fraud proofs and stochastic zero-knowledge checks to enhance security. Proof of Efficient Attribution (PoEA) cryptographically links execution traces to hardware, preventing attacks that could degrade model performance. Furthermore, the system achieves 99% of the throughput of centralized systems with sub-second finality, and at a cost of $0.07 per query, making it economically competitive with existing web-based services.